From 67014d8d55c4a6fbbf37b534aff442a7c8471303 Mon Sep 17 00:00:00 2001 From: ye shengjie <184427520+ye1457@users.noreply.github.com> Date: Wed, 16 Sep 2026 12:54:56 +0000 Subject: [PATCH 01/13] docs(benchmarks): focus leaderboard on aggregate results --- README.md | 10 +++++++++ README.zh-CN.md | 10 +++++++++ docs/source-en/index.rst | 1 + docs/source-en/rst_source/benchmarks.rst | 25 ++++++++++++++++++++++ docs/source-en/rst_source/overview.rst | 20 +++++++++++++++++ docs/source-en/rst_source/usage/libero.rst | 15 ++++++++----- docs/source-zh/index.rst | 1 + docs/source-zh/rst_source/benchmarks.rst | 25 ++++++++++++++++++++++ docs/source-zh/rst_source/overview.rst | 19 ++++++++++++++++ docs/source-zh/rst_source/usage/libero.rst | 9 ++++++-- 10 files changed, 128 insertions(+), 7 deletions(-) create mode 100644 docs/source-en/rst_source/benchmarks.rst create mode 100644 docs/source-zh/rst_source/benchmarks.rst diff --git a/README.md b/README.md index 6b6fcfb84..6189c87fa 100644 --- a/README.md +++ b/README.md @@ -27,6 +27,15 @@ RPent framework +## Benchmark Results + +Compare RPent with reference methods on LIBERO, LIBERO-PRO, RoboCasa365 Target50, and RoboTwin C2R. Rankings apply to the methods and evaluation coverage shown; see [Benchmark Results](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) for suite results and model configurations. + +Codex / GPT-6 Astra / low / reasoning: **92.63% Overall (741/800)** across all eight LIBERO-PRO suites. See the [suite results](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html), including the separately frozen Long and Spatial/Object/Goal memory batches. + +[![RPent success-rate comparisons across four benchmarks](https://raw.githubusercontent.com/RLinf/misc/aacd664e6b486a1b75cafc765b354f18d88004d2/rpent/benchmarks/leaderboard-en-light.png)](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) + + ## Who Should Consider Using RPent? RPent is built for four kinds of users: @@ -38,6 +47,7 @@ RPent is built for four kinds of users: ## What's NEW! +- [2026/09] Added an interactive leaderboard and consolidated benchmark results for LIBERO, LIBERO-PRO, RoboCasa365, and RoboTwin, with model comparisons and suite-level results. Explore [Benchmark Results](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html). - [2026/09] 🔥 RPent supports Franka single-arm and dual-arm real-robot extensions. Doc: [Franka](https://rpent.readthedocs.io/en/latest/rst_source/usage/franka.html) · [Dual Franka](https://rpent.readthedocs.io/en/latest/rst_source/usage/dual_franka.html). - [2026/08] 🔥 RPent supports RoboCasa with RLDX-1 as manipulation model. See the [RoboCasa setup and Target50 guide](https://rpent.readthedocs.io/en/latest/rst_source/usage/robocasa.html). - [2026/08] 🔥 RPent supports the non-reasoning mode, which reduces average execution time by ~40%. diff --git a/README.zh-CN.md b/README.zh-CN.md index e45334ce8..952ab19e4 100644 --- a/README.zh-CN.md +++ b/README.zh-CN.md @@ -27,6 +27,15 @@ RPent framework +## 基准测试结果 + +对比 RPent 与参考方法在 LIBERO、LIBERO-PRO、RoboCasa365 Target50 和 RoboTwin C2R 上的成功率。排名仅限图中方法及评测范围；套件成绩和模型配置见[基准测试结果](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html)。 + +Codex / GPT-6 Astra / low / reasoning 已完成全部八套 LIBERO-PRO，**Overall 92.63%（741/800）**。详见 [套件汇总](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html)，其中 Long 与 Spatial/Object/Goal 分别使用各自冻结的 memory 批次。 + +[![RPent 四项基准成功率对比](https://raw.githubusercontent.com/RLinf/misc/aacd664e6b486a1b75cafc765b354f18d88004d2/rpent/benchmarks/leaderboard-zh-light.png)](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html) + + ## 适用用户 RPent 面向以下四类用户： @@ -38,6 +47,7 @@ RPent 面向以下四类用户： ## 最新动态 +- [2026/09] 新增交互式排行榜和基准测试结果汇总，覆盖 LIBERO、LIBERO-PRO、RoboCasa365 与 RoboTwin，提供模型对比及套件汇总。查看[基准测试结果](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html)。 - [2026/08] 🔥 支持 RoboCasa，使用 RLDX-1 作为操作模型。参见 [RoboCasa 安装与 Target50 指南](https://rpent.readthedocs.io/zh-cn/latest/rst_source/usage/robocasa.html)。 - [2026/08] 🔥 新增非推理（non-reasoning）模式，平均执行时间降低约 40%。 - [2026/08] 🔥 支持 LIBERO 探索模式。文档：[LIBERO 探索模式](https://rpent.readthedocs.io/zh-cn/latest/rst_source/usage/libero.html#memory)。 diff --git a/docs/source-en/index.rst b/docs/source-en/index.rst index 4a91656b9..21cbaff2d 100644 --- a/docs/source-en/index.rst +++ b/docs/source-en/index.rst @@ -74,6 +74,7 @@ Welcome to RPent Overview Installation Quick Start + Benchmark Results .. toctree:: :maxdepth: 2 diff --git a/docs/source-en/rst_source/benchmarks.rst b/docs/source-en/rst_source/benchmarks.rst new file mode 100644 index 000000000..7ea6b6995 --- /dev/null +++ b/docs/source-en/rst_source/benchmarks.rst @@ -0,0 +1,25 @@ +:html_theme.sidebar_secondary.remove: + +.. _benchmark-results: +.. _benchmark-leaderboard: +.. _leaderboard: + +RPent Leaderboard +================= + +.. raw:: html + + + + + +
+
+

LIBERO-PRO

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus 4.782.50%
Claude Code / Opus-4.8 / max / reasoning82.4%
Codex / GPT-5.6 / xhigh / reasoning78.50%
Task card / Molmo72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
Codex / GPT-5.6 / no reasoning62.50%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%
OpenVLA0.0%
NORA0.0%
+

LIBERO

Method / modelSuccess rate
AtomVLA97.0%
Claude Code / Opus-4.8 / max / reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%
+

RoboCasa365 · Target50

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning59.20%
Codex / GPT-5.5 / xhigh / reasoning57.1%
Claude Code / Opus-4.8 / max / reasoning48.6%
WorldDreamer35.3%
RLDX-130.0%
π0.516.9%
π014.8%
+

RoboTwin

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning72.00%
Codex / GPT-5.5 / xhigh / reasoning62.4%
Codex / GPT-5.6 / xhigh / reasoning61.20%
Claude Code / Opus-4.8 / max / reasoning58.4%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%
+

GPT-6 Astra · low · reasoning: LIBERO-PRO 92.63% (741/800).

+
+
diff --git a/docs/source-en/rst_source/overview.rst b/docs/source-en/rst_source/overview.rst index d00a84cff..3fa4c0e09 100644 --- a/docs/source-en/rst_source/overview.rst +++ b/docs/source-en/rst_source/overview.rst @@ -26,6 +26,26 @@ Together, these principles allow RPent to move beyond traditional robot control and establish an agentic infrastructure for the physical world, where intelligence is not only deployed, but continuously built, expanded, and evolved. +Benchmark Results +----------------- + +Compare success rates on LIBERO, LIBERO-PRO, RoboCasa365 Target50, and RoboTwin +C2R. Rankings apply to the methods and evaluation coverage shown; see +:doc:`benchmarks` for detailed results, configurations, and sources. + +.. image:: https://raw.githubusercontent.com/RLinf/misc/aacd664e6b486a1b75cafc765b354f18d88004d2/rpent/benchmarks/leaderboard-en-light.png + :alt: RPent benchmark results + :class: only-light + :width: 100% + :target: benchmarks.html + +.. image:: https://raw.githubusercontent.com/RLinf/misc/aacd664e6b486a1b75cafc765b354f18d88004d2/rpent/benchmarks/leaderboard-en-dark.png + :alt: RPent benchmark results + :class: only-dark + :width: 100% + :target: benchmarks.html + + Feature Matrix -------------- diff --git a/docs/source-en/rst_source/usage/libero.rst b/docs/source-en/rst_source/usage/libero.rst index df38cab1b..86101b7c4 100644 --- a/docs/source-en/rst_source/usage/libero.rst +++ b/docs/source-en/rst_source/usage/libero.rst @@ -278,11 +278,16 @@ See :doc:`../development/add_primitive` for the full walkthrough. Reproducing results ------------------- -The following results reproduce -:doc:`Harness VLA <../awesome_works/harnessvla>` on two LIBERO-PRO suites. -On the `reproduce/libero -`_ branch, use -``gpt-5.5`` to reproduce these results: +See :doc:`../benchmarks` for the unified RPent model comparison on LIBERO-PRO +Task/Swap and the corresponding model configurations. + +The :doc:`GPT-6 Astra suite results <../benchmarks>` +cover all eight complete suites and 800 verified episodes: 741 successes, +59 failures, and 92.63% Overall, with Codex / GPT-6 Astra / low / reasoning. + +The following historical reproduction records use the `reproduce/libero +`_ branch with +``gpt-5.5`` and ``xhigh`` reasoning effort: - ``libero_10_task``: 70% (70/100) - ``libero_10_swap``: 55% (55/100) diff --git a/docs/source-zh/index.rst b/docs/source-zh/index.rst index 2d607e5e3..e1f4bb2a5 100644 --- a/docs/source-zh/index.rst +++ b/docs/source-zh/index.rst @@ -66,6 +66,7 @@ 概览 安装 快速开始 + 基准测试结果 .. toctree:: :maxdepth: 2 diff --git a/docs/source-zh/rst_source/benchmarks.rst b/docs/source-zh/rst_source/benchmarks.rst new file mode 100644 index 000000000..6f912ec38 --- /dev/null +++ b/docs/source-zh/rst_source/benchmarks.rst @@ -0,0 +1,25 @@ +:html_theme.sidebar_secondary.remove: + +.. _benchmark-results: +.. _benchmark-leaderboard: +.. _leaderboard: + +RPent 排行榜 +================= + +.. raw:: html + + + + + +
+
+

LIBERO-PRO

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus 4.782.50%
Claude Code / Opus-4.8 / max / reasoning82.4%
Codex / GPT-5.6 / xhigh / reasoning78.50%
Task card / Molmo72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
Codex / GPT-5.6 / 无推理62.50%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%
OpenVLA0.0%
NORA0.0%
+

LIBERO

方法 / 模型成功率
AtomVLA97.0%
Claude Code / Opus-4.8 / max / reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%
+

RoboCasa365 · Target50

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning59.20%
Codex / GPT-5.5 / xhigh / reasoning57.1%
Claude Code / Opus-4.8 / max / reasoning48.6%
WorldDreamer35.3%
RLDX-130.0%
π0.516.9%
π014.8%
+

RoboTwin

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning72.00%
Codex / GPT-5.5 / xhigh / reasoning62.4%
Codex / GPT-5.6 / xhigh / reasoning61.20%
Claude Code / Opus-4.8 / max / reasoning58.4%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%
+

GPT-6 Astra · low · reasoning：LIBERO-PRO 92.63%（741/800）。

+
+
diff --git a/docs/source-zh/rst_source/overview.rst b/docs/source-zh/rst_source/overview.rst index 89946e887..4f82ac056 100644 --- a/docs/source-zh/rst_source/overview.rst +++ b/docs/source-zh/rst_source/overview.rst @@ -19,6 +19,25 @@ RPent 建立在三条核心设计原则之上: **服务化、标准化、可组 智能体基础设施 (agentic infrastructure for the physical world) —— 在这里, 智能不只是被部署, 而是被持续构建、扩展与演进。 +基准测试结果 +------------ + +对比 LIBERO、LIBERO-PRO、RoboCasa365 Target50 和 RoboTwin C2R 上的成功率。 +排名仅限图中方法及评测范围；完整结果、模型配置和来源见 :doc:`benchmarks`。 + +.. image:: https://raw.githubusercontent.com/RLinf/misc/aacd664e6b486a1b75cafc765b354f18d88004d2/rpent/benchmarks/leaderboard-zh-light.png + :alt: RPent benchmark results + :class: only-light + :width: 100% + :target: benchmarks.html + +.. image:: https://raw.githubusercontent.com/RLinf/misc/aacd664e6b486a1b75cafc765b354f18d88004d2/rpent/benchmarks/leaderboard-zh-dark.png + :alt: RPent benchmark results + :class: only-dark + :width: 100% + :target: benchmarks.html + + 功能矩阵 -------- diff --git a/docs/source-zh/rst_source/usage/libero.rst b/docs/source-zh/rst_source/usage/libero.rst index 1ee4003d6..172395647 100644 --- a/docs/source-zh/rst_source/usage/libero.rst +++ b/docs/source-zh/rst_source/usage/libero.rst @@ -254,8 +254,13 @@ Dashboard 支持 ``api``、``claude_code`` 和 ``codex`` planner。 结果复现 -------- -以下是在两个 LIBERO-PRO 套件上复现 -:doc:`Harness VLA <../awesome_works/harnessvla>` 得到的结果。实验使用 +RPent 在 LIBERO-PRO Task/Swap 上的统一模型对比及对应配置见 :doc:`../benchmarks`。 + +:doc:`GPT-6 Astra 套件汇总 <../benchmarks>` +记录全部八个完整套件及 800 个已核验回合：741 成功、59 失败，Overall 92.63%， +配置为 Codex / GPT-6 Astra / low / reasoning。 + +以下保留历史复现记录，实验使用 `reproduce/libero `_ 分支和 ``gpt-5.5`` 模型： From c19727c7bbdd277cd86a985aae3f2741ba01df7a Mon Sep 17 00:00:00 2001 From: ye shengjie <184427520+ye1457@users.noreply.github.com> Date: Wed, 16 Sep 2026 14:54:10 +0000 Subject: [PATCH 02/13] docs(benchmarks): apply corrected model results and resource snapshot --- README.md | 2 +- README.zh-CN.md | 2 +- docs/source-en/rst_source/benchmarks.rst | 18 +++++++++--------- docs/source-en/rst_source/overview.rst | 4 ++-- docs/source-zh/rst_source/benchmarks.rst | 18 +++++++++--------- docs/source-zh/rst_source/overview.rst | 4 ++-- 6 files changed, 24 insertions(+), 24 deletions(-) diff --git a/README.md b/README.md index 6189c87fa..6d8f77efd 100644 --- a/README.md +++ b/README.md @@ -33,7 +33,7 @@ Compare RPent with reference methods on LIBERO, LIBERO-PRO, RoboCasa365 Target50 Codex / GPT-6 Astra / low / reasoning: **92.63% Overall (741/800)** across all eight LIBERO-PRO suites. See the [suite results](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html), including the separately frozen Long and Spatial/Object/Goal memory batches. -[![RPent success-rate comparisons across four benchmarks](https://raw.githubusercontent.com/RLinf/misc/aacd664e6b486a1b75cafc765b354f18d88004d2/rpent/benchmarks/leaderboard-en-light.png)](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) +[![RPent success-rate comparisons across four benchmarks](https://raw.githubusercontent.com/RLinf/misc/7551b6b0a0228e9c9fd2b70ef994b0c40f1b1412/rpent/benchmarks/leaderboard-en-light.png)](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) ## Who Should Consider Using RPent? diff --git a/README.zh-CN.md b/README.zh-CN.md index 952ab19e4..267e98818 100644 --- a/README.zh-CN.md +++ b/README.zh-CN.md @@ -33,7 +33,7 @@ Codex / GPT-6 Astra / low / reasoning 已完成全部八套 LIBERO-PRO，**Overall 92.63%（741/800）**。详见 [套件汇总](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html)，其中 Long 与 Spatial/Object/Goal 分别使用各自冻结的 memory 批次。 -[![RPent 四项基准成功率对比](https://raw.githubusercontent.com/RLinf/misc/aacd664e6b486a1b75cafc765b354f18d88004d2/rpent/benchmarks/leaderboard-zh-light.png)](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html) +[![RPent 四项基准成功率对比](https://raw.githubusercontent.com/RLinf/misc/7551b6b0a0228e9c9fd2b70ef994b0c40f1b1412/rpent/benchmarks/leaderboard-zh-light.png)](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html) ## 适用用户 diff --git a/docs/source-en/rst_source/benchmarks.rst b/docs/source-en/rst_source/benchmarks.rst index 7ea6b6995..f20768a1f 100644 --- a/docs/source-en/rst_source/benchmarks.rst +++ b/docs/source-en/rst_source/benchmarks.rst @@ -9,17 +9,17 @@ RPent Leaderboard .. raw:: html - - - - + + + +
+ data-results-url="https://raw.githubusercontent.com/RLinf/misc/7551b6b0a0228e9c9fd2b70ef994b0c40f1b1412/rpent/benchmarks/results.json">
-

LIBERO-PRO

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus 4.782.50%
Claude Code / Opus-4.8 / max / reasoning82.4%
Codex / GPT-5.6 / xhigh / reasoning78.50%
Task card / Molmo72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
Codex / GPT-5.6 / no reasoning62.50%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%
OpenVLA0.0%
NORA0.0%
-

LIBERO

Method / modelSuccess rate
AtomVLA97.0%
Claude Code / Opus-4.8 / max / reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%
-

RoboCasa365 · Target50

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning59.20%
Codex / GPT-5.5 / xhigh / reasoning57.1%
Claude Code / Opus-4.8 / max / reasoning48.6%
WorldDreamer35.3%
RLDX-130.0%
π0.516.9%
π014.8%
-

RoboTwin

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning72.00%
Codex / GPT-5.5 / xhigh / reasoning62.4%
Codex / GPT-5.6 / xhigh / reasoning61.20%
Claude Code / Opus-4.8 / max / reasoning58.4%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%
+

LIBERO-PRO

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus 4.782.4%
Task card / Molmo72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%
NORA0.0%
OpenVLA0.0%
+

LIBERO

Method / modelSuccess rate
AtomVLA97.0%
Claude Code / Opus-4.7 / max / reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%
+

RoboCasa365 · Target50

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning59.20%
Codex / GPT-5.5 / xhigh / reasoning57.1%
Claude Code / Opus-4.7 / max / reasoning48.6%
WorldDreamer35.3%
RLDX-130.0%
π0.516.9%
π014.8%
+

RoboTwin

Method / modelSuccess rate
Claude Code / Opus-4.7 / max / reasoning58.4%
Codex / GPT-5.5 / xhigh / reasoning58.0%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%

GPT-6 Astra · low · reasoning: LIBERO-PRO 92.63% (741/800).

diff --git a/docs/source-en/rst_source/overview.rst b/docs/source-en/rst_source/overview.rst index 3fa4c0e09..6ac5f430e 100644 --- a/docs/source-en/rst_source/overview.rst +++ b/docs/source-en/rst_source/overview.rst @@ -33,13 +33,13 @@ Compare success rates on LIBERO, LIBERO-PRO, RoboCasa365 Target50, and RoboTwin C2R. Rankings apply to the methods and evaluation coverage shown; see :doc:`benchmarks` for detailed results, configurations, and sources. -.. image:: https://raw.githubusercontent.com/RLinf/misc/aacd664e6b486a1b75cafc765b354f18d88004d2/rpent/benchmarks/leaderboard-en-light.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/7551b6b0a0228e9c9fd2b70ef994b0c40f1b1412/rpent/benchmarks/leaderboard-en-light.png :alt: RPent benchmark results :class: only-light :width: 100% :target: benchmarks.html -.. image:: https://raw.githubusercontent.com/RLinf/misc/aacd664e6b486a1b75cafc765b354f18d88004d2/rpent/benchmarks/leaderboard-en-dark.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/7551b6b0a0228e9c9fd2b70ef994b0c40f1b1412/rpent/benchmarks/leaderboard-en-dark.png :alt: RPent benchmark results :class: only-dark :width: 100% diff --git a/docs/source-zh/rst_source/benchmarks.rst b/docs/source-zh/rst_source/benchmarks.rst index 6f912ec38..e6f581f40 100644 --- a/docs/source-zh/rst_source/benchmarks.rst +++ b/docs/source-zh/rst_source/benchmarks.rst @@ -9,17 +9,17 @@ RPent 排行榜 .. raw:: html - - - - + + + +
+ data-results-url="https://raw.githubusercontent.com/RLinf/misc/7551b6b0a0228e9c9fd2b70ef994b0c40f1b1412/rpent/benchmarks/results.json">
-

LIBERO-PRO

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus 4.782.50%
Claude Code / Opus-4.8 / max / reasoning82.4%
Codex / GPT-5.6 / xhigh / reasoning78.50%
Task card / Molmo72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
Codex / GPT-5.6 / 无推理62.50%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%
OpenVLA0.0%
NORA0.0%
-

LIBERO

方法 / 模型成功率
AtomVLA97.0%
Claude Code / Opus-4.8 / max / reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%
-

RoboCasa365 · Target50

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning59.20%
Codex / GPT-5.5 / xhigh / reasoning57.1%
Claude Code / Opus-4.8 / max / reasoning48.6%
WorldDreamer35.3%
RLDX-130.0%
π0.516.9%
π014.8%
-

RoboTwin

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning72.00%
Codex / GPT-5.5 / xhigh / reasoning62.4%
Codex / GPT-5.6 / xhigh / reasoning61.20%
Claude Code / Opus-4.8 / max / reasoning58.4%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%
+

LIBERO-PRO

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus 4.782.4%
Task card / Molmo72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%
NORA0.0%
OpenVLA0.0%
+

LIBERO

方法 / 模型成功率
AtomVLA97.0%
Claude Code / Opus-4.7 / max / reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%
+

RoboCasa365 · Target50

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning59.20%
Codex / GPT-5.5 / xhigh / reasoning57.1%
Claude Code / Opus-4.7 / max / reasoning48.6%
WorldDreamer35.3%
RLDX-130.0%
π0.516.9%
π014.8%
+

RoboTwin

方法 / 模型成功率
Claude Code / Opus-4.7 / max / reasoning58.4%
Codex / GPT-5.5 / xhigh / reasoning58.0%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%

GPT-6 Astra · low · reasoning：LIBERO-PRO 92.63%（741/800）。

diff --git a/docs/source-zh/rst_source/overview.rst b/docs/source-zh/rst_source/overview.rst index 4f82ac056..4c00be43c 100644 --- a/docs/source-zh/rst_source/overview.rst +++ b/docs/source-zh/rst_source/overview.rst @@ -25,13 +25,13 @@ RPent 建立在三条核心设计原则之上: **服务化、标准化、可组 对比 LIBERO、LIBERO-PRO、RoboCasa365 Target50 和 RoboTwin C2R 上的成功率。 排名仅限图中方法及评测范围；完整结果、模型配置和来源见 :doc:`benchmarks`。 -.. image:: https://raw.githubusercontent.com/RLinf/misc/aacd664e6b486a1b75cafc765b354f18d88004d2/rpent/benchmarks/leaderboard-zh-light.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/7551b6b0a0228e9c9fd2b70ef994b0c40f1b1412/rpent/benchmarks/leaderboard-zh-light.png :alt: RPent benchmark results :class: only-light :width: 100% :target: benchmarks.html -.. image:: https://raw.githubusercontent.com/RLinf/misc/aacd664e6b486a1b75cafc765b354f18d88004d2/rpent/benchmarks/leaderboard-zh-dark.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/7551b6b0a0228e9c9fd2b70ef994b0c40f1b1412/rpent/benchmarks/leaderboard-zh-dark.png :alt: RPent benchmark results :class: only-dark :width: 100% From 118388f08063eee1e1eb0edd8233c897e110d8fd Mon Sep 17 00:00:00 2001 From: ye shengjie <184427520+ye1457@users.noreply.github.com> Date: Wed, 16 Sep 2026 15:29:57 +0000 Subject: [PATCH 03/13] docs(benchmarks): sync Flash Mode and Xiaomi result summaries --- README.md | 2 +- README.zh-CN.md | 2 +- docs/source-en/rst_source/benchmarks.rst | 17 +++++++++-------- docs/source-en/rst_source/overview.rst | 4 ++-- docs/source-zh/rst_source/benchmarks.rst | 17 +++++++++-------- docs/source-zh/rst_source/overview.rst | 4 ++-- 6 files changed, 24 insertions(+), 22 deletions(-) diff --git a/README.md b/README.md index 6d8f77efd..d5c31c248 100644 --- a/README.md +++ b/README.md @@ -33,7 +33,7 @@ Compare RPent with reference methods on LIBERO, LIBERO-PRO, RoboCasa365 Target50 Codex / GPT-6 Astra / low / reasoning: **92.63% Overall (741/800)** across all eight LIBERO-PRO suites. See the [suite results](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html), including the separately frozen Long and Spatial/Object/Goal memory batches. -[![RPent success-rate comparisons across four benchmarks](https://raw.githubusercontent.com/RLinf/misc/7551b6b0a0228e9c9fd2b70ef994b0c40f1b1412/rpent/benchmarks/leaderboard-en-light.png)](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) +[![RPent success-rate comparisons across four benchmarks](https://raw.githubusercontent.com/RLinf/misc/29a9d96588fbd9fa01cb4e035658eba028384fb6/rpent/benchmarks/leaderboard-en-light.png)](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) ## Who Should Consider Using RPent? diff --git a/README.zh-CN.md b/README.zh-CN.md index 267e98818..35757badd 100644 --- a/README.zh-CN.md +++ b/README.zh-CN.md @@ -33,7 +33,7 @@ Codex / GPT-6 Astra / low / reasoning 已完成全部八套 LIBERO-PRO，**Overall 92.63%（741/800）**。详见 [套件汇总](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html)，其中 Long 与 Spatial/Object/Goal 分别使用各自冻结的 memory 批次。 -[![RPent 四项基准成功率对比](https://raw.githubusercontent.com/RLinf/misc/7551b6b0a0228e9c9fd2b70ef994b0c40f1b1412/rpent/benchmarks/leaderboard-zh-light.png)](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html) +[![RPent 四项基准成功率对比](https://raw.githubusercontent.com/RLinf/misc/29a9d96588fbd9fa01cb4e035658eba028384fb6/rpent/benchmarks/leaderboard-zh-light.png)](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html) ## 适用用户 diff --git a/docs/source-en/rst_source/benchmarks.rst b/docs/source-en/rst_source/benchmarks.rst index f20768a1f..6e4446411 100644 --- a/docs/source-en/rst_source/benchmarks.rst +++ b/docs/source-en/rst_source/benchmarks.rst @@ -9,17 +9,18 @@ RPent Leaderboard .. raw:: html - - - - + + + +
+ data-results-url="https://raw.githubusercontent.com/RLinf/misc/29a9d96588fbd9fa01cb4e035658eba028384fb6/rpent/benchmarks/results.json">
-

LIBERO-PRO

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus 4.782.4%
Task card / Molmo72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%
NORA0.0%
OpenVLA0.0%
+

LIBERO-PRO

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus 4.782.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%
NORA0.0%
OpenVLA0.0%

LIBERO

Method / modelSuccess rate
AtomVLA97.0%
Claude Code / Opus-4.7 / max / reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%
-

RoboCasa365 · Target50

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning59.20%
Codex / GPT-5.5 / xhigh / reasoning57.1%
Claude Code / Opus-4.7 / max / reasoning48.6%
WorldDreamer35.3%
RLDX-130.0%
π0.516.9%
π014.8%
-

RoboTwin

Method / modelSuccess rate
Claude Code / Opus-4.7 / max / reasoning58.4%
Codex / GPT-5.5 / xhigh / reasoning58.0%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%
+

RoboCasa365 · Target50

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning59.20%
Xiaomi-Robotics-157.4%
Codex / GPT-5.5 / xhigh / reasoning57.1%
Claude Code / Opus-4.7 / max / reasoning48.6%
WorldDreamer35.3%
RLDX-130.0%
π0.516.9%
π014.8%
+

RoboTwin

Method / modelSuccess rate
Codex / GPT-5.5 / xhigh / reasoning62.4%
Claude Code / Opus-4.7 / max / reasoning58.4%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%

GPT-6 Astra · low · reasoning: LIBERO-PRO 92.63% (741/800).

+

RPent Flash Mode: 72.63% (581/800). Molmo is used for visual localization only.

diff --git a/docs/source-en/rst_source/overview.rst b/docs/source-en/rst_source/overview.rst index 6ac5f430e..7a44285c4 100644 --- a/docs/source-en/rst_source/overview.rst +++ b/docs/source-en/rst_source/overview.rst @@ -33,13 +33,13 @@ Compare success rates on LIBERO, LIBERO-PRO, RoboCasa365 Target50, and RoboTwin C2R. Rankings apply to the methods and evaluation coverage shown; see :doc:`benchmarks` for detailed results, configurations, and sources. -.. image:: https://raw.githubusercontent.com/RLinf/misc/7551b6b0a0228e9c9fd2b70ef994b0c40f1b1412/rpent/benchmarks/leaderboard-en-light.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/29a9d96588fbd9fa01cb4e035658eba028384fb6/rpent/benchmarks/leaderboard-en-light.png :alt: RPent benchmark results :class: only-light :width: 100% :target: benchmarks.html -.. image:: https://raw.githubusercontent.com/RLinf/misc/7551b6b0a0228e9c9fd2b70ef994b0c40f1b1412/rpent/benchmarks/leaderboard-en-dark.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/29a9d96588fbd9fa01cb4e035658eba028384fb6/rpent/benchmarks/leaderboard-en-dark.png :alt: RPent benchmark results :class: only-dark :width: 100% diff --git a/docs/source-zh/rst_source/benchmarks.rst b/docs/source-zh/rst_source/benchmarks.rst index e6f581f40..1ada8814d 100644 --- a/docs/source-zh/rst_source/benchmarks.rst +++ b/docs/source-zh/rst_source/benchmarks.rst @@ -9,17 +9,18 @@ RPent 排行榜 .. raw:: html - - - - + + + +
+ data-results-url="https://raw.githubusercontent.com/RLinf/misc/29a9d96588fbd9fa01cb4e035658eba028384fb6/rpent/benchmarks/results.json">
-

LIBERO-PRO

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus 4.782.4%
Task card / Molmo72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%
NORA0.0%
OpenVLA0.0%
+

LIBERO-PRO

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus 4.782.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%
NORA0.0%
OpenVLA0.0%

LIBERO

方法 / 模型成功率
AtomVLA97.0%
Claude Code / Opus-4.7 / max / reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%
-

RoboCasa365 · Target50

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning59.20%
Codex / GPT-5.5 / xhigh / reasoning57.1%
Claude Code / Opus-4.7 / max / reasoning48.6%
WorldDreamer35.3%
RLDX-130.0%
π0.516.9%
π014.8%
-

RoboTwin

方法 / 模型成功率
Claude Code / Opus-4.7 / max / reasoning58.4%
Codex / GPT-5.5 / xhigh / reasoning58.0%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%
+

RoboCasa365 · Target50

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning59.20%
Xiaomi-Robotics-157.4%
Codex / GPT-5.5 / xhigh / reasoning57.1%
Claude Code / Opus-4.7 / max / reasoning48.6%
WorldDreamer35.3%
RLDX-130.0%
π0.516.9%
π014.8%
+

RoboTwin

方法 / 模型成功率
Codex / GPT-5.5 / xhigh / reasoning62.4%
Claude Code / Opus-4.7 / max / reasoning58.4%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%

GPT-6 Astra · low · reasoning：LIBERO-PRO 92.63%（741/800）。

+

RPent Flash Mode: 72.63% (581/800). Molmo 仅用于视觉定位。

diff --git a/docs/source-zh/rst_source/overview.rst b/docs/source-zh/rst_source/overview.rst index 4c00be43c..75c983b23 100644 --- a/docs/source-zh/rst_source/overview.rst +++ b/docs/source-zh/rst_source/overview.rst @@ -25,13 +25,13 @@ RPent 建立在三条核心设计原则之上: **服务化、标准化、可组 对比 LIBERO、LIBERO-PRO、RoboCasa365 Target50 和 RoboTwin C2R 上的成功率。 排名仅限图中方法及评测范围；完整结果、模型配置和来源见 :doc:`benchmarks`。 -.. image:: https://raw.githubusercontent.com/RLinf/misc/7551b6b0a0228e9c9fd2b70ef994b0c40f1b1412/rpent/benchmarks/leaderboard-zh-light.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/29a9d96588fbd9fa01cb4e035658eba028384fb6/rpent/benchmarks/leaderboard-zh-light.png :alt: RPent benchmark results :class: only-light :width: 100% :target: benchmarks.html -.. image:: https://raw.githubusercontent.com/RLinf/misc/7551b6b0a0228e9c9fd2b70ef994b0c40f1b1412/rpent/benchmarks/leaderboard-zh-dark.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/29a9d96588fbd9fa01cb4e035658eba028384fb6/rpent/benchmarks/leaderboard-zh-dark.png :alt: RPent benchmark results :class: only-dark :width: 100% From f78a66e61dbcf9f1e6be886ab48838886e08af87 Mon Sep 17 00:00:00 2001 From: ye shengjie <184427520+ye1457@users.noreply.github.com> Date: Wed, 16 Sep 2026 15:58:17 +0000 Subject: [PATCH 04/13] docs(benchmarks): document ASPIRE and Molmo2 Flash breakdowns --- README.md | 2 +- README.zh-CN.md | 2 +- docs/source-en/rst_source/benchmarks.rst | 16 ++++++++-------- docs/source-en/rst_source/overview.rst | 4 ++-- docs/source-zh/rst_source/benchmarks.rst | 16 ++++++++-------- docs/source-zh/rst_source/overview.rst | 4 ++-- 6 files changed, 22 insertions(+), 22 deletions(-) diff --git a/README.md b/README.md index d5c31c248..1b98c77d5 100644 --- a/README.md +++ b/README.md @@ -33,7 +33,7 @@ Compare RPent with reference methods on LIBERO, LIBERO-PRO, RoboCasa365 Target50 Codex / GPT-6 Astra / low / reasoning: **92.63% Overall (741/800)** across all eight LIBERO-PRO suites. See the [suite results](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html), including the separately frozen Long and Spatial/Object/Goal memory batches. -[![RPent success-rate comparisons across four benchmarks](https://raw.githubusercontent.com/RLinf/misc/29a9d96588fbd9fa01cb4e035658eba028384fb6/rpent/benchmarks/leaderboard-en-light.png)](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) +[![RPent success-rate comparisons across four benchmarks](https://raw.githubusercontent.com/RLinf/misc/07c25bdb9c036ecc437f8e83c8117a080e4a6d2e/rpent/benchmarks/leaderboard-en-light.png)](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) ## Who Should Consider Using RPent? diff --git a/README.zh-CN.md b/README.zh-CN.md index 35757badd..869d5ecd8 100644 --- a/README.zh-CN.md +++ b/README.zh-CN.md @@ -33,7 +33,7 @@ Codex / GPT-6 Astra / low / reasoning 已完成全部八套 LIBERO-PRO，**Overall 92.63%（741/800）**。详见 [套件汇总](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html)，其中 Long 与 Spatial/Object/Goal 分别使用各自冻结的 memory 批次。 -[![RPent 四项基准成功率对比](https://raw.githubusercontent.com/RLinf/misc/29a9d96588fbd9fa01cb4e035658eba028384fb6/rpent/benchmarks/leaderboard-zh-light.png)](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html) +[![RPent 四项基准成功率对比](https://raw.githubusercontent.com/RLinf/misc/07c25bdb9c036ecc437f8e83c8117a080e4a6d2e/rpent/benchmarks/leaderboard-zh-light.png)](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html) ## 适用用户 diff --git a/docs/source-en/rst_source/benchmarks.rst b/docs/source-en/rst_source/benchmarks.rst index 6e4446411..9ea2a2e28 100644 --- a/docs/source-en/rst_source/benchmarks.rst +++ b/docs/source-en/rst_source/benchmarks.rst @@ -9,18 +9,18 @@ RPent Leaderboard .. raw:: html - - - - + + + +
+ data-results-url="https://raw.githubusercontent.com/RLinf/misc/07c25bdb9c036ecc437f8e83c8117a080e4a6d2e/rpent/benchmarks/results.json">
-

LIBERO-PRO

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus 4.782.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%
NORA0.0%
OpenVLA0.0%
+

LIBERO-PRO

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus 4.782.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%

LIBERO

Method / modelSuccess rate
AtomVLA97.0%
Claude Code / Opus-4.7 / max / reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%

RoboCasa365 · Target50

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning59.20%
Xiaomi-Robotics-157.4%
Codex / GPT-5.5 / xhigh / reasoning57.1%
Claude Code / Opus-4.7 / max / reasoning48.6%
WorldDreamer35.3%
RLDX-130.0%
π0.516.9%
π014.8%

RoboTwin

Method / modelSuccess rate
Codex / GPT-5.5 / xhigh / reasoning62.4%
Claude Code / Opus-4.7 / max / reasoning58.4%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%
-

GPT-6 Astra · low · reasoning: LIBERO-PRO 92.63% (741/800).

-

RPent Flash Mode: 72.63% (581/800). Molmo is used for visual localization only.

+

GPT-6 Astra: 92.63% (741/800). RPent Flash Mode / Molmo2-8B: 72.63% (581/800).

+

ASPIRE: six main-suite scores; Long uses zero-shot transfer from LIBERO-90. No eight-suite Overall is reported. Flash Mode uses Molmo2-8B for visual localization, not planning.

ASPIRE

SuiteSuccess rateEvaluation
Object Swap98.0%Held-out seeds 1-50
Object Task95.0%Held-out seeds 1-50
Goal Swap81.0%Held-out seeds 1-50
Spatial Task60.0%Held-out seeds 1-50
Spatial Swap51.0%Held-out seeds 1-50
Goal Task45.0%Held-out seeds 1-50
Long Task38.3%Long zero-shot transfer (LIBERO-90 library)
Long Swap22.6%Long zero-shot transfer (LIBERO-90 library)

RPent Flash Mode · Task + Swap totals

Exact combined counts printed in the published figure; individual Task/Swap counts are not reported in text. These are not eight-suite component scores.

SuiteSuccess rateSuccess / evaluated
Object89.5%179/200
Spatial74.5%149/200
Goal69.5%139/200
Long57.0%114/200
diff --git a/docs/source-en/rst_source/overview.rst b/docs/source-en/rst_source/overview.rst index 7a44285c4..0075b13fd 100644 --- a/docs/source-en/rst_source/overview.rst +++ b/docs/source-en/rst_source/overview.rst @@ -33,13 +33,13 @@ Compare success rates on LIBERO, LIBERO-PRO, RoboCasa365 Target50, and RoboTwin C2R. Rankings apply to the methods and evaluation coverage shown; see :doc:`benchmarks` for detailed results, configurations, and sources. -.. image:: https://raw.githubusercontent.com/RLinf/misc/29a9d96588fbd9fa01cb4e035658eba028384fb6/rpent/benchmarks/leaderboard-en-light.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/07c25bdb9c036ecc437f8e83c8117a080e4a6d2e/rpent/benchmarks/leaderboard-en-light.png :alt: RPent benchmark results :class: only-light :width: 100% :target: benchmarks.html -.. image:: https://raw.githubusercontent.com/RLinf/misc/29a9d96588fbd9fa01cb4e035658eba028384fb6/rpent/benchmarks/leaderboard-en-dark.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/07c25bdb9c036ecc437f8e83c8117a080e4a6d2e/rpent/benchmarks/leaderboard-en-dark.png :alt: RPent benchmark results :class: only-dark :width: 100% diff --git a/docs/source-zh/rst_source/benchmarks.rst b/docs/source-zh/rst_source/benchmarks.rst index 1ada8814d..bcc773e97 100644 --- a/docs/source-zh/rst_source/benchmarks.rst +++ b/docs/source-zh/rst_source/benchmarks.rst @@ -9,18 +9,18 @@ RPent 排行榜 .. raw:: html - - - - + + + +
+ data-results-url="https://raw.githubusercontent.com/RLinf/misc/07c25bdb9c036ecc437f8e83c8117a080e4a6d2e/rpent/benchmarks/results.json">
-

LIBERO-PRO

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus 4.782.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%
NORA0.0%
OpenVLA0.0%
+

LIBERO-PRO

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus 4.782.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%

LIBERO

方法 / 模型成功率
AtomVLA97.0%
Claude Code / Opus-4.7 / max / reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%

RoboCasa365 · Target50

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning59.20%
Xiaomi-Robotics-157.4%
Codex / GPT-5.5 / xhigh / reasoning57.1%
Claude Code / Opus-4.7 / max / reasoning48.6%
WorldDreamer35.3%
RLDX-130.0%
π0.516.9%
π014.8%

RoboTwin

方法 / 模型成功率
Codex / GPT-5.5 / xhigh / reasoning62.4%
Claude Code / Opus-4.7 / max / reasoning58.4%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%
-

GPT-6 Astra · low · reasoning：LIBERO-PRO 92.63%（741/800）。

-

RPent Flash Mode: 72.63% (581/800). Molmo 仅用于视觉定位。

+

GPT-6 Astra: 92.63% (741/800). RPent Flash Mode / Molmo2-8B: 72.63% (581/800).

+

ASPIRE：六项常规评测；Long 为 LIBERO-90 技能库零样本迁移，未报告八套件 Overall。Flash Mode 使用 Molmo2-8B 进行视觉定位，不用于规划。

ASPIRE

分项成功率评测
Object Swap98.0%保留测试 seed 1-50
Object Task95.0%保留测试 seed 1-50
Goal Swap81.0%保留测试 seed 1-50
Spatial Task60.0%保留测试 seed 1-50
Spatial Swap51.0%保留测试 seed 1-50
Goal Task45.0%保留测试 seed 1-50
Long Task38.3%Long 零样本迁移（LIBERO-90 技能库）
Long Swap22.6%Long 零样本迁移（LIBERO-90 技能库）

RPent Flash Mode · Task + Swap 汇总

采用公开图中明确标注的合计次数；未按柱高估算各 Task/Swap 次数，不能将合计作为单个分项。

套件成功率成功 / 总回合
Object89.5%179/200
Spatial74.5%149/200
Goal69.5%139/200
Long57.0%114/200
diff --git a/docs/source-zh/rst_source/overview.rst b/docs/source-zh/rst_source/overview.rst index 75c983b23..f0d75aca9 100644 --- a/docs/source-zh/rst_source/overview.rst +++ b/docs/source-zh/rst_source/overview.rst @@ -25,13 +25,13 @@ RPent 建立在三条核心设计原则之上: **服务化、标准化、可组 对比 LIBERO、LIBERO-PRO、RoboCasa365 Target50 和 RoboTwin C2R 上的成功率。 排名仅限图中方法及评测范围；完整结果、模型配置和来源见 :doc:`benchmarks`。 -.. image:: https://raw.githubusercontent.com/RLinf/misc/29a9d96588fbd9fa01cb4e035658eba028384fb6/rpent/benchmarks/leaderboard-zh-light.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/07c25bdb9c036ecc437f8e83c8117a080e4a6d2e/rpent/benchmarks/leaderboard-zh-light.png :alt: RPent benchmark results :class: only-light :width: 100% :target: benchmarks.html -.. image:: https://raw.githubusercontent.com/RLinf/misc/29a9d96588fbd9fa01cb4e035658eba028384fb6/rpent/benchmarks/leaderboard-zh-dark.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/07c25bdb9c036ecc437f8e83c8117a080e4a6d2e/rpent/benchmarks/leaderboard-zh-dark.png :alt: RPent benchmark results :class: only-dark :width: 100% From 08a3ba0468d34bf05f4954a0d3d3111c0c796979 Mon Sep 17 00:00:00 2001 From: ye shengjie <184427520+ye1457@users.noreply.github.com> Date: Wed, 16 Sep 2026 16:21:57 +0000 Subject: [PATCH 05/13] docs(benchmarks): rank ASPIRE mean and leave Flash components pending --- README.md | 2 +- README.zh-CN.md | 2 +- docs/source-en/rst_source/benchmarks.rst | 14 +++++++------- docs/source-en/rst_source/overview.rst | 4 ++-- docs/source-zh/rst_source/benchmarks.rst | 14 +++++++------- docs/source-zh/rst_source/overview.rst | 4 ++-- 6 files changed, 20 insertions(+), 20 deletions(-) diff --git a/README.md b/README.md index 1b98c77d5..c2ffb1d9b 100644 --- a/README.md +++ b/README.md @@ -33,7 +33,7 @@ Compare RPent with reference methods on LIBERO, LIBERO-PRO, RoboCasa365 Target50 Codex / GPT-6 Astra / low / reasoning: **92.63% Overall (741/800)** across all eight LIBERO-PRO suites. See the [suite results](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html), including the separately frozen Long and Spatial/Object/Goal memory batches. -[![RPent success-rate comparisons across four benchmarks](https://raw.githubusercontent.com/RLinf/misc/07c25bdb9c036ecc437f8e83c8117a080e4a6d2e/rpent/benchmarks/leaderboard-en-light.png)](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) +[![RPent success-rate comparisons across four benchmarks](https://raw.githubusercontent.com/RLinf/misc/10140de07db9c6a0a8c0e1683dc27351e0e319e2/rpent/benchmarks/leaderboard-en-light.png)](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) ## Who Should Consider Using RPent? diff --git a/README.zh-CN.md b/README.zh-CN.md index 869d5ecd8..befe15df6 100644 --- a/README.zh-CN.md +++ b/README.zh-CN.md @@ -33,7 +33,7 @@ Codex / GPT-6 Astra / low / reasoning 已完成全部八套 LIBERO-PRO，**Overall 92.63%（741/800）**。详见 [套件汇总](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html)，其中 Long 与 Spatial/Object/Goal 分别使用各自冻结的 memory 批次。 -[![RPent 四项基准成功率对比](https://raw.githubusercontent.com/RLinf/misc/07c25bdb9c036ecc437f8e83c8117a080e4a6d2e/rpent/benchmarks/leaderboard-zh-light.png)](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html) +[![RPent 四项基准成功率对比](https://raw.githubusercontent.com/RLinf/misc/10140de07db9c6a0a8c0e1683dc27351e0e319e2/rpent/benchmarks/leaderboard-zh-light.png)](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html) ## 适用用户 diff --git a/docs/source-en/rst_source/benchmarks.rst b/docs/source-en/rst_source/benchmarks.rst index 9ea2a2e28..a15283fc6 100644 --- a/docs/source-en/rst_source/benchmarks.rst +++ b/docs/source-en/rst_source/benchmarks.rst @@ -9,18 +9,18 @@ RPent Leaderboard .. raw:: html - - - - + + + +
+ data-results-url="https://raw.githubusercontent.com/RLinf/misc/10140de07db9c6a0a8c0e1683dc27351e0e319e2/rpent/benchmarks/results.json">
-

LIBERO-PRO

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus 4.782.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%
+

LIBERO-PRO

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus 4.782.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
ASPIRE61.36%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%

LIBERO

Method / modelSuccess rate
AtomVLA97.0%
Claude Code / Opus-4.7 / max / reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%

RoboCasa365 · Target50

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning59.20%
Xiaomi-Robotics-157.4%
Codex / GPT-5.5 / xhigh / reasoning57.1%
Claude Code / Opus-4.7 / max / reasoning48.6%
WorldDreamer35.3%
RLDX-130.0%
π0.516.9%
π014.8%

RoboTwin

Method / modelSuccess rate
Codex / GPT-5.5 / xhigh / reasoning62.4%
Claude Code / Opus-4.7 / max / reasoning58.4%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%

GPT-6 Astra: 92.63% (741/800). RPent Flash Mode / Molmo2-8B: 72.63% (581/800).

-

ASPIRE: six main-suite scores; Long uses zero-shot transfer from LIBERO-90. No eight-suite Overall is reported. Flash Mode uses Molmo2-8B for visual localization, not planning.

ASPIRE

SuiteSuccess rateEvaluation
Object Swap98.0%Held-out seeds 1-50
Object Task95.0%Held-out seeds 1-50
Goal Swap81.0%Held-out seeds 1-50
Spatial Task60.0%Held-out seeds 1-50
Spatial Swap51.0%Held-out seeds 1-50
Goal Task45.0%Held-out seeds 1-50
Long Task38.3%Long zero-shot transfer (LIBERO-90 library)
Long Swap22.6%Long zero-shot transfer (LIBERO-90 library)

RPent Flash Mode · Task + Swap totals

Exact combined counts printed in the published figure; individual Task/Swap counts are not reported in text. These are not eight-suite component scores.

SuiteSuccess rateSuccess / evaluated
Object89.5%179/200
Spatial74.5%149/200
Goal69.5%139/200
Long57.0%114/200
+

ASPIRE: six main-suite evaluations; Long is zero-shot transfer from the LIBERO-90 skill library.

All methods & reported scores

Method / modelOverallSpatial TaskSpatial SwapObject TaskObject SwapGoal TaskGoal SwapLong TaskLong Swap
GPT-6 Astra92.63%100%98%100%99%88%99%85%72%
Opus 4.782.4%————————
RPent Flash Mode / Molmo2-8B72.63%————————
GPT-5.572.1%81.0%69.0%94.0%91.0%75.0%66.0%52.0%49.0%
ASPIRE (eight-score arithmetic mean)61.36%60.0%51.0%95.0%98.0%45.0%81.0%38.3%22.6%
π_RLinf50.0%42.0%59.0%71.0%78.0%45.0%42.0%49.0%14.0%
π0.511.0%1.0%20.0%1.0%17.0%2.0%38.0%1.0%8.0%
AtomVLA6.3%1.0%16.0%0.0%10.0%11.0%2.0%9.0%1.0%
X-VLA3.8%0.0%0.0%8.0%2.0%9.0%1.0%10.0%0.0%
MolmoAct1.5%0.0%0.0%0.0%6.0%0.0%0.0%6.0%0.0%
π00.3%0.0%0.0%0.0%2.0%0.0%0.0%0.0%0.0%
Cap-X—14.0%12.0%18.0%22.0%17.0%26.0%——
RATS—31.0%29.0%63.0%61.0%36.0%43.0%——
diff --git a/docs/source-en/rst_source/overview.rst b/docs/source-en/rst_source/overview.rst index 0075b13fd..ba7c5fada 100644 --- a/docs/source-en/rst_source/overview.rst +++ b/docs/source-en/rst_source/overview.rst @@ -33,13 +33,13 @@ Compare success rates on LIBERO, LIBERO-PRO, RoboCasa365 Target50, and RoboTwin C2R. Rankings apply to the methods and evaluation coverage shown; see :doc:`benchmarks` for detailed results, configurations, and sources. -.. image:: https://raw.githubusercontent.com/RLinf/misc/07c25bdb9c036ecc437f8e83c8117a080e4a6d2e/rpent/benchmarks/leaderboard-en-light.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/10140de07db9c6a0a8c0e1683dc27351e0e319e2/rpent/benchmarks/leaderboard-en-light.png :alt: RPent benchmark results :class: only-light :width: 100% :target: benchmarks.html -.. image:: https://raw.githubusercontent.com/RLinf/misc/07c25bdb9c036ecc437f8e83c8117a080e4a6d2e/rpent/benchmarks/leaderboard-en-dark.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/10140de07db9c6a0a8c0e1683dc27351e0e319e2/rpent/benchmarks/leaderboard-en-dark.png :alt: RPent benchmark results :class: only-dark :width: 100% diff --git a/docs/source-zh/rst_source/benchmarks.rst b/docs/source-zh/rst_source/benchmarks.rst index bcc773e97..6525fcc73 100644 --- a/docs/source-zh/rst_source/benchmarks.rst +++ b/docs/source-zh/rst_source/benchmarks.rst @@ -9,18 +9,18 @@ RPent 排行榜 .. raw:: html - - - - + + + +
+ data-results-url="https://raw.githubusercontent.com/RLinf/misc/10140de07db9c6a0a8c0e1683dc27351e0e319e2/rpent/benchmarks/results.json">
-

LIBERO-PRO

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus 4.782.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%
+

LIBERO-PRO

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus 4.782.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
ASPIRE61.36%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%

LIBERO

方法 / 模型成功率
AtomVLA97.0%
Claude Code / Opus-4.7 / max / reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%

RoboCasa365 · Target50

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning59.20%
Xiaomi-Robotics-157.4%
Codex / GPT-5.5 / xhigh / reasoning57.1%
Claude Code / Opus-4.7 / max / reasoning48.6%
WorldDreamer35.3%
RLDX-130.0%
π0.516.9%
π014.8%

RoboTwin

方法 / 模型成功率
Codex / GPT-5.5 / xhigh / reasoning62.4%
Claude Code / Opus-4.7 / max / reasoning58.4%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%

GPT-6 Astra: 92.63% (741/800). RPent Flash Mode / Molmo2-8B: 72.63% (581/800).

-

ASPIRE：六项常规评测；Long 为 LIBERO-90 技能库零样本迁移，未报告八套件 Overall。Flash Mode 使用 Molmo2-8B 进行视觉定位，不用于规划。

ASPIRE

分项成功率评测
Object Swap98.0%保留测试 seed 1-50
Object Task95.0%保留测试 seed 1-50
Goal Swap81.0%保留测试 seed 1-50
Spatial Task60.0%保留测试 seed 1-50
Spatial Swap51.0%保留测试 seed 1-50
Goal Task45.0%保留测试 seed 1-50
Long Task38.3%Long 零样本迁移（LIBERO-90 技能库）
Long Swap22.6%Long 零样本迁移（LIBERO-90 技能库）

RPent Flash Mode · Task + Swap 汇总

采用公开图中明确标注的合计次数；未按柱高估算各 Task/Swap 次数，不能将合计作为单个分项。

套件成功率成功 / 总回合
Object89.5%179/200
Spatial74.5%149/200
Goal69.5%139/200
Long57.0%114/200
+

ASPIRE：六项常规评测；Long 为 LIBERO-90 技能库零样本迁移。

完整方法与分项成绩

方法 / 模型总体Spatial TaskSpatial SwapObject TaskObject SwapGoal TaskGoal SwapLong TaskLong Swap
GPT-6 Astra92.63%100%98%100%99%88%99%85%72%
Opus 4.782.4%————————
RPent Flash Mode / Molmo2-8B72.63%————————
GPT-5.572.1%81.0%69.0%94.0%91.0%75.0%66.0%52.0%49.0%
ASPIRE（八项算术平均）61.36%60.0%51.0%95.0%98.0%45.0%81.0%38.3%22.6%
π_RLinf50.0%42.0%59.0%71.0%78.0%45.0%42.0%49.0%14.0%
π0.511.0%1.0%20.0%1.0%17.0%2.0%38.0%1.0%8.0%
AtomVLA6.3%1.0%16.0%0.0%10.0%11.0%2.0%9.0%1.0%
X-VLA3.8%0.0%0.0%8.0%2.0%9.0%1.0%10.0%0.0%
MolmoAct1.5%0.0%0.0%0.0%6.0%0.0%0.0%6.0%0.0%
π00.3%0.0%0.0%0.0%2.0%0.0%0.0%0.0%0.0%
Cap-X—14.0%12.0%18.0%22.0%17.0%26.0%——
RATS—31.0%29.0%63.0%61.0%36.0%43.0%——
diff --git a/docs/source-zh/rst_source/overview.rst b/docs/source-zh/rst_source/overview.rst index f0d75aca9..49ebba4cd 100644 --- a/docs/source-zh/rst_source/overview.rst +++ b/docs/source-zh/rst_source/overview.rst @@ -25,13 +25,13 @@ RPent 建立在三条核心设计原则之上: **服务化、标准化、可组 对比 LIBERO、LIBERO-PRO、RoboCasa365 Target50 和 RoboTwin C2R 上的成功率。 排名仅限图中方法及评测范围；完整结果、模型配置和来源见 :doc:`benchmarks`。 -.. image:: https://raw.githubusercontent.com/RLinf/misc/07c25bdb9c036ecc437f8e83c8117a080e4a6d2e/rpent/benchmarks/leaderboard-zh-light.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/10140de07db9c6a0a8c0e1683dc27351e0e319e2/rpent/benchmarks/leaderboard-zh-light.png :alt: RPent benchmark results :class: only-light :width: 100% :target: benchmarks.html -.. image:: https://raw.githubusercontent.com/RLinf/misc/07c25bdb9c036ecc437f8e83c8117a080e4a6d2e/rpent/benchmarks/leaderboard-zh-dark.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/10140de07db9c6a0a8c0e1683dc27351e0e319e2/rpent/benchmarks/leaderboard-zh-dark.png :alt: RPent benchmark results :class: only-dark :width: 100% From 0ade8dc17a4fe2e98ebc141737d8ba002bec91b4 Mon Sep 17 00:00:00 2001 From: ye shengjie <184427520+ye1457@users.noreply.github.com> Date: Wed, 16 Sep 2026 16:38:10 +0000 Subject: [PATCH 06/13] docs(benchmarks): fill verified Flash Mode suite scores --- README.md | 2 +- README.zh-CN.md | 2 +- docs/source-en/rst_source/benchmarks.rst | 12 ++++++------ docs/source-en/rst_source/overview.rst | 4 ++-- docs/source-zh/rst_source/benchmarks.rst | 12 ++++++------ docs/source-zh/rst_source/overview.rst | 4 ++-- 6 files changed, 18 insertions(+), 18 deletions(-) diff --git a/README.md b/README.md index c2ffb1d9b..563b1931f 100644 --- a/README.md +++ b/README.md @@ -33,7 +33,7 @@ Compare RPent with reference methods on LIBERO, LIBERO-PRO, RoboCasa365 Target50 Codex / GPT-6 Astra / low / reasoning: **92.63% Overall (741/800)** across all eight LIBERO-PRO suites. See the [suite results](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html), including the separately frozen Long and Spatial/Object/Goal memory batches. -[![RPent success-rate comparisons across four benchmarks](https://raw.githubusercontent.com/RLinf/misc/10140de07db9c6a0a8c0e1683dc27351e0e319e2/rpent/benchmarks/leaderboard-en-light.png)](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) +[![RPent success-rate comparisons across four benchmarks](https://raw.githubusercontent.com/RLinf/misc/b69d555969600e853d8071eb19112502583b1e06/rpent/benchmarks/leaderboard-en-light.png)](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) ## Who Should Consider Using RPent? diff --git a/README.zh-CN.md b/README.zh-CN.md index befe15df6..e83a2bf54 100644 --- a/README.zh-CN.md +++ b/README.zh-CN.md @@ -33,7 +33,7 @@ Codex / GPT-6 Astra / low / reasoning 已完成全部八套 LIBERO-PRO，**Overall 92.63%（741/800）**。详见 [套件汇总](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html)，其中 Long 与 Spatial/Object/Goal 分别使用各自冻结的 memory 批次。 -[![RPent 四项基准成功率对比](https://raw.githubusercontent.com/RLinf/misc/10140de07db9c6a0a8c0e1683dc27351e0e319e2/rpent/benchmarks/leaderboard-zh-light.png)](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html) +[![RPent 四项基准成功率对比](https://raw.githubusercontent.com/RLinf/misc/b69d555969600e853d8071eb19112502583b1e06/rpent/benchmarks/leaderboard-zh-light.png)](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html) ## 适用用户 diff --git a/docs/source-en/rst_source/benchmarks.rst b/docs/source-en/rst_source/benchmarks.rst index a15283fc6..f3891c619 100644 --- a/docs/source-en/rst_source/benchmarks.rst +++ b/docs/source-en/rst_source/benchmarks.rst @@ -9,18 +9,18 @@ RPent Leaderboard .. raw:: html - - - - + + + +
+ data-results-url="https://raw.githubusercontent.com/RLinf/misc/b69d555969600e853d8071eb19112502583b1e06/rpent/benchmarks/results.json">

LIBERO-PRO

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus 4.782.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
ASPIRE61.36%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%

LIBERO

Method / modelSuccess rate
AtomVLA97.0%
Claude Code / Opus-4.7 / max / reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%

RoboCasa365 · Target50

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning59.20%
Xiaomi-Robotics-157.4%
Codex / GPT-5.5 / xhigh / reasoning57.1%
Claude Code / Opus-4.7 / max / reasoning48.6%
WorldDreamer35.3%
RLDX-130.0%
π0.516.9%
π014.8%

RoboTwin

Method / modelSuccess rate
Codex / GPT-5.5 / xhigh / reasoning62.4%
Claude Code / Opus-4.7 / max / reasoning58.4%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%

GPT-6 Astra: 92.63% (741/800). RPent Flash Mode / Molmo2-8B: 72.63% (581/800).

-

ASPIRE: six main-suite evaluations; Long is zero-shot transfer from the LIBERO-90 skill library.

All methods & reported scores

Method / modelOverallSpatial TaskSpatial SwapObject TaskObject SwapGoal TaskGoal SwapLong TaskLong Swap
GPT-6 Astra92.63%100%98%100%99%88%99%85%72%
Opus 4.782.4%————————
RPent Flash Mode / Molmo2-8B72.63%————————
GPT-5.572.1%81.0%69.0%94.0%91.0%75.0%66.0%52.0%49.0%
ASPIRE (eight-score arithmetic mean)61.36%60.0%51.0%95.0%98.0%45.0%81.0%38.3%22.6%
π_RLinf50.0%42.0%59.0%71.0%78.0%45.0%42.0%49.0%14.0%
π0.511.0%1.0%20.0%1.0%17.0%2.0%38.0%1.0%8.0%
AtomVLA6.3%1.0%16.0%0.0%10.0%11.0%2.0%9.0%1.0%
X-VLA3.8%0.0%0.0%8.0%2.0%9.0%1.0%10.0%0.0%
MolmoAct1.5%0.0%0.0%0.0%6.0%0.0%0.0%6.0%0.0%
π00.3%0.0%0.0%0.0%2.0%0.0%0.0%0.0%0.0%
Cap-X—14.0%12.0%18.0%22.0%17.0%26.0%——
RATS—31.0%29.0%63.0%61.0%36.0%43.0%——
+

ASPIRE: six main-suite evaluations; Long is zero-shot transfer from the LIBERO-90 skill library.

All methods & reported scores

Method / modelOverallSpatial TaskSpatial SwapObject TaskObject SwapGoal TaskGoal SwapLong TaskLong Swap
GPT-6 Astra92.63%100%98%100%99%88%99%85%72%
Opus 4.782.4%————————
RPent Flash Mode / Molmo2-8B72.63%79.00%70.00%86.00%93.00%74.00%65.00%60.00%54.00%
GPT-5.572.1%81.0%69.0%94.0%91.0%75.0%66.0%52.0%49.0%
ASPIRE (eight-score arithmetic mean)61.36%60.0%51.0%95.0%98.0%45.0%81.0%38.3%22.6%
π_RLinf50.0%42.0%59.0%71.0%78.0%45.0%42.0%49.0%14.0%
π0.511.0%1.0%20.0%1.0%17.0%2.0%38.0%1.0%8.0%
AtomVLA6.3%1.0%16.0%0.0%10.0%11.0%2.0%9.0%1.0%
X-VLA3.8%0.0%0.0%8.0%2.0%9.0%1.0%10.0%0.0%
MolmoAct1.5%0.0%0.0%0.0%6.0%0.0%0.0%6.0%0.0%
π00.3%0.0%0.0%0.0%2.0%0.0%0.0%0.0%0.0%
Cap-X—14.0%12.0%18.0%22.0%17.0%26.0%——
RATS—31.0%29.0%63.0%61.0%36.0%43.0%——
diff --git a/docs/source-en/rst_source/overview.rst b/docs/source-en/rst_source/overview.rst index ba7c5fada..b37e0be67 100644 --- a/docs/source-en/rst_source/overview.rst +++ b/docs/source-en/rst_source/overview.rst @@ -33,13 +33,13 @@ Compare success rates on LIBERO, LIBERO-PRO, RoboCasa365 Target50, and RoboTwin C2R. Rankings apply to the methods and evaluation coverage shown; see :doc:`benchmarks` for detailed results, configurations, and sources. -.. image:: https://raw.githubusercontent.com/RLinf/misc/10140de07db9c6a0a8c0e1683dc27351e0e319e2/rpent/benchmarks/leaderboard-en-light.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/b69d555969600e853d8071eb19112502583b1e06/rpent/benchmarks/leaderboard-en-light.png :alt: RPent benchmark results :class: only-light :width: 100% :target: benchmarks.html -.. image:: https://raw.githubusercontent.com/RLinf/misc/10140de07db9c6a0a8c0e1683dc27351e0e319e2/rpent/benchmarks/leaderboard-en-dark.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/b69d555969600e853d8071eb19112502583b1e06/rpent/benchmarks/leaderboard-en-dark.png :alt: RPent benchmark results :class: only-dark :width: 100% diff --git a/docs/source-zh/rst_source/benchmarks.rst b/docs/source-zh/rst_source/benchmarks.rst index 6525fcc73..8019eb9a6 100644 --- a/docs/source-zh/rst_source/benchmarks.rst +++ b/docs/source-zh/rst_source/benchmarks.rst @@ -9,18 +9,18 @@ RPent 排行榜 .. raw:: html - - - - + + + +
+ data-results-url="https://raw.githubusercontent.com/RLinf/misc/b69d555969600e853d8071eb19112502583b1e06/rpent/benchmarks/results.json">

LIBERO-PRO

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus 4.782.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
ASPIRE61.36%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%

LIBERO

方法 / 模型成功率
AtomVLA97.0%
Claude Code / Opus-4.7 / max / reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%

RoboCasa365 · Target50

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning59.20%
Xiaomi-Robotics-157.4%
Codex / GPT-5.5 / xhigh / reasoning57.1%
Claude Code / Opus-4.7 / max / reasoning48.6%
WorldDreamer35.3%
RLDX-130.0%
π0.516.9%
π014.8%

RoboTwin

方法 / 模型成功率
Codex / GPT-5.5 / xhigh / reasoning62.4%
Claude Code / Opus-4.7 / max / reasoning58.4%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%

GPT-6 Astra: 92.63% (741/800). RPent Flash Mode / Molmo2-8B: 72.63% (581/800).

-

ASPIRE：六项常规评测；Long 为 LIBERO-90 技能库零样本迁移。

完整方法与分项成绩

方法 / 模型总体Spatial TaskSpatial SwapObject TaskObject SwapGoal TaskGoal SwapLong TaskLong Swap
GPT-6 Astra92.63%100%98%100%99%88%99%85%72%
Opus 4.782.4%————————
RPent Flash Mode / Molmo2-8B72.63%————————
GPT-5.572.1%81.0%69.0%94.0%91.0%75.0%66.0%52.0%49.0%
ASPIRE（八项算术平均）61.36%60.0%51.0%95.0%98.0%45.0%81.0%38.3%22.6%
π_RLinf50.0%42.0%59.0%71.0%78.0%45.0%42.0%49.0%14.0%
π0.511.0%1.0%20.0%1.0%17.0%2.0%38.0%1.0%8.0%
AtomVLA6.3%1.0%16.0%0.0%10.0%11.0%2.0%9.0%1.0%
X-VLA3.8%0.0%0.0%8.0%2.0%9.0%1.0%10.0%0.0%
MolmoAct1.5%0.0%0.0%0.0%6.0%0.0%0.0%6.0%0.0%
π00.3%0.0%0.0%0.0%2.0%0.0%0.0%0.0%0.0%
Cap-X—14.0%12.0%18.0%22.0%17.0%26.0%——
RATS—31.0%29.0%63.0%61.0%36.0%43.0%——
+

ASPIRE：六项常规评测；Long 为 LIBERO-90 技能库零样本迁移。

完整方法与分项成绩

方法 / 模型总体Spatial TaskSpatial SwapObject TaskObject SwapGoal TaskGoal SwapLong TaskLong Swap
GPT-6 Astra92.63%100%98%100%99%88%99%85%72%
Opus 4.782.4%————————
RPent Flash Mode / Molmo2-8B72.63%79.00%70.00%86.00%93.00%74.00%65.00%60.00%54.00%
GPT-5.572.1%81.0%69.0%94.0%91.0%75.0%66.0%52.0%49.0%
ASPIRE（八项算术平均）61.36%60.0%51.0%95.0%98.0%45.0%81.0%38.3%22.6%
π_RLinf50.0%42.0%59.0%71.0%78.0%45.0%42.0%49.0%14.0%
π0.511.0%1.0%20.0%1.0%17.0%2.0%38.0%1.0%8.0%
AtomVLA6.3%1.0%16.0%0.0%10.0%11.0%2.0%9.0%1.0%
X-VLA3.8%0.0%0.0%8.0%2.0%9.0%1.0%10.0%0.0%
MolmoAct1.5%0.0%0.0%0.0%6.0%0.0%0.0%6.0%0.0%
π00.3%0.0%0.0%0.0%2.0%0.0%0.0%0.0%0.0%
Cap-X—14.0%12.0%18.0%22.0%17.0%26.0%——
RATS—31.0%29.0%63.0%61.0%36.0%43.0%——
diff --git a/docs/source-zh/rst_source/overview.rst b/docs/source-zh/rst_source/overview.rst index 49ebba4cd..82c507e9b 100644 --- a/docs/source-zh/rst_source/overview.rst +++ b/docs/source-zh/rst_source/overview.rst @@ -25,13 +25,13 @@ RPent 建立在三条核心设计原则之上: **服务化、标准化、可组 对比 LIBERO、LIBERO-PRO、RoboCasa365 Target50 和 RoboTwin C2R 上的成功率。 排名仅限图中方法及评测范围；完整结果、模型配置和来源见 :doc:`benchmarks`。 -.. image:: https://raw.githubusercontent.com/RLinf/misc/10140de07db9c6a0a8c0e1683dc27351e0e319e2/rpent/benchmarks/leaderboard-zh-light.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/b69d555969600e853d8071eb19112502583b1e06/rpent/benchmarks/leaderboard-zh-light.png :alt: RPent benchmark results :class: only-light :width: 100% :target: benchmarks.html -.. image:: https://raw.githubusercontent.com/RLinf/misc/10140de07db9c6a0a8c0e1683dc27351e0e319e2/rpent/benchmarks/leaderboard-zh-dark.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/b69d555969600e853d8071eb19112502583b1e06/rpent/benchmarks/leaderboard-zh-dark.png :alt: RPent benchmark results :class: only-dark :width: 100% From db31f2665229a2a601eaf2d2fa77dc3b879ca9eb Mon Sep 17 00:00:00 2001 From: ye shengjie <184427520+ye1457@users.noreply.github.com> Date: Thu, 17 Sep 2026 03:58:15 +0000 Subject: [PATCH 07/13] docs(benchmarks): sync text-only leaderboard and Opus reasoning labels --- README.md | 2 +- README.zh-CN.md | 2 +- docs/source-en/rst_source/benchmarks.rst | 20 ++++++++++---------- docs/source-en/rst_source/overview.rst | 4 ++-- docs/source-zh/rst_source/benchmarks.rst | 20 ++++++++++---------- docs/source-zh/rst_source/overview.rst | 4 ++-- 6 files changed, 26 insertions(+), 26 deletions(-) diff --git a/README.md b/README.md index 563b1931f..a8c43d7ba 100644 --- a/README.md +++ b/README.md @@ -33,7 +33,7 @@ Compare RPent with reference methods on LIBERO, LIBERO-PRO, RoboCasa365 Target50 Codex / GPT-6 Astra / low / reasoning: **92.63% Overall (741/800)** across all eight LIBERO-PRO suites. See the [suite results](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html), including the separately frozen Long and Spatial/Object/Goal memory batches. -[![RPent success-rate comparisons across four benchmarks](https://raw.githubusercontent.com/RLinf/misc/b69d555969600e853d8071eb19112502583b1e06/rpent/benchmarks/leaderboard-en-light.png)](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) +[![RPent success-rate comparisons across four benchmarks](https://raw.githubusercontent.com/RLinf/misc/36f26c2d0cdb2e00b78b6ea3166e4c265c45b51f/rpent/benchmarks/leaderboard-en-light.png)](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) ## Who Should Consider Using RPent? diff --git a/README.zh-CN.md b/README.zh-CN.md index e83a2bf54..900351e72 100644 --- a/README.zh-CN.md +++ b/README.zh-CN.md @@ -33,7 +33,7 @@ Codex / GPT-6 Astra / low / reasoning 已完成全部八套 LIBERO-PRO，**Overall 92.63%（741/800）**。详见 [套件汇总](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html)，其中 Long 与 Spatial/Object/Goal 分别使用各自冻结的 memory 批次。 -[![RPent 四项基准成功率对比](https://raw.githubusercontent.com/RLinf/misc/b69d555969600e853d8071eb19112502583b1e06/rpent/benchmarks/leaderboard-zh-light.png)](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html) +[![RPent 四项基准成功率对比](https://raw.githubusercontent.com/RLinf/misc/36f26c2d0cdb2e00b78b6ea3166e4c265c45b51f/rpent/benchmarks/leaderboard-zh-light.png)](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html) ## 适用用户 diff --git a/docs/source-en/rst_source/benchmarks.rst b/docs/source-en/rst_source/benchmarks.rst index f3891c619..189085a74 100644 --- a/docs/source-en/rst_source/benchmarks.rst +++ b/docs/source-en/rst_source/benchmarks.rst @@ -9,18 +9,18 @@ RPent Leaderboard .. raw:: html - - - - + + + +
+ data-results-url="https://raw.githubusercontent.com/RLinf/misc/36f26c2d0cdb2e00b78b6ea3166e4c265c45b51f/rpent/benchmarks/results.json">
-

LIBERO-PRO

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus 4.782.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
ASPIRE61.36%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%
-

LIBERO

Method / modelSuccess rate
AtomVLA97.0%
Claude Code / Opus-4.7 / max / reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%
-

RoboCasa365 · Target50

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning59.20%
Xiaomi-Robotics-157.4%
Codex / GPT-5.5 / xhigh / reasoning57.1%
Claude Code / Opus-4.7 / max / reasoning48.6%
WorldDreamer35.3%
RLDX-130.0%
π0.516.9%
π014.8%
-

RoboTwin

Method / modelSuccess rate
Codex / GPT-5.5 / xhigh / reasoning62.4%
Claude Code / Opus-4.7 / max / reasoning58.4%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%
+

LIBERO-PRO

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus-4.7 / max.reasoning82.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
ASPIRE61.36%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%
+

LIBERO

Method / modelSuccess rate
AtomVLA97.0%
Claude Code / Opus-4.7 / max.reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%
+

RoboCasa365 · Target50

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning59.20%
Xiaomi-Robotics-157.4%
Codex / GPT-5.5 / xhigh / reasoning57.1%
Claude Code / Opus-4.7 / max.reasoning48.6%
WorldDreamer35.3%
RLDX-130.0%
π0.516.9%
π014.8%
+

RoboTwin

Method / modelSuccess rate
Codex / GPT-5.5 / xhigh / reasoning62.4%
Claude Code / Opus-4.7 / max.reasoning58.4%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%

GPT-6 Astra: 92.63% (741/800). RPent Flash Mode / Molmo2-8B: 72.63% (581/800).

-

ASPIRE: six main-suite evaluations; Long is zero-shot transfer from the LIBERO-90 skill library.

All methods & reported scores

Method / modelOverallSpatial TaskSpatial SwapObject TaskObject SwapGoal TaskGoal SwapLong TaskLong Swap
GPT-6 Astra92.63%100%98%100%99%88%99%85%72%
Opus 4.782.4%————————
RPent Flash Mode / Molmo2-8B72.63%79.00%70.00%86.00%93.00%74.00%65.00%60.00%54.00%
GPT-5.572.1%81.0%69.0%94.0%91.0%75.0%66.0%52.0%49.0%
ASPIRE (eight-score arithmetic mean)61.36%60.0%51.0%95.0%98.0%45.0%81.0%38.3%22.6%
π_RLinf50.0%42.0%59.0%71.0%78.0%45.0%42.0%49.0%14.0%
π0.511.0%1.0%20.0%1.0%17.0%2.0%38.0%1.0%8.0%
AtomVLA6.3%1.0%16.0%0.0%10.0%11.0%2.0%9.0%1.0%
X-VLA3.8%0.0%0.0%8.0%2.0%9.0%1.0%10.0%0.0%
MolmoAct1.5%0.0%0.0%0.0%6.0%0.0%0.0%6.0%0.0%
π00.3%0.0%0.0%0.0%2.0%0.0%0.0%0.0%0.0%
Cap-X—14.0%12.0%18.0%22.0%17.0%26.0%——
RATS—31.0%29.0%63.0%61.0%36.0%43.0%——
+

ASPIRE: six main-suite evaluations; Long is zero-shot transfer from the LIBERO-90 skill library.

All methods & reported scores

Method / modelOverallSpatial TaskSpatial SwapObject TaskObject SwapGoal TaskGoal SwapLong TaskLong Swap
GPT-6 Astra92.63%100%98%100%99%88%99%85%72%
Opus-4.7 / max.reasoning82.4%————————
RPent Flash Mode / Molmo2-8B72.63%79.00%70.00%86.00%93.00%74.00%65.00%60.00%54.00%
GPT-5.572.1%81.0%69.0%94.0%91.0%75.0%66.0%52.0%49.0%
ASPIRE (eight-score arithmetic mean)61.36%60.0%51.0%95.0%98.0%45.0%81.0%38.3%22.6%
π_RLinf50.0%42.0%59.0%71.0%78.0%45.0%42.0%49.0%14.0%
π0.511.0%1.0%20.0%1.0%17.0%2.0%38.0%1.0%8.0%
AtomVLA6.3%1.0%16.0%0.0%10.0%11.0%2.0%9.0%1.0%
X-VLA3.8%0.0%0.0%8.0%2.0%9.0%1.0%10.0%0.0%
MolmoAct1.5%0.0%0.0%0.0%6.0%0.0%0.0%6.0%0.0%
π00.3%0.0%0.0%0.0%2.0%0.0%0.0%0.0%0.0%
Cap-X—14.0%12.0%18.0%22.0%17.0%26.0%——
RATS—31.0%29.0%63.0%61.0%36.0%43.0%——
diff --git a/docs/source-en/rst_source/overview.rst b/docs/source-en/rst_source/overview.rst index b37e0be67..125402c58 100644 --- a/docs/source-en/rst_source/overview.rst +++ b/docs/source-en/rst_source/overview.rst @@ -33,13 +33,13 @@ Compare success rates on LIBERO, LIBERO-PRO, RoboCasa365 Target50, and RoboTwin C2R. Rankings apply to the methods and evaluation coverage shown; see :doc:`benchmarks` for detailed results, configurations, and sources. -.. image:: https://raw.githubusercontent.com/RLinf/misc/b69d555969600e853d8071eb19112502583b1e06/rpent/benchmarks/leaderboard-en-light.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/36f26c2d0cdb2e00b78b6ea3166e4c265c45b51f/rpent/benchmarks/leaderboard-en-light.png :alt: RPent benchmark results :class: only-light :width: 100% :target: benchmarks.html -.. image:: https://raw.githubusercontent.com/RLinf/misc/b69d555969600e853d8071eb19112502583b1e06/rpent/benchmarks/leaderboard-en-dark.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/36f26c2d0cdb2e00b78b6ea3166e4c265c45b51f/rpent/benchmarks/leaderboard-en-dark.png :alt: RPent benchmark results :class: only-dark :width: 100% diff --git a/docs/source-zh/rst_source/benchmarks.rst b/docs/source-zh/rst_source/benchmarks.rst index 8019eb9a6..a144397d3 100644 --- a/docs/source-zh/rst_source/benchmarks.rst +++ b/docs/source-zh/rst_source/benchmarks.rst @@ -9,18 +9,18 @@ RPent 排行榜 .. raw:: html - - - - + + + +
+ data-results-url="https://raw.githubusercontent.com/RLinf/misc/36f26c2d0cdb2e00b78b6ea3166e4c265c45b51f/rpent/benchmarks/results.json">
-

LIBERO-PRO

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus 4.782.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
ASPIRE61.36%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%
-

LIBERO

方法 / 模型成功率
AtomVLA97.0%
Claude Code / Opus-4.7 / max / reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%
-

RoboCasa365 · Target50

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning59.20%
Xiaomi-Robotics-157.4%
Codex / GPT-5.5 / xhigh / reasoning57.1%
Claude Code / Opus-4.7 / max / reasoning48.6%
WorldDreamer35.3%
RLDX-130.0%
π0.516.9%
π014.8%
-

RoboTwin

方法 / 模型成功率
Codex / GPT-5.5 / xhigh / reasoning62.4%
Claude Code / Opus-4.7 / max / reasoning58.4%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%
+

LIBERO-PRO

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus-4.7 / max.reasoning82.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
ASPIRE61.36%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%
+

LIBERO

方法 / 模型成功率
AtomVLA97.0%
Claude Code / Opus-4.7 / max.reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%
+

RoboCasa365 · Target50

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning59.20%
Xiaomi-Robotics-157.4%
Codex / GPT-5.5 / xhigh / reasoning57.1%
Claude Code / Opus-4.7 / max.reasoning48.6%
WorldDreamer35.3%
RLDX-130.0%
π0.516.9%
π014.8%
+

RoboTwin

方法 / 模型成功率
Codex / GPT-5.5 / xhigh / reasoning62.4%
Claude Code / Opus-4.7 / max.reasoning58.4%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%

GPT-6 Astra: 92.63% (741/800). RPent Flash Mode / Molmo2-8B: 72.63% (581/800).

-

ASPIRE：六项常规评测；Long 为 LIBERO-90 技能库零样本迁移。

完整方法与分项成绩

方法 / 模型总体Spatial TaskSpatial SwapObject TaskObject SwapGoal TaskGoal SwapLong TaskLong Swap
GPT-6 Astra92.63%100%98%100%99%88%99%85%72%
Opus 4.782.4%————————
RPent Flash Mode / Molmo2-8B72.63%79.00%70.00%86.00%93.00%74.00%65.00%60.00%54.00%
GPT-5.572.1%81.0%69.0%94.0%91.0%75.0%66.0%52.0%49.0%
ASPIRE（八项算术平均）61.36%60.0%51.0%95.0%98.0%45.0%81.0%38.3%22.6%
π_RLinf50.0%42.0%59.0%71.0%78.0%45.0%42.0%49.0%14.0%
π0.511.0%1.0%20.0%1.0%17.0%2.0%38.0%1.0%8.0%
AtomVLA6.3%1.0%16.0%0.0%10.0%11.0%2.0%9.0%1.0%
X-VLA3.8%0.0%0.0%8.0%2.0%9.0%1.0%10.0%0.0%
MolmoAct1.5%0.0%0.0%0.0%6.0%0.0%0.0%6.0%0.0%
π00.3%0.0%0.0%0.0%2.0%0.0%0.0%0.0%0.0%
Cap-X—14.0%12.0%18.0%22.0%17.0%26.0%——
RATS—31.0%29.0%63.0%61.0%36.0%43.0%——
+

ASPIRE：六项常规评测；Long 为 LIBERO-90 技能库零样本迁移。

完整方法与分项成绩

方法 / 模型总体Spatial TaskSpatial SwapObject TaskObject SwapGoal TaskGoal SwapLong TaskLong Swap
GPT-6 Astra92.63%100%98%100%99%88%99%85%72%
Opus-4.7 / max.reasoning82.4%————————
RPent Flash Mode / Molmo2-8B72.63%79.00%70.00%86.00%93.00%74.00%65.00%60.00%54.00%
GPT-5.572.1%81.0%69.0%94.0%91.0%75.0%66.0%52.0%49.0%
ASPIRE（八项算术平均）61.36%60.0%51.0%95.0%98.0%45.0%81.0%38.3%22.6%
π_RLinf50.0%42.0%59.0%71.0%78.0%45.0%42.0%49.0%14.0%
π0.511.0%1.0%20.0%1.0%17.0%2.0%38.0%1.0%8.0%
AtomVLA6.3%1.0%16.0%0.0%10.0%11.0%2.0%9.0%1.0%
X-VLA3.8%0.0%0.0%8.0%2.0%9.0%1.0%10.0%0.0%
MolmoAct1.5%0.0%0.0%0.0%6.0%0.0%0.0%6.0%0.0%
π00.3%0.0%0.0%0.0%2.0%0.0%0.0%0.0%0.0%
Cap-X—14.0%12.0%18.0%22.0%17.0%26.0%——
RATS—31.0%29.0%63.0%61.0%36.0%43.0%——
diff --git a/docs/source-zh/rst_source/overview.rst b/docs/source-zh/rst_source/overview.rst index 82c507e9b..cca6de3a3 100644 --- a/docs/source-zh/rst_source/overview.rst +++ b/docs/source-zh/rst_source/overview.rst @@ -25,13 +25,13 @@ RPent 建立在三条核心设计原则之上: **服务化、标准化、可组 对比 LIBERO、LIBERO-PRO、RoboCasa365 Target50 和 RoboTwin C2R 上的成功率。 排名仅限图中方法及评测范围；完整结果、模型配置和来源见 :doc:`benchmarks`。 -.. image:: https://raw.githubusercontent.com/RLinf/misc/b69d555969600e853d8071eb19112502583b1e06/rpent/benchmarks/leaderboard-zh-light.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/36f26c2d0cdb2e00b78b6ea3166e4c265c45b51f/rpent/benchmarks/leaderboard-zh-light.png :alt: RPent benchmark results :class: only-light :width: 100% :target: benchmarks.html -.. image:: https://raw.githubusercontent.com/RLinf/misc/b69d555969600e853d8071eb19112502583b1e06/rpent/benchmarks/leaderboard-zh-dark.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/36f26c2d0cdb2e00b78b6ea3166e4c265c45b51f/rpent/benchmarks/leaderboard-zh-dark.png :alt: RPent benchmark results :class: only-dark :width: 100% From c5c973fcca9349de5a956645d352482a77e43851 Mon Sep 17 00:00:00 2001 From: ye shengjie <184427520+ye1457@users.noreply.github.com> Date: Thu, 17 Sep 2026 04:09:08 +0000 Subject: [PATCH 08/13] docs(benchmarks): keep ASPIRE zero-shot explanation below results --- README.md | 2 +- README.zh-CN.md | 2 +- docs/source-en/rst_source/benchmarks.rst | 12 ++++++------ docs/source-en/rst_source/overview.rst | 4 ++-- docs/source-zh/rst_source/benchmarks.rst | 12 ++++++------ docs/source-zh/rst_source/overview.rst | 4 ++-- 6 files changed, 18 insertions(+), 18 deletions(-) diff --git a/README.md b/README.md index a8c43d7ba..c84ae8249 100644 --- a/README.md +++ b/README.md @@ -33,7 +33,7 @@ Compare RPent with reference methods on LIBERO, LIBERO-PRO, RoboCasa365 Target50 Codex / GPT-6 Astra / low / reasoning: **92.63% Overall (741/800)** across all eight LIBERO-PRO suites. See the [suite results](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html), including the separately frozen Long and Spatial/Object/Goal memory batches. -[![RPent success-rate comparisons across four benchmarks](https://raw.githubusercontent.com/RLinf/misc/36f26c2d0cdb2e00b78b6ea3166e4c265c45b51f/rpent/benchmarks/leaderboard-en-light.png)](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) +[![RPent success-rate comparisons across four benchmarks](https://raw.githubusercontent.com/RLinf/misc/fbb158d608554595dcd2d4aae6fcc42e8f71c86b/rpent/benchmarks/leaderboard-en-light.png)](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) ## Who Should Consider Using RPent? diff --git a/README.zh-CN.md b/README.zh-CN.md index 900351e72..921de9920 100644 --- a/README.zh-CN.md +++ b/README.zh-CN.md @@ -33,7 +33,7 @@ Codex / GPT-6 Astra / low / reasoning 已完成全部八套 LIBERO-PRO，**Overall 92.63%（741/800）**。详见 [套件汇总](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html)，其中 Long 与 Spatial/Object/Goal 分别使用各自冻结的 memory 批次。 -[![RPent 四项基准成功率对比](https://raw.githubusercontent.com/RLinf/misc/36f26c2d0cdb2e00b78b6ea3166e4c265c45b51f/rpent/benchmarks/leaderboard-zh-light.png)](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html) +[![RPent 四项基准成功率对比](https://raw.githubusercontent.com/RLinf/misc/fbb158d608554595dcd2d4aae6fcc42e8f71c86b/rpent/benchmarks/leaderboard-zh-light.png)](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html) ## 适用用户 diff --git a/docs/source-en/rst_source/benchmarks.rst b/docs/source-en/rst_source/benchmarks.rst index 189085a74..1a558c010 100644 --- a/docs/source-en/rst_source/benchmarks.rst +++ b/docs/source-en/rst_source/benchmarks.rst @@ -9,18 +9,18 @@ RPent Leaderboard .. raw:: html - - - - + + + +
+ data-results-url="https://raw.githubusercontent.com/RLinf/misc/fbb158d608554595dcd2d4aae6fcc42e8f71c86b/rpent/benchmarks/results.json">

LIBERO-PRO

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus-4.7 / max.reasoning82.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
ASPIRE61.36%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%

LIBERO

Method / modelSuccess rate
AtomVLA97.0%
Claude Code / Opus-4.7 / max.reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%

RoboCasa365 · Target50

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning59.20%
Xiaomi-Robotics-157.4%
Codex / GPT-5.5 / xhigh / reasoning57.1%
Claude Code / Opus-4.7 / max.reasoning48.6%
WorldDreamer35.3%
RLDX-130.0%
π0.516.9%
π014.8%

RoboTwin

Method / modelSuccess rate
Codex / GPT-5.5 / xhigh / reasoning62.4%
Claude Code / Opus-4.7 / max.reasoning58.4%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%

GPT-6 Astra: 92.63% (741/800). RPent Flash Mode / Molmo2-8B: 72.63% (581/800).

-

ASPIRE: six main-suite evaluations; Long is zero-shot transfer from the LIBERO-90 skill library.

All methods & reported scores

Method / modelOverallSpatial TaskSpatial SwapObject TaskObject SwapGoal TaskGoal SwapLong TaskLong Swap
GPT-6 Astra92.63%100%98%100%99%88%99%85%72%
Opus-4.7 / max.reasoning82.4%————————
RPent Flash Mode / Molmo2-8B72.63%79.00%70.00%86.00%93.00%74.00%65.00%60.00%54.00%
GPT-5.572.1%81.0%69.0%94.0%91.0%75.0%66.0%52.0%49.0%
ASPIRE (eight-score arithmetic mean)61.36%60.0%51.0%95.0%98.0%45.0%81.0%38.3%22.6%
π_RLinf50.0%42.0%59.0%71.0%78.0%45.0%42.0%49.0%14.0%
π0.511.0%1.0%20.0%1.0%17.0%2.0%38.0%1.0%8.0%
AtomVLA6.3%1.0%16.0%0.0%10.0%11.0%2.0%9.0%1.0%
X-VLA3.8%0.0%0.0%8.0%2.0%9.0%1.0%10.0%0.0%
MolmoAct1.5%0.0%0.0%0.0%6.0%0.0%0.0%6.0%0.0%
π00.3%0.0%0.0%0.0%2.0%0.0%0.0%0.0%0.0%
Cap-X—14.0%12.0%18.0%22.0%17.0%26.0%——
RATS—31.0%29.0%63.0%61.0%36.0%43.0%——
+

All methods & reported scores

Method / modelOverallSpatial TaskSpatial SwapObject TaskObject SwapGoal TaskGoal SwapLong TaskLong Swap
GPT-6 Astra92.63%100%98%100%99%88%99%85%72%
Opus-4.7 / max.reasoning82.4%————————
RPent Flash Mode / Molmo2-8B72.63%79.00%70.00%86.00%93.00%74.00%65.00%60.00%54.00%
GPT-5.572.1%81.0%69.0%94.0%91.0%75.0%66.0%52.0%49.0%
ASPIRE61.36%60.0%51.0%95.0%98.0%45.0%81.0%38.3%22.6%
π_RLinf50.0%42.0%59.0%71.0%78.0%45.0%42.0%49.0%14.0%
π0.511.0%1.0%20.0%1.0%17.0%2.0%38.0%1.0%8.0%
AtomVLA6.3%1.0%16.0%0.0%10.0%11.0%2.0%9.0%1.0%
X-VLA3.8%0.0%0.0%8.0%2.0%9.0%1.0%10.0%0.0%
MolmoAct1.5%0.0%0.0%0.0%6.0%0.0%0.0%6.0%0.0%
π00.3%0.0%0.0%0.0%2.0%0.0%0.0%0.0%0.0%
Cap-X—14.0%12.0%18.0%22.0%17.0%26.0%——
RATS—31.0%29.0%63.0%61.0%36.0%43.0%——

ASPIRE: Long Task and Long Swap use zero-shot transfer from the LIBERO-90 skill library.

diff --git a/docs/source-en/rst_source/overview.rst b/docs/source-en/rst_source/overview.rst index 125402c58..f9dc584e4 100644 --- a/docs/source-en/rst_source/overview.rst +++ b/docs/source-en/rst_source/overview.rst @@ -33,13 +33,13 @@ Compare success rates on LIBERO, LIBERO-PRO, RoboCasa365 Target50, and RoboTwin C2R. Rankings apply to the methods and evaluation coverage shown; see :doc:`benchmarks` for detailed results, configurations, and sources. -.. image:: https://raw.githubusercontent.com/RLinf/misc/36f26c2d0cdb2e00b78b6ea3166e4c265c45b51f/rpent/benchmarks/leaderboard-en-light.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/fbb158d608554595dcd2d4aae6fcc42e8f71c86b/rpent/benchmarks/leaderboard-en-light.png :alt: RPent benchmark results :class: only-light :width: 100% :target: benchmarks.html -.. image:: https://raw.githubusercontent.com/RLinf/misc/36f26c2d0cdb2e00b78b6ea3166e4c265c45b51f/rpent/benchmarks/leaderboard-en-dark.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/fbb158d608554595dcd2d4aae6fcc42e8f71c86b/rpent/benchmarks/leaderboard-en-dark.png :alt: RPent benchmark results :class: only-dark :width: 100% diff --git a/docs/source-zh/rst_source/benchmarks.rst b/docs/source-zh/rst_source/benchmarks.rst index a144397d3..d6a18dc1c 100644 --- a/docs/source-zh/rst_source/benchmarks.rst +++ b/docs/source-zh/rst_source/benchmarks.rst @@ -9,18 +9,18 @@ RPent 排行榜 .. raw:: html - - - - + + + +
+ data-results-url="https://raw.githubusercontent.com/RLinf/misc/fbb158d608554595dcd2d4aae6fcc42e8f71c86b/rpent/benchmarks/results.json">

LIBERO-PRO

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus-4.7 / max.reasoning82.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
ASPIRE61.36%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%

LIBERO

方法 / 模型成功率
AtomVLA97.0%
Claude Code / Opus-4.7 / max.reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%

RoboCasa365 · Target50

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning59.20%
Xiaomi-Robotics-157.4%
Codex / GPT-5.5 / xhigh / reasoning57.1%
Claude Code / Opus-4.7 / max.reasoning48.6%
WorldDreamer35.3%
RLDX-130.0%
π0.516.9%
π014.8%

RoboTwin

方法 / 模型成功率
Codex / GPT-5.5 / xhigh / reasoning62.4%
Claude Code / Opus-4.7 / max.reasoning58.4%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%

GPT-6 Astra: 92.63% (741/800). RPent Flash Mode / Molmo2-8B: 72.63% (581/800).

-

ASPIRE：六项常规评测；Long 为 LIBERO-90 技能库零样本迁移。

完整方法与分项成绩

方法 / 模型总体Spatial TaskSpatial SwapObject TaskObject SwapGoal TaskGoal SwapLong TaskLong Swap
GPT-6 Astra92.63%100%98%100%99%88%99%85%72%
Opus-4.7 / max.reasoning82.4%————————
RPent Flash Mode / Molmo2-8B72.63%79.00%70.00%86.00%93.00%74.00%65.00%60.00%54.00%
GPT-5.572.1%81.0%69.0%94.0%91.0%75.0%66.0%52.0%49.0%
ASPIRE（八项算术平均）61.36%60.0%51.0%95.0%98.0%45.0%81.0%38.3%22.6%
π_RLinf50.0%42.0%59.0%71.0%78.0%45.0%42.0%49.0%14.0%
π0.511.0%1.0%20.0%1.0%17.0%2.0%38.0%1.0%8.0%
AtomVLA6.3%1.0%16.0%0.0%10.0%11.0%2.0%9.0%1.0%
X-VLA3.8%0.0%0.0%8.0%2.0%9.0%1.0%10.0%0.0%
MolmoAct1.5%0.0%0.0%0.0%6.0%0.0%0.0%6.0%0.0%
π00.3%0.0%0.0%0.0%2.0%0.0%0.0%0.0%0.0%
Cap-X—14.0%12.0%18.0%22.0%17.0%26.0%——
RATS—31.0%29.0%63.0%61.0%36.0%43.0%——
+

完整方法与分项成绩

方法 / 模型总体Spatial TaskSpatial SwapObject TaskObject SwapGoal TaskGoal SwapLong TaskLong Swap
GPT-6 Astra92.63%100%98%100%99%88%99%85%72%
Opus-4.7 / max.reasoning82.4%————————
RPent Flash Mode / Molmo2-8B72.63%79.00%70.00%86.00%93.00%74.00%65.00%60.00%54.00%
GPT-5.572.1%81.0%69.0%94.0%91.0%75.0%66.0%52.0%49.0%
ASPIRE61.36%60.0%51.0%95.0%98.0%45.0%81.0%38.3%22.6%
π_RLinf50.0%42.0%59.0%71.0%78.0%45.0%42.0%49.0%14.0%
π0.511.0%1.0%20.0%1.0%17.0%2.0%38.0%1.0%8.0%
AtomVLA6.3%1.0%16.0%0.0%10.0%11.0%2.0%9.0%1.0%
X-VLA3.8%0.0%0.0%8.0%2.0%9.0%1.0%10.0%0.0%
MolmoAct1.5%0.0%0.0%0.0%6.0%0.0%0.0%6.0%0.0%
π00.3%0.0%0.0%0.0%2.0%0.0%0.0%0.0%0.0%
Cap-X—14.0%12.0%18.0%22.0%17.0%26.0%——
RATS—31.0%29.0%63.0%61.0%36.0%43.0%——

ASPIRE：Long Task 和 Long Swap 使用 LIBERO-90 技能库进行 zero-shot 迁移。

diff --git a/docs/source-zh/rst_source/overview.rst b/docs/source-zh/rst_source/overview.rst index cca6de3a3..e63bf633c 100644 --- a/docs/source-zh/rst_source/overview.rst +++ b/docs/source-zh/rst_source/overview.rst @@ -25,13 +25,13 @@ RPent 建立在三条核心设计原则之上: **服务化、标准化、可组 对比 LIBERO、LIBERO-PRO、RoboCasa365 Target50 和 RoboTwin C2R 上的成功率。 排名仅限图中方法及评测范围；完整结果、模型配置和来源见 :doc:`benchmarks`。 -.. image:: https://raw.githubusercontent.com/RLinf/misc/36f26c2d0cdb2e00b78b6ea3166e4c265c45b51f/rpent/benchmarks/leaderboard-zh-light.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/fbb158d608554595dcd2d4aae6fcc42e8f71c86b/rpent/benchmarks/leaderboard-zh-light.png :alt: RPent benchmark results :class: only-light :width: 100% :target: benchmarks.html -.. image:: https://raw.githubusercontent.com/RLinf/misc/36f26c2d0cdb2e00b78b6ea3166e4c265c45b51f/rpent/benchmarks/leaderboard-zh-dark.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/fbb158d608554595dcd2d4aae6fcc42e8f71c86b/rpent/benchmarks/leaderboard-zh-dark.png :alt: RPent benchmark results :class: only-dark :width: 100% From 7b20311b49996d5212c0ff2245b2c4fa3b9ff71c Mon Sep 17 00:00:00 2001 From: ye shengjie <184427520+ye1457@users.noreply.github.com> Date: Thu, 17 Sep 2026 04:15:32 +0000 Subject: [PATCH 09/13] docs(benchmarks): sync RoboCasa Astra splits and verified episode counts --- README.md | 2 +- README.zh-CN.md | 2 +- docs/source-en/rst_source/benchmarks.rst | 11 ++++++----- docs/source-en/rst_source/overview.rst | 4 ++-- docs/source-zh/rst_source/benchmarks.rst | 11 ++++++----- docs/source-zh/rst_source/overview.rst | 4 ++-- 6 files changed, 18 insertions(+), 16 deletions(-) diff --git a/README.md b/README.md index c84ae8249..8cc5b295a 100644 --- a/README.md +++ b/README.md @@ -33,7 +33,7 @@ Compare RPent with reference methods on LIBERO, LIBERO-PRO, RoboCasa365 Target50 Codex / GPT-6 Astra / low / reasoning: **92.63% Overall (741/800)** across all eight LIBERO-PRO suites. See the [suite results](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html), including the separately frozen Long and Spatial/Object/Goal memory batches. -[![RPent success-rate comparisons across four benchmarks](https://raw.githubusercontent.com/RLinf/misc/fbb158d608554595dcd2d4aae6fcc42e8f71c86b/rpent/benchmarks/leaderboard-en-light.png)](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) +[![RPent success-rate comparisons across four benchmarks](https://raw.githubusercontent.com/RLinf/misc/15da23a9cbc402c92cd6ea05a4f71adedcaf6a99/rpent/benchmarks/leaderboard-en-light.png)](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) ## Who Should Consider Using RPent? diff --git a/README.zh-CN.md b/README.zh-CN.md index 921de9920..597208857 100644 --- a/README.zh-CN.md +++ b/README.zh-CN.md @@ -33,7 +33,7 @@ Codex / GPT-6 Astra / low / reasoning 已完成全部八套 LIBERO-PRO，**Overall 92.63%（741/800）**。详见 [套件汇总](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html)，其中 Long 与 Spatial/Object/Goal 分别使用各自冻结的 memory 批次。 -[![RPent 四项基准成功率对比](https://raw.githubusercontent.com/RLinf/misc/fbb158d608554595dcd2d4aae6fcc42e8f71c86b/rpent/benchmarks/leaderboard-zh-light.png)](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html) +[![RPent 四项基准成功率对比](https://raw.githubusercontent.com/RLinf/misc/15da23a9cbc402c92cd6ea05a4f71adedcaf6a99/rpent/benchmarks/leaderboard-zh-light.png)](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html) ## 适用用户 diff --git a/docs/source-en/rst_source/benchmarks.rst b/docs/source-en/rst_source/benchmarks.rst index 1a558c010..d72cd74b0 100644 --- a/docs/source-en/rst_source/benchmarks.rst +++ b/docs/source-en/rst_source/benchmarks.rst @@ -9,12 +9,12 @@ RPent Leaderboard .. raw:: html - - - - + + + +
+ data-results-url="https://raw.githubusercontent.com/RLinf/misc/15da23a9cbc402c92cd6ea05a4f71adedcaf6a99/rpent/benchmarks/results.json">

LIBERO-PRO

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus-4.7 / max.reasoning82.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
ASPIRE61.36%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%

LIBERO

Method / modelSuccess rate
AtomVLA97.0%
Claude Code / Opus-4.7 / max.reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%
@@ -22,5 +22,6 @@ RPent Leaderboard

RoboTwin

Method / modelSuccess rate
Codex / GPT-5.5 / xhigh / reasoning62.4%
Claude Code / Opus-4.7 / max.reasoning58.4%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%

GPT-6 Astra: 92.63% (741/800). RPent Flash Mode / Molmo2-8B: 72.63% (581/800).

All methods & reported scores

Method / modelOverallSpatial TaskSpatial SwapObject TaskObject SwapGoal TaskGoal SwapLong TaskLong Swap
GPT-6 Astra92.63%100%98%100%99%88%99%85%72%
Opus-4.7 / max.reasoning82.4%————————
RPent Flash Mode / Molmo2-8B72.63%79.00%70.00%86.00%93.00%74.00%65.00%60.00%54.00%
GPT-5.572.1%81.0%69.0%94.0%91.0%75.0%66.0%52.0%49.0%
ASPIRE61.36%60.0%51.0%95.0%98.0%45.0%81.0%38.3%22.6%
π_RLinf50.0%42.0%59.0%71.0%78.0%45.0%42.0%49.0%14.0%
π0.511.0%1.0%20.0%1.0%17.0%2.0%38.0%1.0%8.0%
AtomVLA6.3%1.0%16.0%0.0%10.0%11.0%2.0%9.0%1.0%
X-VLA3.8%0.0%0.0%8.0%2.0%9.0%1.0%10.0%0.0%
MolmoAct1.5%0.0%0.0%0.0%6.0%0.0%0.0%6.0%0.0%
π00.3%0.0%0.0%0.0%2.0%0.0%0.0%0.0%0.0%
Cap-X—14.0%12.0%18.0%22.0%17.0%26.0%——
RATS—31.0%29.0%63.0%61.0%36.0%43.0%——

ASPIRE: Long Task and Long Swap use zero-shot transfer from the LIBERO-90 skill library.

+

RoboCasa365 · Target50 · All methods & reported scores

Method / modelOverallAtomic-SeenComposite-SeenComposite-Unseen
GPT-6 Astra59.20%87.78%43.75%42.50%
Xiaomi-Robotics-157.4%80.2%57.1%32.1%
GPT-5.557.1%92.0%61.0%13.8%
Opus-4.7 / max.reasoning48.6%79.4%47.5%15.0%
WorldDreamer35.3%66.3%26.7%9.0%
RLDX-130.0%60.0%21.3%5.0%
π0.516.9%39.6%7.1%1.2%
π014.8%34.6%6.1%1.1%

GPT-6 Astra: seeds 1–5 for each of the 50 tasks, 250 episodes in total.

diff --git a/docs/source-en/rst_source/overview.rst b/docs/source-en/rst_source/overview.rst index f9dc584e4..643211056 100644 --- a/docs/source-en/rst_source/overview.rst +++ b/docs/source-en/rst_source/overview.rst @@ -33,13 +33,13 @@ Compare success rates on LIBERO, LIBERO-PRO, RoboCasa365 Target50, and RoboTwin C2R. Rankings apply to the methods and evaluation coverage shown; see :doc:`benchmarks` for detailed results, configurations, and sources. -.. image:: https://raw.githubusercontent.com/RLinf/misc/fbb158d608554595dcd2d4aae6fcc42e8f71c86b/rpent/benchmarks/leaderboard-en-light.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/15da23a9cbc402c92cd6ea05a4f71adedcaf6a99/rpent/benchmarks/leaderboard-en-light.png :alt: RPent benchmark results :class: only-light :width: 100% :target: benchmarks.html -.. image:: https://raw.githubusercontent.com/RLinf/misc/fbb158d608554595dcd2d4aae6fcc42e8f71c86b/rpent/benchmarks/leaderboard-en-dark.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/15da23a9cbc402c92cd6ea05a4f71adedcaf6a99/rpent/benchmarks/leaderboard-en-dark.png :alt: RPent benchmark results :class: only-dark :width: 100% diff --git a/docs/source-zh/rst_source/benchmarks.rst b/docs/source-zh/rst_source/benchmarks.rst index d6a18dc1c..79e6f35cc 100644 --- a/docs/source-zh/rst_source/benchmarks.rst +++ b/docs/source-zh/rst_source/benchmarks.rst @@ -9,12 +9,12 @@ RPent 排行榜 .. raw:: html - - - - + + + +
+ data-results-url="https://raw.githubusercontent.com/RLinf/misc/15da23a9cbc402c92cd6ea05a4f71adedcaf6a99/rpent/benchmarks/results.json">

LIBERO-PRO

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus-4.7 / max.reasoning82.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
ASPIRE61.36%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%

LIBERO

方法 / 模型成功率
AtomVLA97.0%
Claude Code / Opus-4.7 / max.reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%
@@ -22,5 +22,6 @@ RPent 排行榜

RoboTwin

方法 / 模型成功率
Codex / GPT-5.5 / xhigh / reasoning62.4%
Claude Code / Opus-4.7 / max.reasoning58.4%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%

GPT-6 Astra: 92.63% (741/800). RPent Flash Mode / Molmo2-8B: 72.63% (581/800).

完整方法与分项成绩

方法 / 模型总体Spatial TaskSpatial SwapObject TaskObject SwapGoal TaskGoal SwapLong TaskLong Swap
GPT-6 Astra92.63%100%98%100%99%88%99%85%72%
Opus-4.7 / max.reasoning82.4%————————
RPent Flash Mode / Molmo2-8B72.63%79.00%70.00%86.00%93.00%74.00%65.00%60.00%54.00%
GPT-5.572.1%81.0%69.0%94.0%91.0%75.0%66.0%52.0%49.0%
ASPIRE61.36%60.0%51.0%95.0%98.0%45.0%81.0%38.3%22.6%
π_RLinf50.0%42.0%59.0%71.0%78.0%45.0%42.0%49.0%14.0%
π0.511.0%1.0%20.0%1.0%17.0%2.0%38.0%1.0%8.0%
AtomVLA6.3%1.0%16.0%0.0%10.0%11.0%2.0%9.0%1.0%
X-VLA3.8%0.0%0.0%8.0%2.0%9.0%1.0%10.0%0.0%
MolmoAct1.5%0.0%0.0%0.0%6.0%0.0%0.0%6.0%0.0%
π00.3%0.0%0.0%0.0%2.0%0.0%0.0%0.0%0.0%
Cap-X—14.0%12.0%18.0%22.0%17.0%26.0%——
RATS—31.0%29.0%63.0%61.0%36.0%43.0%——

ASPIRE：Long Task 和 Long Swap 使用 LIBERO-90 技能库进行 zero-shot 迁移。

+

RoboCasa365 · Target50 · 完整方法与分项成绩

方法 / 模型总体Atomic-SeenComposite-SeenComposite-Unseen
GPT-6 Astra59.20%87.78%43.75%42.50%
Xiaomi-Robotics-157.4%80.2%57.1%32.1%
GPT-5.557.1%92.0%61.0%13.8%
Opus-4.7 / max.reasoning48.6%79.4%47.5%15.0%
WorldDreamer35.3%66.3%26.7%9.0%
RLDX-130.0%60.0%21.3%5.0%
π0.516.9%39.6%7.1%1.2%
π014.8%34.6%6.1%1.1%

GPT-6 Astra：50 个任务各使用 seed 1–5，共 250 回合。

diff --git a/docs/source-zh/rst_source/overview.rst b/docs/source-zh/rst_source/overview.rst index e63bf633c..0cde9614b 100644 --- a/docs/source-zh/rst_source/overview.rst +++ b/docs/source-zh/rst_source/overview.rst @@ -25,13 +25,13 @@ RPent 建立在三条核心设计原则之上: **服务化、标准化、可组 对比 LIBERO、LIBERO-PRO、RoboCasa365 Target50 和 RoboTwin C2R 上的成功率。 排名仅限图中方法及评测范围；完整结果、模型配置和来源见 :doc:`benchmarks`。 -.. image:: https://raw.githubusercontent.com/RLinf/misc/fbb158d608554595dcd2d4aae6fcc42e8f71c86b/rpent/benchmarks/leaderboard-zh-light.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/15da23a9cbc402c92cd6ea05a4f71adedcaf6a99/rpent/benchmarks/leaderboard-zh-light.png :alt: RPent benchmark results :class: only-light :width: 100% :target: benchmarks.html -.. image:: https://raw.githubusercontent.com/RLinf/misc/fbb158d608554595dcd2d4aae6fcc42e8f71c86b/rpent/benchmarks/leaderboard-zh-dark.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/15da23a9cbc402c92cd6ea05a4f71adedcaf6a99/rpent/benchmarks/leaderboard-zh-dark.png :alt: RPent benchmark results :class: only-dark :width: 100% From d3d1c758a58d2368ee1d1b0e828f822d95906afc Mon Sep 17 00:00:00 2001 From: ye shengjie <184427520+ye1457@users.noreply.github.com> Date: Thu, 17 Sep 2026 04:24:43 +0000 Subject: [PATCH 10/13] docs: remove RoboCasa Astra scope note in both languages --- README.md | 2 +- README.zh-CN.md | 2 +- docs/source-en/rst_source/benchmarks.rst | 12 ++++++------ docs/source-en/rst_source/overview.rst | 4 ++-- docs/source-zh/rst_source/benchmarks.rst | 12 ++++++------ docs/source-zh/rst_source/overview.rst | 4 ++-- 6 files changed, 18 insertions(+), 18 deletions(-) diff --git a/README.md b/README.md index 8cc5b295a..eacb2f723 100644 --- a/README.md +++ b/README.md @@ -33,7 +33,7 @@ Compare RPent with reference methods on LIBERO, LIBERO-PRO, RoboCasa365 Target50 Codex / GPT-6 Astra / low / reasoning: **92.63% Overall (741/800)** across all eight LIBERO-PRO suites. See the [suite results](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html), including the separately frozen Long and Spatial/Object/Goal memory batches. -[![RPent success-rate comparisons across four benchmarks](https://raw.githubusercontent.com/RLinf/misc/15da23a9cbc402c92cd6ea05a4f71adedcaf6a99/rpent/benchmarks/leaderboard-en-light.png)](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) +[![RPent success-rate comparisons across four benchmarks](https://raw.githubusercontent.com/RLinf/misc/f3ddb1a933f631cc5c9650cd8d345c56bfc32648/rpent/benchmarks/leaderboard-en-light.png)](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) ## Who Should Consider Using RPent? diff --git a/README.zh-CN.md b/README.zh-CN.md index 597208857..85920cf15 100644 --- a/README.zh-CN.md +++ b/README.zh-CN.md @@ -33,7 +33,7 @@ Codex / GPT-6 Astra / low / reasoning 已完成全部八套 LIBERO-PRO，**Overall 92.63%（741/800）**。详见 [套件汇总](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html)，其中 Long 与 Spatial/Object/Goal 分别使用各自冻结的 memory 批次。 -[![RPent 四项基准成功率对比](https://raw.githubusercontent.com/RLinf/misc/15da23a9cbc402c92cd6ea05a4f71adedcaf6a99/rpent/benchmarks/leaderboard-zh-light.png)](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html) +[![RPent 四项基准成功率对比](https://raw.githubusercontent.com/RLinf/misc/f3ddb1a933f631cc5c9650cd8d345c56bfc32648/rpent/benchmarks/leaderboard-zh-light.png)](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html) ## 适用用户 diff --git a/docs/source-en/rst_source/benchmarks.rst b/docs/source-en/rst_source/benchmarks.rst index d72cd74b0..c0dc45887 100644 --- a/docs/source-en/rst_source/benchmarks.rst +++ b/docs/source-en/rst_source/benchmarks.rst @@ -9,12 +9,12 @@ RPent Leaderboard .. raw:: html - - - - + + + +
+ data-results-url="https://raw.githubusercontent.com/RLinf/misc/f3ddb1a933f631cc5c9650cd8d345c56bfc32648/rpent/benchmarks/results.json">

LIBERO-PRO

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus-4.7 / max.reasoning82.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
ASPIRE61.36%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%

LIBERO

Method / modelSuccess rate
AtomVLA97.0%
Claude Code / Opus-4.7 / max.reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%
@@ -22,6 +22,6 @@ RPent Leaderboard

RoboTwin

Method / modelSuccess rate
Codex / GPT-5.5 / xhigh / reasoning62.4%
Claude Code / Opus-4.7 / max.reasoning58.4%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%

GPT-6 Astra: 92.63% (741/800). RPent Flash Mode / Molmo2-8B: 72.63% (581/800).

All methods & reported scores

Method / modelOverallSpatial TaskSpatial SwapObject TaskObject SwapGoal TaskGoal SwapLong TaskLong Swap
GPT-6 Astra92.63%100%98%100%99%88%99%85%72%
Opus-4.7 / max.reasoning82.4%————————
RPent Flash Mode / Molmo2-8B72.63%79.00%70.00%86.00%93.00%74.00%65.00%60.00%54.00%
GPT-5.572.1%81.0%69.0%94.0%91.0%75.0%66.0%52.0%49.0%
ASPIRE61.36%60.0%51.0%95.0%98.0%45.0%81.0%38.3%22.6%
π_RLinf50.0%42.0%59.0%71.0%78.0%45.0%42.0%49.0%14.0%
π0.511.0%1.0%20.0%1.0%17.0%2.0%38.0%1.0%8.0%
AtomVLA6.3%1.0%16.0%0.0%10.0%11.0%2.0%9.0%1.0%
X-VLA3.8%0.0%0.0%8.0%2.0%9.0%1.0%10.0%0.0%
MolmoAct1.5%0.0%0.0%0.0%6.0%0.0%0.0%6.0%0.0%
π00.3%0.0%0.0%0.0%2.0%0.0%0.0%0.0%0.0%
Cap-X—14.0%12.0%18.0%22.0%17.0%26.0%——
RATS—31.0%29.0%63.0%61.0%36.0%43.0%——

ASPIRE: Long Task and Long Swap use zero-shot transfer from the LIBERO-90 skill library.

-

RoboCasa365 · Target50 · All methods & reported scores

Method / modelOverallAtomic-SeenComposite-SeenComposite-Unseen
GPT-6 Astra59.20%87.78%43.75%42.50%
Xiaomi-Robotics-157.4%80.2%57.1%32.1%
GPT-5.557.1%92.0%61.0%13.8%
Opus-4.7 / max.reasoning48.6%79.4%47.5%15.0%
WorldDreamer35.3%66.3%26.7%9.0%
RLDX-130.0%60.0%21.3%5.0%
π0.516.9%39.6%7.1%1.2%
π014.8%34.6%6.1%1.1%

GPT-6 Astra: seeds 1–5 for each of the 50 tasks, 250 episodes in total.

+

RoboCasa365 · Target50 · All methods & reported scores

Method / modelOverallAtomic-SeenComposite-SeenComposite-Unseen
GPT-6 Astra59.20%87.78%43.75%42.50%
Xiaomi-Robotics-157.4%80.2%57.1%32.1%
GPT-5.557.1%92.0%61.0%13.8%
Opus-4.7 / max.reasoning48.6%79.4%47.5%15.0%
WorldDreamer35.3%66.3%26.7%9.0%
RLDX-130.0%60.0%21.3%5.0%
π0.516.9%39.6%7.1%1.2%
π014.8%34.6%6.1%1.1%
diff --git a/docs/source-en/rst_source/overview.rst b/docs/source-en/rst_source/overview.rst index 643211056..e4624d05a 100644 --- a/docs/source-en/rst_source/overview.rst +++ b/docs/source-en/rst_source/overview.rst @@ -33,13 +33,13 @@ Compare success rates on LIBERO, LIBERO-PRO, RoboCasa365 Target50, and RoboTwin C2R. Rankings apply to the methods and evaluation coverage shown; see :doc:`benchmarks` for detailed results, configurations, and sources. -.. image:: https://raw.githubusercontent.com/RLinf/misc/15da23a9cbc402c92cd6ea05a4f71adedcaf6a99/rpent/benchmarks/leaderboard-en-light.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/f3ddb1a933f631cc5c9650cd8d345c56bfc32648/rpent/benchmarks/leaderboard-en-light.png :alt: RPent benchmark results :class: only-light :width: 100% :target: benchmarks.html -.. image:: https://raw.githubusercontent.com/RLinf/misc/15da23a9cbc402c92cd6ea05a4f71adedcaf6a99/rpent/benchmarks/leaderboard-en-dark.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/f3ddb1a933f631cc5c9650cd8d345c56bfc32648/rpent/benchmarks/leaderboard-en-dark.png :alt: RPent benchmark results :class: only-dark :width: 100% diff --git a/docs/source-zh/rst_source/benchmarks.rst b/docs/source-zh/rst_source/benchmarks.rst index 79e6f35cc..9870bf087 100644 --- a/docs/source-zh/rst_source/benchmarks.rst +++ b/docs/source-zh/rst_source/benchmarks.rst @@ -9,12 +9,12 @@ RPent 排行榜 .. raw:: html - - - - + + + +
+ data-results-url="https://raw.githubusercontent.com/RLinf/misc/f3ddb1a933f631cc5c9650cd8d345c56bfc32648/rpent/benchmarks/results.json">

LIBERO-PRO

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus-4.7 / max.reasoning82.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
ASPIRE61.36%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%

LIBERO

方法 / 模型成功率
AtomVLA97.0%
Claude Code / Opus-4.7 / max.reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%
@@ -22,6 +22,6 @@ RPent 排行榜

RoboTwin

方法 / 模型成功率
Codex / GPT-5.5 / xhigh / reasoning62.4%
Claude Code / Opus-4.7 / max.reasoning58.4%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%

GPT-6 Astra: 92.63% (741/800). RPent Flash Mode / Molmo2-8B: 72.63% (581/800).

完整方法与分项成绩

方法 / 模型总体Spatial TaskSpatial SwapObject TaskObject SwapGoal TaskGoal SwapLong TaskLong Swap
GPT-6 Astra92.63%100%98%100%99%88%99%85%72%
Opus-4.7 / max.reasoning82.4%————————
RPent Flash Mode / Molmo2-8B72.63%79.00%70.00%86.00%93.00%74.00%65.00%60.00%54.00%
GPT-5.572.1%81.0%69.0%94.0%91.0%75.0%66.0%52.0%49.0%
ASPIRE61.36%60.0%51.0%95.0%98.0%45.0%81.0%38.3%22.6%
π_RLinf50.0%42.0%59.0%71.0%78.0%45.0%42.0%49.0%14.0%
π0.511.0%1.0%20.0%1.0%17.0%2.0%38.0%1.0%8.0%
AtomVLA6.3%1.0%16.0%0.0%10.0%11.0%2.0%9.0%1.0%
X-VLA3.8%0.0%0.0%8.0%2.0%9.0%1.0%10.0%0.0%
MolmoAct1.5%0.0%0.0%0.0%6.0%0.0%0.0%6.0%0.0%
π00.3%0.0%0.0%0.0%2.0%0.0%0.0%0.0%0.0%
Cap-X—14.0%12.0%18.0%22.0%17.0%26.0%——
RATS—31.0%29.0%63.0%61.0%36.0%43.0%——

ASPIRE：Long Task 和 Long Swap 使用 LIBERO-90 技能库进行 zero-shot 迁移。

-

RoboCasa365 · Target50 · 完整方法与分项成绩

方法 / 模型总体Atomic-SeenComposite-SeenComposite-Unseen
GPT-6 Astra59.20%87.78%43.75%42.50%
Xiaomi-Robotics-157.4%80.2%57.1%32.1%
GPT-5.557.1%92.0%61.0%13.8%
Opus-4.7 / max.reasoning48.6%79.4%47.5%15.0%
WorldDreamer35.3%66.3%26.7%9.0%
RLDX-130.0%60.0%21.3%5.0%
π0.516.9%39.6%7.1%1.2%
π014.8%34.6%6.1%1.1%

GPT-6 Astra：50 个任务各使用 seed 1–5，共 250 回合。

+

RoboCasa365 · Target50 · 完整方法与分项成绩

方法 / 模型总体Atomic-SeenComposite-SeenComposite-Unseen
GPT-6 Astra59.20%87.78%43.75%42.50%
Xiaomi-Robotics-157.4%80.2%57.1%32.1%
GPT-5.557.1%92.0%61.0%13.8%
Opus-4.7 / max.reasoning48.6%79.4%47.5%15.0%
WorldDreamer35.3%66.3%26.7%9.0%
RLDX-130.0%60.0%21.3%5.0%
π0.516.9%39.6%7.1%1.2%
π014.8%34.6%6.1%1.1%
diff --git a/docs/source-zh/rst_source/overview.rst b/docs/source-zh/rst_source/overview.rst index 0cde9614b..7acc65dc0 100644 --- a/docs/source-zh/rst_source/overview.rst +++ b/docs/source-zh/rst_source/overview.rst @@ -25,13 +25,13 @@ RPent 建立在三条核心设计原则之上: **服务化、标准化、可组 对比 LIBERO、LIBERO-PRO、RoboCasa365 Target50 和 RoboTwin C2R 上的成功率。 排名仅限图中方法及评测范围；完整结果、模型配置和来源见 :doc:`benchmarks`。 -.. image:: https://raw.githubusercontent.com/RLinf/misc/15da23a9cbc402c92cd6ea05a4f71adedcaf6a99/rpent/benchmarks/leaderboard-zh-light.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/f3ddb1a933f631cc5c9650cd8d345c56bfc32648/rpent/benchmarks/leaderboard-zh-light.png :alt: RPent benchmark results :class: only-light :width: 100% :target: benchmarks.html -.. image:: https://raw.githubusercontent.com/RLinf/misc/15da23a9cbc402c92cd6ea05a4f71adedcaf6a99/rpent/benchmarks/leaderboard-zh-dark.png +.. image:: https://raw.githubusercontent.com/RLinf/misc/f3ddb1a933f631cc5c9650cd8d345c56bfc32648/rpent/benchmarks/leaderboard-zh-dark.png :alt: RPent benchmark results :class: only-dark :width: 100% From ed2f896e9164e58a1b48b91828e9874df800b903 Mon Sep 17 00:00:00 2001 From: ye shengjie <184427520+ye1457@users.noreply.github.com> Date: Thu, 17 Sep 2026 04:52:29 +0000 Subject: [PATCH 11/13] docs: address leaderboard theme, grouping and memory review --- README.md | 4 ++-- README.zh-CN.md | 4 ++-- docs/source-en/rst_source/benchmarks.rst | 21 +++++++++------------ docs/source-en/rst_source/overview.rst | 4 ++-- docs/source-zh/rst_source/benchmarks.rst | 21 +++++++++------------ docs/source-zh/rst_source/overview.rst | 4 ++-- 6 files changed, 26 insertions(+), 32 deletions(-) diff --git a/README.md b/README.md index eacb2f723..01eed9f68 100644 --- a/README.md +++ b/README.md @@ -31,9 +31,9 @@ Compare RPent with reference methods on LIBERO, LIBERO-PRO, RoboCasa365 Target50, and RoboTwin C2R. Rankings apply to the methods and evaluation coverage shown; see [Benchmark Results](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) for suite results and model configurations. -Codex / GPT-6 Astra / low / reasoning: **92.63% Overall (741/800)** across all eight LIBERO-PRO suites. See the [suite results](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html), including the separately frozen Long and Spatial/Object/Goal memory batches. +Codex / GPT-6 Astra / low / reasoning: **92.63% Overall (741/800)** across all eight LIBERO-PRO suites. See the [suite results and memory-batch explanation](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html#libero-pro-astra-memory), including the separately frozen Long and Spatial/Object/Goal memory batches. -[![RPent success-rate comparisons across four benchmarks](https://raw.githubusercontent.com/RLinf/misc/f3ddb1a933f631cc5c9650cd8d345c56bfc32648/rpent/benchmarks/leaderboard-en-light.png)](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) +[![RPent success-rate comparisons across four benchmarks](https://cdn.jsdelivr.net/gh/RLinf/misc@44420395c8a31ff5065f1ebede08ed19255b3711/rpent/benchmarks/leaderboard-en-light.png)](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) ## Who Should Consider Using RPent? diff --git a/README.zh-CN.md b/README.zh-CN.md index 85920cf15..6ae641fc1 100644 --- a/README.zh-CN.md +++ b/README.zh-CN.md @@ -31,9 +31,9 @@ 对比 RPent 与参考方法在 LIBERO、LIBERO-PRO、RoboCasa365 Target50 和 RoboTwin C2R 上的成功率。排名仅限图中方法及评测范围；套件成绩和模型配置见[基准测试结果](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html)。 -Codex / GPT-6 Astra / low / reasoning 已完成全部八套 LIBERO-PRO，**Overall 92.63%（741/800）**。详见 [套件汇总](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html)，其中 Long 与 Spatial/Object/Goal 分别使用各自冻结的 memory 批次。 +Codex / GPT-6 Astra / low / reasoning 已完成全部八套 LIBERO-PRO，**Overall 92.63%（741/800）**。详见 [套件汇总与 memory 批次说明](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html#libero-pro-astra-memory)，其中 Long 与 Spatial/Object/Goal 分别使用各自冻结的 memory 批次。 -[![RPent 四项基准成功率对比](https://raw.githubusercontent.com/RLinf/misc/f3ddb1a933f631cc5c9650cd8d345c56bfc32648/rpent/benchmarks/leaderboard-zh-light.png)](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html) +[![RPent 四项基准成功率对比](https://cdn.jsdelivr.net/gh/RLinf/misc@44420395c8a31ff5065f1ebede08ed19255b3711/rpent/benchmarks/leaderboard-zh-light.png)](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html) ## 适用用户 diff --git a/docs/source-en/rst_source/benchmarks.rst b/docs/source-en/rst_source/benchmarks.rst index c0dc45887..15d783c41 100644 --- a/docs/source-en/rst_source/benchmarks.rst +++ b/docs/source-en/rst_source/benchmarks.rst @@ -9,19 +9,16 @@ RPent Leaderboard .. raw:: html - - - - + + + +
+ data-results-url="https://cdn.jsdelivr.net/gh/RLinf/misc@44420395c8a31ff5065f1ebede08ed19255b3711/rpent/benchmarks/results.json">
-

LIBERO-PRO

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus-4.7 / max.reasoning82.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
ASPIRE61.36%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%
-

LIBERO

Method / modelSuccess rate
AtomVLA97.0%
Claude Code / Opus-4.7 / max.reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%
-

RoboCasa365 · Target50

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning59.20%
Xiaomi-Robotics-157.4%
Codex / GPT-5.5 / xhigh / reasoning57.1%
Claude Code / Opus-4.7 / max.reasoning48.6%
WorldDreamer35.3%
RLDX-130.0%
π0.516.9%
π014.8%
-

RoboTwin

Method / modelSuccess rate
Codex / GPT-5.5 / xhigh / reasoning62.4%
Claude Code / Opus-4.7 / max.reasoning58.4%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%
-

GPT-6 Astra: 92.63% (741/800). RPent Flash Mode / Molmo2-8B: 72.63% (581/800).

-

All methods & reported scores

Method / modelOverallSpatial TaskSpatial SwapObject TaskObject SwapGoal TaskGoal SwapLong TaskLong Swap
GPT-6 Astra92.63%100%98%100%99%88%99%85%72%
Opus-4.7 / max.reasoning82.4%————————
RPent Flash Mode / Molmo2-8B72.63%79.00%70.00%86.00%93.00%74.00%65.00%60.00%54.00%
GPT-5.572.1%81.0%69.0%94.0%91.0%75.0%66.0%52.0%49.0%
ASPIRE61.36%60.0%51.0%95.0%98.0%45.0%81.0%38.3%22.6%
π_RLinf50.0%42.0%59.0%71.0%78.0%45.0%42.0%49.0%14.0%
π0.511.0%1.0%20.0%1.0%17.0%2.0%38.0%1.0%8.0%
AtomVLA6.3%1.0%16.0%0.0%10.0%11.0%2.0%9.0%1.0%
X-VLA3.8%0.0%0.0%8.0%2.0%9.0%1.0%10.0%0.0%
MolmoAct1.5%0.0%0.0%0.0%6.0%0.0%0.0%6.0%0.0%
π00.3%0.0%0.0%0.0%2.0%0.0%0.0%0.0%0.0%
Cap-X—14.0%12.0%18.0%22.0%17.0%26.0%——
RATS—31.0%29.0%63.0%61.0%36.0%43.0%——

ASPIRE: Long Task and Long Swap use zero-shot transfer from the LIBERO-90 skill library.

-

RoboCasa365 · Target50 · All methods & reported scores

Method / modelOverallAtomic-SeenComposite-SeenComposite-Unseen
GPT-6 Astra59.20%87.78%43.75%42.50%
Xiaomi-Robotics-157.4%80.2%57.1%32.1%
GPT-5.557.1%92.0%61.0%13.8%
Opus-4.7 / max.reasoning48.6%79.4%47.5%15.0%
WorldDreamer35.3%66.3%26.7%9.0%
RLDX-130.0%60.0%21.3%5.0%
π0.516.9%39.6%7.1%1.2%
π014.8%34.6%6.1%1.1%
+

LIBERO-PRO

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus-4.7 / max.reasoning82.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
ASPIRE61.36%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%

GPT-6 Astra: 92.63% (741/800). RPent Flash Mode / Molmo2-8B: 72.63% (581/800).

All methods & reported scores

Method / modelOverallSpatial TaskSpatial SwapObject TaskObject SwapGoal TaskGoal SwapLong TaskLong Swap
GPT-6 Astra92.63%100%98%100%99%88%99%85%72%
Opus-4.7 / max.reasoning82.4%————————
RPent Flash Mode / Molmo2-8B72.63%79.00%70.00%86.00%93.00%74.00%65.00%60.00%54.00%
GPT-5.572.1%81.0%69.0%94.0%91.0%75.0%66.0%52.0%49.0%
ASPIRE61.36%60.0%51.0%95.0%98.0%45.0%81.0%38.3%22.6%
π_RLinf50.0%42.0%59.0%71.0%78.0%45.0%42.0%49.0%14.0%
π0.511.0%1.0%20.0%1.0%17.0%2.0%38.0%1.0%8.0%
AtomVLA6.3%1.0%16.0%0.0%10.0%11.0%2.0%9.0%1.0%
X-VLA3.8%0.0%0.0%8.0%2.0%9.0%1.0%10.0%0.0%
MolmoAct1.5%0.0%0.0%0.0%6.0%0.0%0.0%6.0%0.0%
π00.3%0.0%0.0%0.0%2.0%0.0%0.0%0.0%0.0%
Cap-X—14.0%12.0%18.0%22.0%17.0%26.0%——
RATS—31.0%29.0%63.0%61.0%36.0%43.0%——

ASPIRE: Long Task and Long Swap use zero-shot transfer from the LIBERO-90 skill library.

GPT-6 Astra: Long Task/Swap and the other six suites use separate memory-file snapshots frozen after their respective exploration phases, with no updates during evaluation. Overall combines two non-overlapping batches: Long 157/200 plus the other suites 584/600, giving 741/800 (92.63%); the 800 episodes do not share a single memory snapshot.

+

LIBERO

Method / modelSuccess rate
AtomVLA97.0%
Claude Code / Opus-4.7 / max.reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%
+

RoboCasa365 · Target50

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning59.20%
Xiaomi-Robotics-157.4%
Codex / GPT-5.5 / xhigh / reasoning57.1%
Claude Code / Opus-4.7 / max.reasoning48.6%
WorldDreamer35.3%
RLDX-130.0%
π0.516.9%
π014.8%

RoboCasa365 · Target50 · All methods & reported scores

Method / modelOverallAtomic-SeenComposite-SeenComposite-Unseen
GPT-6 Astra59.20%87.78%43.75%42.50%
Xiaomi-Robotics-157.4%80.2%57.1%32.1%
GPT-5.557.1%92.0%61.0%13.8%
Opus-4.7 / max.reasoning48.6%79.4%47.5%15.0%
WorldDreamer35.3%66.3%26.7%9.0%
RLDX-130.0%60.0%21.3%5.0%
π0.516.9%39.6%7.1%1.2%
π014.8%34.6%6.1%1.1%
+

RoboTwin

Method / modelSuccess rate
Codex / GPT-5.5 / xhigh / reasoning62.4%
Claude Code / Opus-4.7 / max.reasoning58.4%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%
diff --git a/docs/source-en/rst_source/overview.rst b/docs/source-en/rst_source/overview.rst index e4624d05a..34f9cefb7 100644 --- a/docs/source-en/rst_source/overview.rst +++ b/docs/source-en/rst_source/overview.rst @@ -33,13 +33,13 @@ Compare success rates on LIBERO, LIBERO-PRO, RoboCasa365 Target50, and RoboTwin C2R. Rankings apply to the methods and evaluation coverage shown; see :doc:`benchmarks` for detailed results, configurations, and sources. -.. image:: https://raw.githubusercontent.com/RLinf/misc/f3ddb1a933f631cc5c9650cd8d345c56bfc32648/rpent/benchmarks/leaderboard-en-light.png +.. image:: https://cdn.jsdelivr.net/gh/RLinf/misc@44420395c8a31ff5065f1ebede08ed19255b3711/rpent/benchmarks/leaderboard-en-light.png :alt: RPent benchmark results :class: only-light :width: 100% :target: benchmarks.html -.. image:: https://raw.githubusercontent.com/RLinf/misc/f3ddb1a933f631cc5c9650cd8d345c56bfc32648/rpent/benchmarks/leaderboard-en-dark.png +.. image:: https://cdn.jsdelivr.net/gh/RLinf/misc@44420395c8a31ff5065f1ebede08ed19255b3711/rpent/benchmarks/leaderboard-en-dark.png :alt: RPent benchmark results :class: only-dark :width: 100% diff --git a/docs/source-zh/rst_source/benchmarks.rst b/docs/source-zh/rst_source/benchmarks.rst index 9870bf087..abb88a8e8 100644 --- a/docs/source-zh/rst_source/benchmarks.rst +++ b/docs/source-zh/rst_source/benchmarks.rst @@ -9,19 +9,16 @@ RPent 排行榜 .. raw:: html - - - - + + + +
+ data-results-url="https://cdn.jsdelivr.net/gh/RLinf/misc@44420395c8a31ff5065f1ebede08ed19255b3711/rpent/benchmarks/results.json">
-

LIBERO-PRO

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus-4.7 / max.reasoning82.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
ASPIRE61.36%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%
-

LIBERO

方法 / 模型成功率
AtomVLA97.0%
Claude Code / Opus-4.7 / max.reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%
-

RoboCasa365 · Target50

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning59.20%
Xiaomi-Robotics-157.4%
Codex / GPT-5.5 / xhigh / reasoning57.1%
Claude Code / Opus-4.7 / max.reasoning48.6%
WorldDreamer35.3%
RLDX-130.0%
π0.516.9%
π014.8%
-

RoboTwin

方法 / 模型成功率
Codex / GPT-5.5 / xhigh / reasoning62.4%
Claude Code / Opus-4.7 / max.reasoning58.4%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%
-

GPT-6 Astra: 92.63% (741/800). RPent Flash Mode / Molmo2-8B: 72.63% (581/800).

-

完整方法与分项成绩

方法 / 模型总体Spatial TaskSpatial SwapObject TaskObject SwapGoal TaskGoal SwapLong TaskLong Swap
GPT-6 Astra92.63%100%98%100%99%88%99%85%72%
Opus-4.7 / max.reasoning82.4%————————
RPent Flash Mode / Molmo2-8B72.63%79.00%70.00%86.00%93.00%74.00%65.00%60.00%54.00%
GPT-5.572.1%81.0%69.0%94.0%91.0%75.0%66.0%52.0%49.0%
ASPIRE61.36%60.0%51.0%95.0%98.0%45.0%81.0%38.3%22.6%
π_RLinf50.0%42.0%59.0%71.0%78.0%45.0%42.0%49.0%14.0%
π0.511.0%1.0%20.0%1.0%17.0%2.0%38.0%1.0%8.0%
AtomVLA6.3%1.0%16.0%0.0%10.0%11.0%2.0%9.0%1.0%
X-VLA3.8%0.0%0.0%8.0%2.0%9.0%1.0%10.0%0.0%
MolmoAct1.5%0.0%0.0%0.0%6.0%0.0%0.0%6.0%0.0%
π00.3%0.0%0.0%0.0%2.0%0.0%0.0%0.0%0.0%
Cap-X—14.0%12.0%18.0%22.0%17.0%26.0%——
RATS—31.0%29.0%63.0%61.0%36.0%43.0%——

ASPIRE：Long Task 和 Long Swap 使用 LIBERO-90 技能库进行 zero-shot 迁移。

-

RoboCasa365 · Target50 · 完整方法与分项成绩

方法 / 模型总体Atomic-SeenComposite-SeenComposite-Unseen
GPT-6 Astra59.20%87.78%43.75%42.50%
Xiaomi-Robotics-157.4%80.2%57.1%32.1%
GPT-5.557.1%92.0%61.0%13.8%
Opus-4.7 / max.reasoning48.6%79.4%47.5%15.0%
WorldDreamer35.3%66.3%26.7%9.0%
RLDX-130.0%60.0%21.3%5.0%
π0.516.9%39.6%7.1%1.2%
π014.8%34.6%6.1%1.1%
+

LIBERO-PRO

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus-4.7 / max.reasoning82.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
ASPIRE61.36%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%

GPT-6 Astra: 92.63% (741/800). RPent Flash Mode / Molmo2-8B: 72.63% (581/800).

完整方法与分项成绩

方法 / 模型总体Spatial TaskSpatial SwapObject TaskObject SwapGoal TaskGoal SwapLong TaskLong Swap
GPT-6 Astra92.63%100%98%100%99%88%99%85%72%
Opus-4.7 / max.reasoning82.4%————————
RPent Flash Mode / Molmo2-8B72.63%79.00%70.00%86.00%93.00%74.00%65.00%60.00%54.00%
GPT-5.572.1%81.0%69.0%94.0%91.0%75.0%66.0%52.0%49.0%
ASPIRE61.36%60.0%51.0%95.0%98.0%45.0%81.0%38.3%22.6%
π_RLinf50.0%42.0%59.0%71.0%78.0%45.0%42.0%49.0%14.0%
π0.511.0%1.0%20.0%1.0%17.0%2.0%38.0%1.0%8.0%
AtomVLA6.3%1.0%16.0%0.0%10.0%11.0%2.0%9.0%1.0%
X-VLA3.8%0.0%0.0%8.0%2.0%9.0%1.0%10.0%0.0%
MolmoAct1.5%0.0%0.0%0.0%6.0%0.0%0.0%6.0%0.0%
π00.3%0.0%0.0%0.0%2.0%0.0%0.0%0.0%0.0%
Cap-X—14.0%12.0%18.0%22.0%17.0%26.0%——
RATS—31.0%29.0%63.0%61.0%36.0%43.0%——

ASPIRE：Long Task 和 Long Swap 使用 LIBERO-90 技能库进行 zero-shot 迁移。

GPT-6 Astra：Long Task/Swap 与其余六套件使用各自探索后冻结的 memory 文件快照，评测期间不更新。Overall 合并两个不重叠批次：Long 157/200，加上其余套件 584/600，得到 741/800（92.63%）；并非全部回合共享同一份 memory 快照。

+

LIBERO

方法 / 模型成功率
AtomVLA97.0%
Claude Code / Opus-4.7 / max.reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%
+

RoboCasa365 · Target50

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning59.20%
Xiaomi-Robotics-157.4%
Codex / GPT-5.5 / xhigh / reasoning57.1%
Claude Code / Opus-4.7 / max.reasoning48.6%
WorldDreamer35.3%
RLDX-130.0%
π0.516.9%
π014.8%

RoboCasa365 · Target50 · 完整方法与分项成绩

方法 / 模型总体Atomic-SeenComposite-SeenComposite-Unseen
GPT-6 Astra59.20%87.78%43.75%42.50%
Xiaomi-Robotics-157.4%80.2%57.1%32.1%
GPT-5.557.1%92.0%61.0%13.8%
Opus-4.7 / max.reasoning48.6%79.4%47.5%15.0%
WorldDreamer35.3%66.3%26.7%9.0%
RLDX-130.0%60.0%21.3%5.0%
π0.516.9%39.6%7.1%1.2%
π014.8%34.6%6.1%1.1%
+

RoboTwin

方法 / 模型成功率
Codex / GPT-5.5 / xhigh / reasoning62.4%
Claude Code / Opus-4.7 / max.reasoning58.4%
LingBot-VLA50.4%
π0.547.9%
GR00T-N1.720.7%
StarVLA10.6%
diff --git a/docs/source-zh/rst_source/overview.rst b/docs/source-zh/rst_source/overview.rst index 7acc65dc0..19403e2a8 100644 --- a/docs/source-zh/rst_source/overview.rst +++ b/docs/source-zh/rst_source/overview.rst @@ -25,13 +25,13 @@ RPent 建立在三条核心设计原则之上: **服务化、标准化、可组 对比 LIBERO、LIBERO-PRO、RoboCasa365 Target50 和 RoboTwin C2R 上的成功率。 排名仅限图中方法及评测范围；完整结果、模型配置和来源见 :doc:`benchmarks`。 -.. image:: https://raw.githubusercontent.com/RLinf/misc/f3ddb1a933f631cc5c9650cd8d345c56bfc32648/rpent/benchmarks/leaderboard-zh-light.png +.. image:: https://cdn.jsdelivr.net/gh/RLinf/misc@44420395c8a31ff5065f1ebede08ed19255b3711/rpent/benchmarks/leaderboard-zh-light.png :alt: RPent benchmark results :class: only-light :width: 100% :target: benchmarks.html -.. image:: https://raw.githubusercontent.com/RLinf/misc/f3ddb1a933f631cc5c9650cd8d345c56bfc32648/rpent/benchmarks/leaderboard-zh-dark.png +.. image:: https://cdn.jsdelivr.net/gh/RLinf/misc@44420395c8a31ff5065f1ebede08ed19255b3711/rpent/benchmarks/leaderboard-zh-dark.png :alt: RPent benchmark results :class: only-dark :width: 100% From dcc229dda0c5ddcdde9b51b0337d14ec39393b00 Mon Sep 17 00:00:00 2001 From: ye shengjie <184427520+ye1457@users.noreply.github.com> Date: Thu, 17 Sep 2026 06:33:26 +0000 Subject: [PATCH 12/13] docs: fix leaderboard top link and align announcement icons --- README.md | 4 ++-- README.zh-CN.md | 4 ++-- docs/source-en/rst_source/benchmarks.rst | 10 +++++----- docs/source-en/rst_source/overview.rst | 4 ++-- docs/source-zh/rst_source/benchmarks.rst | 10 +++++----- docs/source-zh/rst_source/overview.rst | 4 ++-- 6 files changed, 18 insertions(+), 18 deletions(-) diff --git a/README.md b/README.md index 01eed9f68..c83c52874 100644 --- a/README.md +++ b/README.md @@ -33,7 +33,7 @@ Compare RPent with reference methods on LIBERO, LIBERO-PRO, RoboCasa365 Target50 Codex / GPT-6 Astra / low / reasoning: **92.63% Overall (741/800)** across all eight LIBERO-PRO suites. See the [suite results and memory-batch explanation](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html#libero-pro-astra-memory), including the separately frozen Long and Spatial/Object/Goal memory batches. -[![RPent success-rate comparisons across four benchmarks](https://cdn.jsdelivr.net/gh/RLinf/misc@44420395c8a31ff5065f1ebede08ed19255b3711/rpent/benchmarks/leaderboard-en-light.png)](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) +[![RPent success-rate comparisons across four benchmarks](https://cdn.jsdelivr.net/gh/RLinf/misc@881ca8850640e1ac5e728865926349e4660032e7/rpent/benchmarks/leaderboard-en-light.png)](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) ## Who Should Consider Using RPent? @@ -47,7 +47,7 @@ RPent is built for four kinds of users: ## What's NEW! -- [2026/09] Added an interactive leaderboard and consolidated benchmark results for LIBERO, LIBERO-PRO, RoboCasa365, and RoboTwin, with model comparisons and suite-level results. Explore [Benchmark Results](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html). +- [2026/09] 🔥 Added an interactive leaderboard and consolidated benchmark results for LIBERO, LIBERO-PRO, RoboCasa365, and RoboTwin, with model comparisons and suite-level results. Explore [Benchmark Results](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html). - [2026/09] 🔥 RPent supports Franka single-arm and dual-arm real-robot extensions. Doc: [Franka](https://rpent.readthedocs.io/en/latest/rst_source/usage/franka.html) · [Dual Franka](https://rpent.readthedocs.io/en/latest/rst_source/usage/dual_franka.html). - [2026/08] 🔥 RPent supports RoboCasa with RLDX-1 as manipulation model. See the [RoboCasa setup and Target50 guide](https://rpent.readthedocs.io/en/latest/rst_source/usage/robocasa.html). - [2026/08] 🔥 RPent supports the non-reasoning mode, which reduces average execution time by ~40%. diff --git a/README.zh-CN.md b/README.zh-CN.md index 6ae641fc1..ea73d34a0 100644 --- a/README.zh-CN.md +++ b/README.zh-CN.md @@ -33,7 +33,7 @@ Codex / GPT-6 Astra / low / reasoning 已完成全部八套 LIBERO-PRO，**Overall 92.63%（741/800）**。详见 [套件汇总与 memory 批次说明](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html#libero-pro-astra-memory)，其中 Long 与 Spatial/Object/Goal 分别使用各自冻结的 memory 批次。 -[![RPent 四项基准成功率对比](https://cdn.jsdelivr.net/gh/RLinf/misc@44420395c8a31ff5065f1ebede08ed19255b3711/rpent/benchmarks/leaderboard-zh-light.png)](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html) +[![RPent 四项基准成功率对比](https://cdn.jsdelivr.net/gh/RLinf/misc@881ca8850640e1ac5e728865926349e4660032e7/rpent/benchmarks/leaderboard-zh-light.png)](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html) ## 适用用户 @@ -47,7 +47,7 @@ RPent 面向以下四类用户： ## 最新动态 -- [2026/09] 新增交互式排行榜和基准测试结果汇总，覆盖 LIBERO、LIBERO-PRO、RoboCasa365 与 RoboTwin，提供模型对比及套件汇总。查看[基准测试结果](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html)。 +- [2026/09] 🔥 新增交互式排行榜和基准测试结果汇总，覆盖 LIBERO、LIBERO-PRO、RoboCasa365 与 RoboTwin，提供模型对比及套件汇总。查看[基准测试结果](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html)。 - [2026/08] 🔥 支持 RoboCasa，使用 RLDX-1 作为操作模型。参见 [RoboCasa 安装与 Target50 指南](https://rpent.readthedocs.io/zh-cn/latest/rst_source/usage/robocasa.html)。 - [2026/08] 🔥 新增非推理（non-reasoning）模式，平均执行时间降低约 40%。 - [2026/08] 🔥 支持 LIBERO 探索模式。文档：[LIBERO 探索模式](https://rpent.readthedocs.io/zh-cn/latest/rst_source/usage/libero.html#memory)。 diff --git a/docs/source-en/rst_source/benchmarks.rst b/docs/source-en/rst_source/benchmarks.rst index 15d783c41..bf183dcfe 100644 --- a/docs/source-en/rst_source/benchmarks.rst +++ b/docs/source-en/rst_source/benchmarks.rst @@ -9,12 +9,12 @@ RPent Leaderboard .. raw:: html - - - - + + + +
+ data-results-url="https://cdn.jsdelivr.net/gh/RLinf/misc@881ca8850640e1ac5e728865926349e4660032e7/rpent/benchmarks/results.json">

LIBERO-PRO

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus-4.7 / max.reasoning82.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
ASPIRE61.36%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%

GPT-6 Astra: 92.63% (741/800). RPent Flash Mode / Molmo2-8B: 72.63% (581/800).

All methods & reported scores

Method / modelOverallSpatial TaskSpatial SwapObject TaskObject SwapGoal TaskGoal SwapLong TaskLong Swap
GPT-6 Astra92.63%100%98%100%99%88%99%85%72%
Opus-4.7 / max.reasoning82.4%————————
RPent Flash Mode / Molmo2-8B72.63%79.00%70.00%86.00%93.00%74.00%65.00%60.00%54.00%
GPT-5.572.1%81.0%69.0%94.0%91.0%75.0%66.0%52.0%49.0%
ASPIRE61.36%60.0%51.0%95.0%98.0%45.0%81.0%38.3%22.6%
π_RLinf50.0%42.0%59.0%71.0%78.0%45.0%42.0%49.0%14.0%
π0.511.0%1.0%20.0%1.0%17.0%2.0%38.0%1.0%8.0%
AtomVLA6.3%1.0%16.0%0.0%10.0%11.0%2.0%9.0%1.0%
X-VLA3.8%0.0%0.0%8.0%2.0%9.0%1.0%10.0%0.0%
MolmoAct1.5%0.0%0.0%0.0%6.0%0.0%0.0%6.0%0.0%
π00.3%0.0%0.0%0.0%2.0%0.0%0.0%0.0%0.0%
Cap-X—14.0%12.0%18.0%22.0%17.0%26.0%——
RATS—31.0%29.0%63.0%61.0%36.0%43.0%——

ASPIRE: Long Task and Long Swap use zero-shot transfer from the LIBERO-90 skill library.

GPT-6 Astra: Long Task/Swap and the other six suites use separate memory-file snapshots frozen after their respective exploration phases, with no updates during evaluation. Overall combines two non-overlapping batches: Long 157/200 plus the other suites 584/600, giving 741/800 (92.63%); the 800 episodes do not share a single memory snapshot.

LIBERO

Method / modelSuccess rate
AtomVLA97.0%
Claude Code / Opus-4.7 / max.reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%
diff --git a/docs/source-en/rst_source/overview.rst b/docs/source-en/rst_source/overview.rst index 34f9cefb7..2c98d0a02 100644 --- a/docs/source-en/rst_source/overview.rst +++ b/docs/source-en/rst_source/overview.rst @@ -33,13 +33,13 @@ Compare success rates on LIBERO, LIBERO-PRO, RoboCasa365 Target50, and RoboTwin C2R. Rankings apply to the methods and evaluation coverage shown; see :doc:`benchmarks` for detailed results, configurations, and sources. -.. image:: https://cdn.jsdelivr.net/gh/RLinf/misc@44420395c8a31ff5065f1ebede08ed19255b3711/rpent/benchmarks/leaderboard-en-light.png +.. image:: https://cdn.jsdelivr.net/gh/RLinf/misc@881ca8850640e1ac5e728865926349e4660032e7/rpent/benchmarks/leaderboard-en-light.png :alt: RPent benchmark results :class: only-light :width: 100% :target: benchmarks.html -.. image:: https://cdn.jsdelivr.net/gh/RLinf/misc@44420395c8a31ff5065f1ebede08ed19255b3711/rpent/benchmarks/leaderboard-en-dark.png +.. image:: https://cdn.jsdelivr.net/gh/RLinf/misc@881ca8850640e1ac5e728865926349e4660032e7/rpent/benchmarks/leaderboard-en-dark.png :alt: RPent benchmark results :class: only-dark :width: 100% diff --git a/docs/source-zh/rst_source/benchmarks.rst b/docs/source-zh/rst_source/benchmarks.rst index abb88a8e8..f00bf090f 100644 --- a/docs/source-zh/rst_source/benchmarks.rst +++ b/docs/source-zh/rst_source/benchmarks.rst @@ -9,12 +9,12 @@ RPent 排行榜 .. raw:: html - - - - + + + +
+ data-results-url="https://cdn.jsdelivr.net/gh/RLinf/misc@881ca8850640e1ac5e728865926349e4660032e7/rpent/benchmarks/results.json">

LIBERO-PRO

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus-4.7 / max.reasoning82.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
ASPIRE61.36%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%

GPT-6 Astra: 92.63% (741/800). RPent Flash Mode / Molmo2-8B: 72.63% (581/800).

完整方法与分项成绩

方法 / 模型总体Spatial TaskSpatial SwapObject TaskObject SwapGoal TaskGoal SwapLong TaskLong Swap
GPT-6 Astra92.63%100%98%100%99%88%99%85%72%
Opus-4.7 / max.reasoning82.4%————————
RPent Flash Mode / Molmo2-8B72.63%79.00%70.00%86.00%93.00%74.00%65.00%60.00%54.00%
GPT-5.572.1%81.0%69.0%94.0%91.0%75.0%66.0%52.0%49.0%
ASPIRE61.36%60.0%51.0%95.0%98.0%45.0%81.0%38.3%22.6%
π_RLinf50.0%42.0%59.0%71.0%78.0%45.0%42.0%49.0%14.0%
π0.511.0%1.0%20.0%1.0%17.0%2.0%38.0%1.0%8.0%
AtomVLA6.3%1.0%16.0%0.0%10.0%11.0%2.0%9.0%1.0%
X-VLA3.8%0.0%0.0%8.0%2.0%9.0%1.0%10.0%0.0%
MolmoAct1.5%0.0%0.0%0.0%6.0%0.0%0.0%6.0%0.0%
π00.3%0.0%0.0%0.0%2.0%0.0%0.0%0.0%0.0%
Cap-X—14.0%12.0%18.0%22.0%17.0%26.0%——
RATS—31.0%29.0%63.0%61.0%36.0%43.0%——

ASPIRE：Long Task 和 Long Swap 使用 LIBERO-90 技能库进行 zero-shot 迁移。

GPT-6 Astra：Long Task/Swap 与其余六套件使用各自探索后冻结的 memory 文件快照，评测期间不更新。Overall 合并两个不重叠批次：Long 157/200，加上其余套件 584/600，得到 741/800（92.63%）；并非全部回合共享同一份 memory 快照。

LIBERO

方法 / 模型成功率
AtomVLA97.0%
Claude Code / Opus-4.7 / max.reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%
diff --git a/docs/source-zh/rst_source/overview.rst b/docs/source-zh/rst_source/overview.rst index 19403e2a8..36f7188e0 100644 --- a/docs/source-zh/rst_source/overview.rst +++ b/docs/source-zh/rst_source/overview.rst @@ -25,13 +25,13 @@ RPent 建立在三条核心设计原则之上: **服务化、标准化、可组 对比 LIBERO、LIBERO-PRO、RoboCasa365 Target50 和 RoboTwin C2R 上的成功率。 排名仅限图中方法及评测范围；完整结果、模型配置和来源见 :doc:`benchmarks`。 -.. image:: https://cdn.jsdelivr.net/gh/RLinf/misc@44420395c8a31ff5065f1ebede08ed19255b3711/rpent/benchmarks/leaderboard-zh-light.png +.. image:: https://cdn.jsdelivr.net/gh/RLinf/misc@881ca8850640e1ac5e728865926349e4660032e7/rpent/benchmarks/leaderboard-zh-light.png :alt: RPent benchmark results :class: only-light :width: 100% :target: benchmarks.html -.. image:: https://cdn.jsdelivr.net/gh/RLinf/misc@44420395c8a31ff5065f1ebede08ed19255b3711/rpent/benchmarks/leaderboard-zh-dark.png +.. image:: https://cdn.jsdelivr.net/gh/RLinf/misc@881ca8850640e1ac5e728865926349e4660032e7/rpent/benchmarks/leaderboard-zh-dark.png :alt: RPent benchmark results :class: only-dark :width: 100% From c8b2d89feb32ea9d09e214124ae6224629695965 Mon Sep 17 00:00:00 2001 From: ye shengjie <184427520+ye1457@users.noreply.github.com> Date: Thu, 17 Sep 2026 06:37:42 +0000 Subject: [PATCH 13/13] docs: pin leaderboard scroll clearance for fixed navigation --- README.md | 2 +- README.zh-CN.md | 2 +- docs/source-en/rst_source/benchmarks.rst | 10 +++++----- docs/source-en/rst_source/overview.rst | 4 ++-- docs/source-zh/rst_source/benchmarks.rst | 10 +++++----- docs/source-zh/rst_source/overview.rst | 4 ++-- 6 files changed, 16 insertions(+), 16 deletions(-) diff --git a/README.md b/README.md index c83c52874..80a9ea9bf 100644 --- a/README.md +++ b/README.md @@ -33,7 +33,7 @@ Compare RPent with reference methods on LIBERO, LIBERO-PRO, RoboCasa365 Target50 Codex / GPT-6 Astra / low / reasoning: **92.63% Overall (741/800)** across all eight LIBERO-PRO suites. See the [suite results and memory-batch explanation](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html#libero-pro-astra-memory), including the separately frozen Long and Spatial/Object/Goal memory batches. -[![RPent success-rate comparisons across four benchmarks](https://cdn.jsdelivr.net/gh/RLinf/misc@881ca8850640e1ac5e728865926349e4660032e7/rpent/benchmarks/leaderboard-en-light.png)](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) +[![RPent success-rate comparisons across four benchmarks](https://cdn.jsdelivr.net/gh/RLinf/misc@c3b9b5d4ffa360a8324c5b7aa510e1ed0876aa43/rpent/benchmarks/leaderboard-en-light.png)](https://rpent.readthedocs.io/en/latest/rst_source/benchmarks.html) ## Who Should Consider Using RPent? diff --git a/README.zh-CN.md b/README.zh-CN.md index ea73d34a0..800b6f524 100644 --- a/README.zh-CN.md +++ b/README.zh-CN.md @@ -33,7 +33,7 @@ Codex / GPT-6 Astra / low / reasoning 已完成全部八套 LIBERO-PRO，**Overall 92.63%（741/800）**。详见 [套件汇总与 memory 批次说明](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html#libero-pro-astra-memory)，其中 Long 与 Spatial/Object/Goal 分别使用各自冻结的 memory 批次。 -[![RPent 四项基准成功率对比](https://cdn.jsdelivr.net/gh/RLinf/misc@881ca8850640e1ac5e728865926349e4660032e7/rpent/benchmarks/leaderboard-zh-light.png)](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html) +[![RPent 四项基准成功率对比](https://cdn.jsdelivr.net/gh/RLinf/misc@c3b9b5d4ffa360a8324c5b7aa510e1ed0876aa43/rpent/benchmarks/leaderboard-zh-light.png)](https://rpent.readthedocs.io/zh-cn/latest/rst_source/benchmarks.html) ## 适用用户 diff --git a/docs/source-en/rst_source/benchmarks.rst b/docs/source-en/rst_source/benchmarks.rst index bf183dcfe..c9a203b7f 100644 --- a/docs/source-en/rst_source/benchmarks.rst +++ b/docs/source-en/rst_source/benchmarks.rst @@ -9,12 +9,12 @@ RPent Leaderboard .. raw:: html - - - - + + + +
+ data-results-url="https://cdn.jsdelivr.net/gh/RLinf/misc@c3b9b5d4ffa360a8324c5b7aa510e1ed0876aa43/rpent/benchmarks/results.json">

LIBERO-PRO

Method / modelSuccess rate
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus-4.7 / max.reasoning82.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
ASPIRE61.36%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%

GPT-6 Astra: 92.63% (741/800). RPent Flash Mode / Molmo2-8B: 72.63% (581/800).

All methods & reported scores

Method / modelOverallSpatial TaskSpatial SwapObject TaskObject SwapGoal TaskGoal SwapLong TaskLong Swap
GPT-6 Astra92.63%100%98%100%99%88%99%85%72%
Opus-4.7 / max.reasoning82.4%————————
RPent Flash Mode / Molmo2-8B72.63%79.00%70.00%86.00%93.00%74.00%65.00%60.00%54.00%
GPT-5.572.1%81.0%69.0%94.0%91.0%75.0%66.0%52.0%49.0%
ASPIRE61.36%60.0%51.0%95.0%98.0%45.0%81.0%38.3%22.6%
π_RLinf50.0%42.0%59.0%71.0%78.0%45.0%42.0%49.0%14.0%
π0.511.0%1.0%20.0%1.0%17.0%2.0%38.0%1.0%8.0%
AtomVLA6.3%1.0%16.0%0.0%10.0%11.0%2.0%9.0%1.0%
X-VLA3.8%0.0%0.0%8.0%2.0%9.0%1.0%10.0%0.0%
MolmoAct1.5%0.0%0.0%0.0%6.0%0.0%0.0%6.0%0.0%
π00.3%0.0%0.0%0.0%2.0%0.0%0.0%0.0%0.0%
Cap-X—14.0%12.0%18.0%22.0%17.0%26.0%——
RATS—31.0%29.0%63.0%61.0%36.0%43.0%——

ASPIRE: Long Task and Long Swap use zero-shot transfer from the LIBERO-90 skill library.

GPT-6 Astra: Long Task/Swap and the other six suites use separate memory-file snapshots frozen after their respective exploration phases, with no updates during evaluation. Overall combines two non-overlapping batches: Long 157/200 plus the other suites 584/600, giving 741/800 (92.63%); the 800 episodes do not share a single memory snapshot.

LIBERO

Method / modelSuccess rate
AtomVLA97.0%
Claude Code / Opus-4.7 / max.reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%
diff --git a/docs/source-en/rst_source/overview.rst b/docs/source-en/rst_source/overview.rst index 2c98d0a02..1a8124fc6 100644 --- a/docs/source-en/rst_source/overview.rst +++ b/docs/source-en/rst_source/overview.rst @@ -33,13 +33,13 @@ Compare success rates on LIBERO, LIBERO-PRO, RoboCasa365 Target50, and RoboTwin C2R. Rankings apply to the methods and evaluation coverage shown; see :doc:`benchmarks` for detailed results, configurations, and sources. -.. image:: https://cdn.jsdelivr.net/gh/RLinf/misc@881ca8850640e1ac5e728865926349e4660032e7/rpent/benchmarks/leaderboard-en-light.png +.. image:: https://cdn.jsdelivr.net/gh/RLinf/misc@c3b9b5d4ffa360a8324c5b7aa510e1ed0876aa43/rpent/benchmarks/leaderboard-en-light.png :alt: RPent benchmark results :class: only-light :width: 100% :target: benchmarks.html -.. image:: https://cdn.jsdelivr.net/gh/RLinf/misc@881ca8850640e1ac5e728865926349e4660032e7/rpent/benchmarks/leaderboard-en-dark.png +.. image:: https://cdn.jsdelivr.net/gh/RLinf/misc@c3b9b5d4ffa360a8324c5b7aa510e1ed0876aa43/rpent/benchmarks/leaderboard-en-dark.png :alt: RPent benchmark results :class: only-dark :width: 100% diff --git a/docs/source-zh/rst_source/benchmarks.rst b/docs/source-zh/rst_source/benchmarks.rst index f00bf090f..336ac3e1c 100644 --- a/docs/source-zh/rst_source/benchmarks.rst +++ b/docs/source-zh/rst_source/benchmarks.rst @@ -9,12 +9,12 @@ RPent 排行榜 .. raw:: html - - - - + + + +
+ data-results-url="https://cdn.jsdelivr.net/gh/RLinf/misc@c3b9b5d4ffa360a8324c5b7aa510e1ed0876aa43/rpent/benchmarks/results.json">

LIBERO-PRO

方法 / 模型成功率
Codex / GPT-6 Astra / low / reasoning92.63%
Claude Code / Opus-4.7 / max.reasoning82.4%
RPent Flash Mode72.63%
Codex / GPT-5.5 / xhigh / reasoning72.1%
ASPIRE61.36%
π_RLinf50.0%
π0.511.0%
AtomVLA6.3%
X-VLA3.8%
MolmoAct1.5%
π00.3%

GPT-6 Astra: 92.63% (741/800). RPent Flash Mode / Molmo2-8B: 72.63% (581/800).

完整方法与分项成绩

方法 / 模型总体Spatial TaskSpatial SwapObject TaskObject SwapGoal TaskGoal SwapLong TaskLong Swap
GPT-6 Astra92.63%100%98%100%99%88%99%85%72%
Opus-4.7 / max.reasoning82.4%————————
RPent Flash Mode / Molmo2-8B72.63%79.00%70.00%86.00%93.00%74.00%65.00%60.00%54.00%
GPT-5.572.1%81.0%69.0%94.0%91.0%75.0%66.0%52.0%49.0%
ASPIRE61.36%60.0%51.0%95.0%98.0%45.0%81.0%38.3%22.6%
π_RLinf50.0%42.0%59.0%71.0%78.0%45.0%42.0%49.0%14.0%
π0.511.0%1.0%20.0%1.0%17.0%2.0%38.0%1.0%8.0%
AtomVLA6.3%1.0%16.0%0.0%10.0%11.0%2.0%9.0%1.0%
X-VLA3.8%0.0%0.0%8.0%2.0%9.0%1.0%10.0%0.0%
MolmoAct1.5%0.0%0.0%0.0%6.0%0.0%0.0%6.0%0.0%
π00.3%0.0%0.0%0.0%2.0%0.0%0.0%0.0%0.0%
Cap-X—14.0%12.0%18.0%22.0%17.0%26.0%——
RATS—31.0%29.0%63.0%61.0%36.0%43.0%——

ASPIRE：Long Task 和 Long Swap 使用 LIBERO-90 技能库进行 zero-shot 迁移。

GPT-6 Astra：Long Task/Swap 与其余六套件使用各自探索后冻结的 memory 文件快照，评测期间不更新。Overall 合并两个不重叠批次：Long 157/200，加上其余套件 584/600，得到 741/800（92.63%）；并非全部回合共享同一份 memory 快照。

LIBERO

方法 / 模型成功率
AtomVLA97.0%
Claude Code / Opus-4.7 / max.reasoning96.0%
π_RLinf95.3%
π094.2%
NORA79.5%
OpenVLA76.5%
diff --git a/docs/source-zh/rst_source/overview.rst b/docs/source-zh/rst_source/overview.rst index 36f7188e0..af8c0749d 100644 --- a/docs/source-zh/rst_source/overview.rst +++ b/docs/source-zh/rst_source/overview.rst @@ -25,13 +25,13 @@ RPent 建立在三条核心设计原则之上: **服务化、标准化、可组 对比 LIBERO、LIBERO-PRO、RoboCasa365 Target50 和 RoboTwin C2R 上的成功率。 排名仅限图中方法及评测范围；完整结果、模型配置和来源见 :doc:`benchmarks`。 -.. image:: https://cdn.jsdelivr.net/gh/RLinf/misc@881ca8850640e1ac5e728865926349e4660032e7/rpent/benchmarks/leaderboard-zh-light.png +.. image:: https://cdn.jsdelivr.net/gh/RLinf/misc@c3b9b5d4ffa360a8324c5b7aa510e1ed0876aa43/rpent/benchmarks/leaderboard-zh-light.png :alt: RPent benchmark results :class: only-light :width: 100% :target: benchmarks.html -.. image:: https://cdn.jsdelivr.net/gh/RLinf/misc@881ca8850640e1ac5e728865926349e4660032e7/rpent/benchmarks/leaderboard-zh-dark.png +.. image:: https://cdn.jsdelivr.net/gh/RLinf/misc@c3b9b5d4ffa360a8324c5b7aa510e1ed0876aa43/rpent/benchmarks/leaderboard-zh-dark.png :alt: RPent benchmark results :class: only-dark :width: 100%