From 3000c22ec8a905ac5343c1af6be38facbfab5659 Mon Sep 17 00:00:00 2001 From: Lum0s517 <132797314+Lum0s517@users.noreply.github.com> Date: Thu, 2 Jul 2026 17:09:31 +0800 Subject: [PATCH 1/6] Update README.md --- workspace/README.md | 1 - 1 file changed, 1 deletion(-) diff --git a/workspace/README.md b/workspace/README.md index e4072ca..7d10382 100644 --- a/workspace/README.md +++ b/workspace/README.md @@ -23,7 +23,6 @@ pip install "dmuon @ git+https://github.com/X-Square-Robot/dmuon.git" git clone https://github.com/huggingface/lerobot.git cd lerobot -git checkout c66cd401767e60baece16e1cf68da2824227e076 pip install --no-deps -e . cd - From 1f65c04ac9b9196e87393c1419ae0e473ea399f4 Mon Sep 17 00:00:00 2001 From: Lum0s517 <132797314+Lum0s517@users.noreply.github.com> Date: Thu, 2 Jul 2026 17:10:15 +0800 Subject: [PATCH 2/6] Update fake_inference.py --- scripts/fake_inference.py | 38 +++++++++++++++++++++++++++++++++++--- 1 file changed, 35 insertions(+), 3 deletions(-) diff --git a/scripts/fake_inference.py b/scripts/fake_inference.py index d14d731..d431d2c 100755 --- a/scripts/fake_inference.py +++ b/scripts/fake_inference.py @@ -13,6 +13,7 @@ from pathlib import Path import numpy as np +import torch def _ensure_local_harrix_on_path() -> None: @@ -54,6 +55,38 @@ def _build_fake_observation(seed: int, image_size: int, gripper_dim: int) -> dic } +def _predict_raw_action(adapter, payload: dict) -> np.ndarray: + from wall_x._vendor.harrix.envs.libero_common import encode_proprio + + observation = encode_proprio( + payload["observation"], adapter._train_config, adapter._action_horizon + ) + prefix, postfix = adapter._get_flow_prompt(payload["instruction"]) + batch_inputs = adapter._construct_model_input([observation], [prefix], [postfix]) + + dataset_names = batch_inputs["dataset_names"] + padding = torch.zeros( + (len(dataset_names), 1, adapter._action_dim), + dtype=torch.float32, + ) + padding_action = adapter._normalizer_action.normalize_data( + padding, dataset_names + ).to(batch_inputs["input_ids"].device) + + model_output = adapter._model.generate_flow_action( + action_horizon=adapter._action_horizon, + action_dim=adapter._action_dim, + num_inference_timesteps=adapter._num_inference_timesteps, + padding_action=padding_action, + noise=payload.get("noise"), + **batch_inputs, + ) + predict_action = model_output["predict_action"] + if isinstance(predict_action, torch.Tensor): + predict_action = predict_action.detach().cpu().numpy() + return np.asarray(predict_action[0]) + + def parse_args() -> argparse.Namespace: parser = argparse.ArgumentParser(description=__doc__) parser.add_argument( @@ -64,7 +97,7 @@ def parse_args() -> argparse.Namespace: default=None, help="Optional training config path. Defaults to config.yml/config.yaml next to the checkpoint.", ) - parser.add_argument("--norm-key", default="libero_all") + parser.add_argument("--norm-key", default="x2_normal") parser.add_argument("--architecture", default="qwen2_5") parser.add_argument("--action-mode", default="flow") parser.add_argument( @@ -110,8 +143,7 @@ def main() -> int: "instruction": args.instruction, "noise": None, } - actions = adapter.predict_batch([payload]) - action = np.asarray(actions[0]) + action = _predict_raw_action(adapter, payload) print("Fake inference succeeded.") print(f"action shape: {action.shape}") From 999b5bb73a82749720e8fe9ada427517e39d8dc3 Mon Sep 17 00:00:00 2001 From: Lum0s517 <132797314+Lum0s517@users.noreply.github.com> Date: Thu, 2 Jul 2026 17:12:41 +0800 Subject: [PATCH 3/6] Update fake_inference.py --- scripts/fake_inference.py | 1 + 1 file changed, 1 insertion(+) diff --git a/scripts/fake_inference.py b/scripts/fake_inference.py index d431d2c..1d69425 100755 --- a/scripts/fake_inference.py +++ b/scripts/fake_inference.py @@ -154,3 +154,4 @@ def main() -> int: if __name__ == "__main__": raise SystemExit(main()) + From 52582b34c0e75df52aa92dba54e2d1e9a59d1ade Mon Sep 17 00:00:00 2001 From: Lum0s517 <132797314+Lum0s517@users.noreply.github.com> Date: Tue, 7 Jul 2026 18:40:55 +0800 Subject: [PATCH 4/6] update paper link to arXiv version --- README.md | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/README.md b/README.md index dbbc33a..d38cf48 100644 --- a/README.md +++ b/README.md @@ -41,9 +41,9 @@ package installation. - [June 2026] Wall-X 1.1.0 updates the open-source training and inference stack for Wall-OSS-0.5, including the public serving/evaluation runtime, DMuon training support, and install-time CUDA operator builds. -- [May 2026] We introduce [**WALL-WM: Carving World Action Modeling at the Event Joints**](https://x2robot.com/api/files/file/WALL-WM.pdf), a World Action Model that couples future-video imagination with action prediction at semantic event boundaries. -- [May 2026] We introduce [**Wall-OSS-0.5: A Deployment-Ready VLA with Gradient-Bridged Pretraining**](https://x2robot.com/api/files/file/wall_oss_05.pdf), an open-source model for directly deployable real-robot manipulation and downstream adaptation. -- [Sept 2025] We introduce [**WALL-OSS: Igniting VLMs toward the Embodied Space**](https://x2robot.com/en/research/68bc2cde8497d7f238dde690), an end-to-end embodied foundation model that leverages large-scale multimodal pretraining to achieve embodiment-aware vision-language understanding, language-action association, and robust manipulation capability. +- [May 2026] We introduce [**WALL-WM: Carving World Action Modeling at the Event Joints**](https://arxiv.org/pdf/2606.01955), a World Action Model that couples future-video imagination with action prediction at semantic event boundaries. +- [May 2026] We introduce [**Wall-OSS-0.5: A Deployment-Ready VLA with Gradient-Bridged Pretraining**](https://arxiv.org/pdf/2605.30877v2), an open-source model for directly deployable real-robot manipulation and downstream adaptation. +- [Sept 2025] We introduce [**WALL-OSS: Igniting VLMs toward the Embodied Space**](https://arxiv.org/pdf/2509.11766), an end-to-end embodied foundation model that leverages large-scale multimodal pretraining to achieve embodiment-aware vision-language understanding, language-action association, and robust manipulation capability. ## Models From 6eb1770da2a47f5f3c10abbfd778ed228cd8cc0f Mon Sep 17 00:00:00 2001 From: Lum0s517 <132797314+Lum0s517@users.noreply.github.com> Date: Tue, 7 Jul 2026 18:43:47 +0800 Subject: [PATCH 5/6] update paper link to arXiv version --- README.zh-CN.md | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/README.zh-CN.md b/README.zh-CN.md index 81e73d9..057cd0b 100644 --- a/README.zh-CN.md +++ b/README.zh-CN.md @@ -5,9 +5,9 @@ Wall-X 开源仓库提供 WALL 系列开源具身基础模型的训练与推理 ## 最新动态 - 2026 年 6 月：Wall-X 1.1.0 更新 Wall-OSS-0.5 训练与推理栈，包含公开 serving/evaluation runtime、DMuon 训练支持，以及安装期 CUDA 算子编译。 -- 2026 年 5 月：发布 [WALL-WM: Carving World Action Modeling at the Event Joints](https://x2robot.com/api/files/file/WALL-WM.pdf)。 -- 2026 年 5 月：发布 [Wall-OSS-0.5: A Deployment-Ready VLA with Gradient-Bridged Pretraining](https://x2robot.com/api/files/file/wall_oss_05.pdf)。 -- 2025 年 9 月：发布 [WALL-OSS: Igniting VLMs toward the Embodied Space](https://x2robot.com/en/research/68bc2cde8497d7f238dde690)。 +- 2026 年 5 月：发布 [WALL-WM: Carving World Action Modeling at the Event Joints](https://arxiv.org/pdf/2606.01955)。 +- 2026 年 5 月：发布 [Wall-OSS-0.5: A Deployment-Ready VLA with Gradient-Bridged Pretraining](https://arxiv.org/pdf/2605.30877v2)。 +- 2025 年 9 月：发布 [WALL-OSS: Igniting VLMs toward the Embodied Space](https://arxiv.org/pdf/2509.11766)。 ## 模型 From 79deb6b76c587900f0876712e8f1ae821061552c Mon Sep 17 00:00:00 2001 From: Lum0s517 <132797314+Lum0s517@users.noreply.github.com> Date: Tue, 7 Jul 2026 19:00:47 +0800 Subject: [PATCH 6/6] Update fake_inference.py --- scripts/fake_inference.py | 1 - 1 file changed, 1 deletion(-) diff --git a/scripts/fake_inference.py b/scripts/fake_inference.py index 1d69425..d431d2c 100755 --- a/scripts/fake_inference.py +++ b/scripts/fake_inference.py @@ -154,4 +154,3 @@ def main() -> int: if __name__ == "__main__": raise SystemExit(main()) -