diff --git a/CLAUDE.md b/CLAUDE.md
index c13c49f3d..812701939 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -17,27 +17,43 @@ pytest
 ## Installation
 
 ```bash
+# macOS (MPS)
 uv sync --extra dev
+
+# Linux (CUDA 12.4)
+uv sync --extra dev --extra cu124
 ```
 
 ## Development Workflow
 
-### 1. 代码开发
+### Quick Commands (Makefile)
+
 ```bash
-uv run python scripts/train_rsl_rl.py --task Go1JoystickFlatTerrain
+make format     # Format and lint code
+make type       # Type check with mypy
+make check      # make format && make type
+make test       # Run all tests
+make test-fast  # Run tests excluding slow ones
+make test-all   # make check && make test
 ```
 
-### 2. 提交前检查
+### Manual Commands
+
 ```bash
-# 格式化
+# Format
 uv run ruff format .
+uv run ruff check --fix
 
-# 检查
-uv run ruff check .
+# Type check
+uv run mypy unilab
+
+# Test
+uv run pytest
 ```
 
-### 3. Git 提交
-使用 Conventional Commits：
+## Git Commits
+
+Use Conventional Commits:
 - `feat:` 新功能
 - `fix:` 修复 bug
 - `docs:` 文档
@@ -46,6 +62,10 @@ uv run ruff check .
 - `test:` 测试
 - `chore:` 构建/工具
 
-### 4. 自动化
-- **Pre-commit hooks**（可选）：`pre-commit install`
-- **GitHub CI**：push 到 main 或 PR 时自动运行 lint + test
+## Pre-commit
+
+```bash
+pre-commit install  # Optional
+```
+
+**Always run `make check` before committing.**
diff --git a/Makefile b/Makefile
new file mode 100644
index 000000000..9c362288f
--- /dev/null
+++ b/Makefile
@@ -0,0 +1,26 @@
+.PHONY: sync
+sync:
+	uv sync --extra dev
+
+.PHONY: format
+format:
+	uv run ruff format
+	uv run ruff check --fix
+
+.PHONY: type
+type:
+	uv run mypy unilab
+
+.PHONY: check
+check: format type
+
+.PHONY: test
+test:
+	uv run pytest
+
+.PHONY: test-fast
+test-fast:
+	uv run pytest -m "not slow"
+
+.PHONY: test-all
+test-all: check test
diff --git a/README.md b/README.md
index f28c8f2a1..d4aa506ba 100644
--- a/README.md
+++ b/README.md
@@ -40,8 +40,12 @@ brew install cmake  # macOS
 # sudo apt-get install cmake  # Ubuntu/Debian
 
 # 4. 同步依赖
+# macOS (MPS)
 uv sync
 
+# Linux (CUDA 12.4)
+uv sync --extra cu124
+
 # 5. 可选：Motrix 后端
 uv sync --extra motrix
 ```
diff --git a/benchmark/benchmark_ane_inference.py b/benchmark/benchmark_ane_inference.py
index 25a3bda47..a5f38d6c9 100644
--- a/benchmark/benchmark_ane_inference.py
+++ b/benchmark/benchmark_ane_inference.py
@@ -12,33 +12,30 @@
 
 try:
     from benchmark.core import (
-        bench_callable,
         MLPBenchRecord,
+        bench_callable,
         mlp_param_count,
-        trimmed_mean,
         print_mlp_table,
+        trimmed_mean,
     )
     from benchmark.core.device_info import get_device_info_dict, get_device_info_line
 except ModuleNotFoundError:
-    from core import bench_callable, MLPBenchRecord, mlp_param_count, trimmed_mean, print_mlp_table
+    from core import MLPBenchRecord, bench_callable, mlp_param_count, print_mlp_table, trimmed_mean
     from core.device_info import get_device_info_dict, get_device_info_line
 
 
 import argparse
 import json
-import os
-import warnings
 import shutil
 import tempfile
-from dataclasses import asdict, dataclass
+from dataclasses import asdict
 from datetime import datetime, timezone
 from pathlib import Path
-from typing import Callable, Dict, List, Optional, Tuple
-
+from typing import Dict, List, Optional
 
+import coremltools as ct
 import numpy as np
 import torch
-import coremltools as ct
 
 try:
     import matplotlib
diff --git a/benchmark/benchmark_ane_peak.py b/benchmark/benchmark_ane_peak.py
index e9ca7a3bc..36866f551 100644
--- a/benchmark/benchmark_ane_peak.py
+++ b/benchmark/benchmark_ane_peak.py
@@ -24,7 +24,7 @@
 from dataclasses import asdict, dataclass
 from datetime import datetime, timezone
 from pathlib import Path
-from typing import List, Optional
+from typing import List
 
 import numpy as np
 
diff --git a/benchmark/benchmark_backends.py b/benchmark/benchmark_backends.py
index 4ded71662..f6568c3ba 100644
--- a/benchmark/benchmark_backends.py
+++ b/benchmark/benchmark_backends.py
@@ -8,28 +8,28 @@
 
 try:
     from benchmark.core import (
+        available_backends,
         bench_callable,
-        parse_sizes,
-        parse_dtypes,
+        mlx_dtype,
         normalize_dtypes,
-        available_backends,
         numpy_dtype,
-        torch_dtype,
-        mlx_dtype,
+        parse_dtypes,
+        parse_sizes,
         print_table,
+        torch_dtype,
     )
     from benchmark.core.device_info import get_device_info_dict
 except ModuleNotFoundError:
     from core import (
+        available_backends,
         bench_callable,
-        parse_sizes,
-        parse_dtypes,
+        mlx_dtype,
         normalize_dtypes,
-        available_backends,
         numpy_dtype,
-        torch_dtype,
-        mlx_dtype,
+        parse_dtypes,
+        parse_sizes,
         print_table,
+        torch_dtype,
     )
     from core.device_info import get_device_info_dict
 
@@ -143,7 +143,8 @@ def run_torch_mps(size, warmup, repeat, dtype_name):
     dt = torch_dtype(dtype_name)
     a = torch.randn(size, size, dtype=dt, device="mps")
     b = torch.randn(size, size, dtype=dt, device="mps")
-    sync = lambda: torch.mps.synchronize()
+    def sync():
+        return torch.mps.synchronize()
 
     mm_t = bench_callable(lambda: a @ b, sync, warmup, repeat)
     ew_t = bench_callable(
diff --git a/benchmark/benchmark_conversions.py b/benchmark/benchmark_conversions.py
index 12269ece4..65f0929c1 100644
--- a/benchmark/benchmark_conversions.py
+++ b/benchmark/benchmark_conversions.py
@@ -14,36 +14,35 @@
 import math
 import platform
 import statistics
-import time
 from dataclasses import asdict, dataclass
 from datetime import datetime, timezone
 from pathlib import Path
-from typing import Any, Callable, Dict, List, Tuple
+from typing import Any, Dict, List, Tuple
 
 try:
     from benchmark.core import (
+        available_backends,
         bench_callable,
-        parse_sizes,
-        parse_dtypes,
+        mlx_dtype,
         normalize_dtypes,
-        available_backends,
         numpy_dtype,
-        torch_dtype,
-        mlx_dtype,
+        parse_dtypes,
+        parse_sizes,
         print_table,
+        torch_dtype,
     )
     from benchmark.core.device_info import get_device_info_dict, get_device_info_line
 except ModuleNotFoundError:
     from core import (
+        available_backends,
         bench_callable,
-        parse_sizes,
-        parse_dtypes,
+        mlx_dtype,
         normalize_dtypes,
-        available_backends,
         numpy_dtype,
-        torch_dtype,
-        mlx_dtype,
+        parse_dtypes,
+        parse_sizes,
         print_table,
+        torch_dtype,
     )
     from core.device_info import get_device_info_dict, get_device_info_line
 
diff --git a/benchmark/benchmark_forward_reset_methods.py b/benchmark/benchmark_forward_reset_methods.py
index 1314eeb85..628c3ff3f 100644
--- a/benchmark/benchmark_forward_reset_methods.py
+++ b/benchmark/benchmark_forward_reset_methods.py
@@ -4,13 +4,12 @@
 import argparse
 import sys
 from dataclasses import asdict, dataclass
-from pathlib import Path
-from typing import List, Tuple
 from multiprocessing import cpu_count
+from pathlib import Path
 
 import mujoco
-from mujoco import batch_forward
 import numpy as np
+from mujoco import batch_forward
 
 try:
     from benchmark.core.device_info import get_device_info_dict, get_device_info_line
diff --git a/benchmark/benchmark_gpu_buffer.py b/benchmark/benchmark_gpu_buffer.py
index 20e88e02d..8b912be9b 100644
--- a/benchmark/benchmark_gpu_buffer.py
+++ b/benchmark/benchmark_gpu_buffer.py
@@ -5,11 +5,12 @@
 Usage: python scripts/benchmark_gpu_buffer.py
 """
 
+import sys
+import time
+
+import numpy as np
 import torch
 import torch.nn as nn
-import numpy as np
-import time
-import sys
 
 
 def sync_device(device):
@@ -72,7 +73,7 @@ def benchmark_host_buffer(capacity, obs_dim, action_dim, batch_size, num_updates
         indices = np.random.randint(0, capacity, size=batch_size)
         obs = torch.from_numpy(obs_buf[indices].copy()).to(device, non_blocking=True)
         actions = torch.from_numpy(actions_buf[indices].copy()).to(device, non_blocking=True)
-        next_obs = torch.from_numpy(next_obs_buf[indices].copy()).to(device, non_blocking=True)
+        torch.from_numpy(next_obs_buf[indices].copy()).to(device, non_blocking=True)
         sync_device(device)
 
         # Training step
@@ -111,7 +112,7 @@ def benchmark_gpu_buffer(capacity, obs_dim, action_dim, batch_size, num_updates,
         indices = torch.randint(0, capacity, (batch_size,), device=device)
         obs = obs_buf[indices]
         actions = actions_buf[indices]
-        next_obs = next_obs_buf[indices]
+        next_obs_buf[indices]
 
         # Training step
         pred_actions = actor(obs)
diff --git a/benchmark/benchmark_ipc_buffer_layout.py b/benchmark/benchmark_ipc_buffer_layout.py
index ef885ccdc..1dbee6ec1 100644
--- a/benchmark/benchmark_ipc_buffer_layout.py
+++ b/benchmark/benchmark_ipc_buffer_layout.py
@@ -10,13 +10,12 @@
     python benchmark/benchmark_ipc_buffer_layout.py
 """
 
-import time
 import sys
+import time
 
 import numpy as np
 import torch
 
-
 # ---------------------------------------------------------------------------
 # helpers
 # ---------------------------------------------------------------------------
diff --git a/benchmark/benchmark_ipc_index_sort.py b/benchmark/benchmark_ipc_index_sort.py
index 810e1eda5..f362b3f3b 100644
--- a/benchmark/benchmark_ipc_index_sort.py
+++ b/benchmark/benchmark_ipc_index_sort.py
@@ -10,13 +10,12 @@
     python benchmark/benchmark_ipc_index_sort.py
 """
 
-import time
 import sys
+import time
 
 import numpy as np
 import torch
 
-
 # ---------------------------------------------------------------------------
 # helpers
 # ---------------------------------------------------------------------------
diff --git a/benchmark/benchmark_ipc_weight_sync.py b/benchmark/benchmark_ipc_weight_sync.py
index 76bdf6b8b..769041e3f 100644
--- a/benchmark/benchmark_ipc_weight_sync.py
+++ b/benchmark/benchmark_ipc_weight_sync.py
@@ -16,16 +16,15 @@
     python benchmark/benchmark_ipc_weight_sync.py
 """
 
-import time
 import sys
 import threading
+import time
 from multiprocessing.shared_memory import SharedMemory
 
 import numpy as np
 import torch
 import torch.nn as nn
 
-
 # ---------------------------------------------------------------------------
 # helpers
 # ---------------------------------------------------------------------------
diff --git a/benchmark/benchmark_mj_step.py b/benchmark/benchmark_mj_step.py
index 37bce3d89..381e10457 100644
--- a/benchmark/benchmark_mj_step.py
+++ b/benchmark/benchmark_mj_step.py
@@ -21,14 +21,14 @@
 from pathlib import Path
 from typing import Dict, List
 
+import matplotlib
 import mujoco
 from mujoco import rollout as mj_rollout
-import matplotlib
 
 matplotlib.use("Agg")
 import matplotlib.pyplot as plt
-from matplotlib.patches import Patch
 import numpy as np
+from matplotlib.patches import Patch
 
 _IS_MACOS = platform.system() == "Darwin"
 
@@ -536,7 +536,7 @@ def main():
     if _IS_MACOS:
         if not _has_native_mujoco_mlx_step():
             raise RuntimeError("Native MLX step backend unavailable. Requires mujoco.mlx_step.")
-        print(f"macOS: native mlx_step available")
+        print("macOS: native mlx_step available")
     else:
         print(f"Non-macOS: using mujoco.rollout (1t vs {args.nthread}t)")
 
diff --git a/benchmark/benchmark_mlp_inference.py b/benchmark/benchmark_mlp_inference.py
index 02ce30e88..421c52452 100644
--- a/benchmark/benchmark_mlp_inference.py
+++ b/benchmark/benchmark_mlp_inference.py
@@ -21,21 +21,20 @@
 
 try:
     from benchmark.core import (
-        bench_callable,
-        available_backends,
         MLPBenchRecord,
+        available_backends,
+        bench_callable,
         mlp_param_count,
-        trimmed_mean,
         print_mlp_table,
+        trimmed_mean,
     )
     from benchmark.core.device_info import get_device_info_dict, get_device_info_line
 except ModuleNotFoundError:
     from core import (
-        bench_callable,
-        available_backends,
         MLPBenchRecord,
+        available_backends,
+        bench_callable,
         mlp_param_count,
-        trimmed_mean,
         print_mlp_table,
     )
     from core.device_info import get_device_info_dict, get_device_info_line
@@ -44,14 +43,11 @@
 import json
 import os
 import statistics
-import sys
 import time
-import warnings
-from dataclasses import asdict, dataclass
+from dataclasses import asdict
 from datetime import datetime, timezone
 from pathlib import Path
-from typing import Callable, Dict, List, Optional, Tuple
-
+from typing import Dict, List, Optional, Tuple
 
 try:
     import numpy as np
@@ -1266,7 +1262,7 @@ def run_phase(runners: List[tuple], backends_dict: Dict[str, bool]) -> None:
                 )
             except Exception:
                 pass
-    backends_p4 = available_backends(include_lazy=True)
+    available_backends(include_lazy=True)
     if onnx_session is not None:
         print("\nDetected backends (phase 4: ONNX Runtime):", flush=True)
         print("  - onnxruntime: yes", flush=True)
diff --git a/benchmark/benchmark_mujoco_vs_motrix.py b/benchmark/benchmark_mujoco_vs_motrix.py
index 72ea8c7ca..f5da93cb7 100644
--- a/benchmark/benchmark_mujoco_vs_motrix.py
+++ b/benchmark/benchmark_mujoco_vs_motrix.py
@@ -6,7 +6,7 @@
 import argparse
 import os
 import time
-from pathlib import Path
+
 import numpy as np
 
 try:
diff --git a/benchmark/benchmark_postprocess.py b/benchmark/benchmark_postprocess.py
index a42f5634d..30fedcbd6 100644
--- a/benchmark/benchmark_postprocess.py
+++ b/benchmark/benchmark_postprocess.py
@@ -11,9 +11,9 @@
 
 matplotlib.use("Agg")
 import matplotlib.pyplot as plt
+import mlx.core as mx
 import numpy as np
 import torch
-import mlx.core as mx
 
 from unilab.base import registry
 
diff --git a/benchmark/benchmark_sample_overhead.py b/benchmark/benchmark_sample_overhead.py
index 9a3dd45e4..591deabac 100644
--- a/benchmark/benchmark_sample_overhead.py
+++ b/benchmark/benchmark_sample_overhead.py
@@ -1,12 +1,11 @@
 #!/usr/bin/env python3
 """Realistic sampling benchmark - measures actual sample overhead."""
 
-import torch
-import numpy as np
-import time
-import sys
 import multiprocessing as mp
-from multiprocessing import shared_memory
+import time
+
+import numpy as np
+import torch
 
 
 def sync_device(device):
@@ -44,12 +43,12 @@ def benchmark_host_sample_realistic(capacity, obs_dim, action_dim, batch_size, n
             truncated_copy = truncated[indices].copy()
 
         # Transfer to device (non-blocking)
-        obs_t = torch.from_numpy(obs_copy).to(device, non_blocking=True)
-        actions_t = torch.from_numpy(actions_copy).to(device, non_blocking=True)
-        rewards_t = torch.from_numpy(rewards_copy).to(device, non_blocking=True)
-        next_obs_t = torch.from_numpy(next_obs_copy).to(device, non_blocking=True)
-        dones_t = torch.from_numpy(dones_copy).to(device, non_blocking=True)
-        truncated_t = torch.from_numpy(truncated_copy).to(device, non_blocking=True)
+        torch.from_numpy(obs_copy).to(device, non_blocking=True)
+        torch.from_numpy(actions_copy).to(device, non_blocking=True)
+        torch.from_numpy(rewards_copy).to(device, non_blocking=True)
+        torch.from_numpy(next_obs_copy).to(device, non_blocking=True)
+        torch.from_numpy(dones_copy).to(device, non_blocking=True)
+        torch.from_numpy(truncated_copy).to(device, non_blocking=True)
 
         sync_device(device)
         elapsed = time.perf_counter() - start
@@ -74,12 +73,12 @@ def benchmark_gpu_sample(capacity, obs_dim, action_dim, batch_size, num_samples,
         start = time.perf_counter()
 
         indices = torch.randint(0, capacity, (batch_size,), device=device)
-        obs_t = obs[indices]
-        actions_t = actions[indices]
-        rewards_t = rewards[indices]
-        next_obs_t = next_obs[indices]
-        dones_t = dones[indices]
-        truncated_t = truncated[indices]
+        obs[indices]
+        actions[indices]
+        rewards[indices]
+        next_obs[indices]
+        dones[indices]
+        truncated[indices]
 
         sync_device(device)
         elapsed = time.perf_counter() - start
diff --git a/benchmark/benchmark_sim.py b/benchmark/benchmark_sim.py
index fffd8f023..975e70b05 100644
--- a/benchmark/benchmark_sim.py
+++ b/benchmark/benchmark_sim.py
@@ -15,13 +15,12 @@
 
 import argparse
 import json
-import math
 import os
 import time
 from dataclasses import asdict, dataclass
 from datetime import datetime, timezone
 from pathlib import Path
-from typing import Dict, List, Optional
+from typing import List
 
 try:
     from benchmark.core.device_info import get_device_info_dict, get_device_info_line
diff --git a/benchmark/core/__init__.py b/benchmark/core/__init__.py
index 66137d2a9..c5dbac369 100644
--- a/benchmark/core/__init__.py
+++ b/benchmark/core/__init__.py
@@ -1,13 +1,13 @@
 """Core benchmark utilities for standardized benchmarking."""
 
-from .backends import available_backends, numpy_dtype, torch_dtype, mlx_dtype, sync_backend
+from .backends import available_backends, mlx_dtype, numpy_dtype, sync_backend, torch_dtype
 from .device_info import get_device_info_dict, get_device_info_line
-from .mlp_utils import MLPBenchRecord, env_nums_pow2, mlp_param_count, trimmed_mean, print_mlp_table
+from .mlp_utils import MLPBenchRecord, env_nums_pow2, mlp_param_count, print_mlp_table, trimmed_mean
 from .output import print_table, save_json
 from .plotting import save_line_plot
 from .record import BenchRecord
 from .runner import bench_callable, summarize
-from .utils import parse_sizes, pow2_sizes, parse_dtypes, normalize_dtypes
+from .utils import normalize_dtypes, parse_dtypes, parse_sizes, pow2_sizes
 
 __all__ = [
     "available_backends",
diff --git a/benchmark/core/output.py b/benchmark/core/output.py
index 909e465de..6ce9e0174 100644
--- a/benchmark/core/output.py
+++ b/benchmark/core/output.py
@@ -15,7 +15,8 @@ def print_table(records: List[Dict[str, Any]], headers: List[str]) -> None:
     for row in rows:
         for i, v in enumerate(row):
             col_w[i] = max(col_w[i], len(v))
-    fmt = lambda vals: " | ".join(v.ljust(col_w[i]) for i, v in enumerate(vals))
+    def fmt(vals):
+        return " | ".join(v.ljust(col_w[i]) for i, v in enumerate(vals))
     print(fmt(headers))
     print("-+-".join("-" * w for w in col_w))
     for row in rows:
diff --git a/benchmark/core/plotting.py b/benchmark/core/plotting.py
index 13c7270f9..b13f4a082 100644
--- a/benchmark/core/plotting.py
+++ b/benchmark/core/plotting.py
@@ -1,14 +1,13 @@
 """Plotting utilities for benchmark results."""
 
 from pathlib import Path
-from typing import List, Dict, Any
+from typing import Any, Dict, List
 
 try:
     import matplotlib
 
     matplotlib.use("Agg")
     import matplotlib.pyplot as plt
-    import matplotlib.ticker as mticker
 except Exception:
     plt = None
 
diff --git a/benchmark/core/visualize_collection_profile.py b/benchmark/core/visualize_collection_profile.py
index fc22bed36..0145d7773 100644
--- a/benchmark/core/visualize_collection_profile.py
+++ b/benchmark/core/visualize_collection_profile.py
@@ -11,7 +11,6 @@
 
 import matplotlib.pyplot as plt
 
-
 ITER_RE = re.compile(
     r"\[iter\s+(\d+)/(\d+)\].*?collect=([0-9.]+)s.*?"
     r"prof\(act/step/core/post/reset/buf/fin/ep\)="
diff --git a/benchmark/smoke_test.py b/benchmark/smoke_test.py
index 0cb86305b..d3e952dfa 100644
--- a/benchmark/smoke_test.py
+++ b/benchmark/smoke_test.py
@@ -27,7 +27,7 @@
     print(f"Testing {name}...")
     try:
         mod = __import__(f"benchmark.{name}", fromlist=[name])
-        print(f"  ✓ Import OK")
+        print("  ✓ Import OK")
         passed.append(name)
     except Exception as e:
         print(f"  ✗ Import failed: {e}")
diff --git a/pyproject.toml b/pyproject.toml
index f5710c2f5..97aac1125 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -17,7 +17,7 @@ dependencies = [
     "etils",
     "ml_collections",
     "tensordict",
-    "rsl-rl-lib>=3.3.0,<5.0.0",
+    "rsl-rl-lib>=5.0.0",
     "packaging",
     "mediapy",
     "tensorboard",
@@ -48,14 +48,8 @@ name = "test-pypi"
 url = "https://test.pypi.org/simple/"
 explicit = true
 
-[[tool.uv.index]]
-name = "pytorch-cu121"
-url = "https://download.pytorch.org/whl/cu121"
-explicit = true
-
 [tool.uv.sources]
 mujoco-uni = { index = "test-pypi" }
-torch = { index = "pytorch-cu121", marker = "sys_platform == 'linux'" }
 
 [tool.hatch.build.targets.wheel]
 packages = ["unilab"]
diff --git a/scripts/play_interactive.py b/scripts/play_interactive.py
index e6a454746..a2ef0c360 100644
--- a/scripts/play_interactive.py
+++ b/scripts/play_interactive.py
@@ -13,18 +13,18 @@
     Right-drag     - pan
 """
 
+import argparse
+import importlib
 import os
+import pkgutil
 import sys
 import time
-import argparse
-import pkgutil
-import importlib
 from pathlib import Path
 
-import numpy as np
-import torch
 import mujoco
 import mujoco.viewer
+import numpy as np
+import torch
 
 ROOT_DIR = Path(__file__).parent.parent
 sys.path.append(str(ROOT_DIR))
@@ -48,7 +48,7 @@ def ensure_registries():
 
 from unilab.base import registry
 from unilab.config import locomotion_params
-from unilab.utils.rsl_rl_compat import is_rsl_rl_v4, convert_config_v3_to_v4
+from unilab.utils.rsl_rl_compat import convert_config_v3_to_v4, is_rsl_rl_v4
 from unilab.utils.run_utils import get_latest_run
 from unilab.utils.torch_utils import to_torch
 
@@ -60,7 +60,6 @@ def ensure_registries():
 
 from tensordict import TensorDict
 
-
 # ---------------------------------------------------------------------------
 # Minimal env wrapper (mirrors RslRlVecEnvWrapper in train_rsl_rl.py)
 # ---------------------------------------------------------------------------
diff --git a/scripts/train_appo.py b/scripts/train_appo.py
index 705283015..d616f11d2 100644
--- a/scripts/train_appo.py
+++ b/scripts/train_appo.py
@@ -1,13 +1,14 @@
 """Train APPO agent — native multiprocessing."""
 
 import argparse
-import sys
-import os
 import datetime
+import importlib
+import os
+import pkgutil
+import sys
 from pathlib import Path
+
 import torch
-import pkgutil
-import importlib
 
 ROOT_DIR = Path(__file__).parent.parent
 sys.path.append(str(ROOT_DIR))
@@ -43,14 +44,14 @@ def ensure_registries():
 
 def play_appo(args, rl_cfg):
     """Play mode for APPO."""
-    import torch
-    import numpy as np
     import mediapy as media
+    import numpy as np
+    from rsl_rl.utils import resolve_callable
     from tensordict import TensorDict
+
     from unilab.base import registry
     from unilab.utils import render_many
-    from rsl_rl.utils import resolve_callable
-    from unilab.utils.rsl_rl_compat import is_rsl_rl_v4, convert_config_v3_to_v4
+    from unilab.utils.rsl_rl_compat import convert_config_v3_to_v4, is_rsl_rl_v4
 
     # Normalize to plain dict so ConfigDict doesn't cause isinstance issues
     if hasattr(rl_cfg, "to_dict"):
diff --git a/scripts/train_mlx_ppo.py b/scripts/train_mlx_ppo.py
index 9d988f924..20c341dea 100644
--- a/scripts/train_mlx_ppo.py
+++ b/scripts/train_mlx_ppo.py
@@ -5,17 +5,17 @@
 from __future__ import annotations
 
 import argparse
-from collections import deque
 import datetime
 import importlib
 import json
+import math
 import os
 import pickle
 import pkgutil
-import math
 import statistics
 import sys
 import time
+from collections import deque
 from pathlib import Path
 
 import mlx.core as mx
@@ -45,12 +45,12 @@ def ensure_registries() -> None:
 
 ensure_registries()
 
-from unilab.config import locomotion_params
+from unilab.algos.mlx.common import EmpiricalDiscountedVariationNormalization, RolloutBuffer
+from unilab.algos.mlx.ppo import MLPActorCritic, PPOConfig, PPOTrainer
 from unilab.base import registry
+from unilab.config import locomotion_params
 from unilab.utils import render_many
 from unilab.utils.onpolicy_logger import OnPolicyLogger
-from unilab.algos.mlx.common import EmpiricalDiscountedVariationNormalization, RolloutBuffer
-from unilab.algos.mlx.ppo import MLPActorCritic, PPOConfig, PPOTrainer
 
 TASK_STEP_TUNING = {
     # Tuned for faster collection_time on each task.
@@ -159,8 +159,8 @@ def play_mlx_ppo(args, cfg, dtype, use_fp16, resolved_sim_backend, task_log_root
     """Play mode for MLX PPO."""
     import mlx.core as mx
     import numpy as np
+
     from unilab.base import registry
-    from unilab.utils import render_many
 
     play_model_dtype = mx.float32 if use_fp16 else dtype
     play_env_num = args.play_env_num
diff --git a/scripts/train_offpolicy.py b/scripts/train_offpolicy.py
index aa1474b8a..7c2a0b561 100644
--- a/scripts/train_offpolicy.py
+++ b/scripts/train_offpolicy.py
@@ -2,10 +2,10 @@
 
 from __future__ import annotations
 
+import argparse
+import datetime
 import os
 import sys
-import datetime
-import argparse
 from pathlib import Path
 
 ROOT_DIR = Path(__file__).parent.parent
@@ -197,6 +197,7 @@ def play_offpolicy(algo_name: str, args, cfg) -> None:
     import mediapy as media
     import numpy as np
     import torch
+
     from unilab.base import registry
     from unilab.utils import render_many
     from unilab.utils.algo_utils import build_actor
@@ -214,7 +215,7 @@ def play_offpolicy(algo_name: str, args, cfg) -> None:
             "sac", obs_dim, action_dim, cfg.actor_hidden_dim, cfg.use_layer_norm, device
         )
     elif algo_name == "td3":
-        from unilab.algos.torch.fast_td3.learner import TD3Actor, EmpiricalNormalization
+        from unilab.algos.torch.fast_td3.learner import EmpiricalNormalization, TD3Actor
 
         actor = TD3Actor(
             obs_dim,
diff --git a/scripts/train_rsl_rl.py b/scripts/train_rsl_rl.py
index 8caa2237f..c58e718e3 100644
--- a/scripts/train_rsl_rl.py
+++ b/scripts/train_rsl_rl.py
@@ -1,14 +1,14 @@
-import os
-import sys
 import argparse
-import numpy as np
 import datetime
-from pathlib import Path
-import pkgutil
 import importlib
+import os
+import pkgutil
+import sys
+from pathlib import Path
+
+import numpy as np
 import torch
 from tensordict import TensorDict
-import mediapy as media
 
 # Add workspace root to python path dynamically
 ROOT_DIR = Path(__file__).parent.parent
@@ -34,8 +34,7 @@ def ensure_registries():
 
 from unilab.base import registry
 from unilab.config import locomotion_params, manipulation_params
-from unilab.utils import render_many
-from unilab.utils.torch_utils import to_torch, to_numpy
+from unilab.utils.torch_utils import to_numpy, to_torch
 
 # Try importing rsl_rl
 try:
@@ -44,7 +43,7 @@ def ensure_registries():
     print("Could not import rsl_rl. Please ensure it is installed.")
     sys.exit(1)
 
-from unilab.utils.rsl_rl_compat import is_rsl_rl_v4, convert_config_v3_to_v4
+from unilab.utils.rsl_rl_compat import convert_config_v3_to_v4, is_rsl_rl_v4
 from unilab.utils.run_utils import get_latest_run
 
 
@@ -180,8 +179,8 @@ def step(self, actions):
 def play_rsl_rl(args, cfg, device):
     """Play mode for RSL-RL."""
     import torch
+
     from unilab.base import registry
-    from unilab.utils.torch_utils import to_numpy
 
     env = registry.make(args.task, num_envs=args.play_env_num, sim_backend=args.sim_backend)
     wrapped_env = RslRlVecEnvWrapper(env, device=device)
@@ -258,6 +257,7 @@ def play_rsl_rl(args, cfg, device):
     else:
         # MuJoCo backend: render to video
         import mediapy as media
+
         from unilab.utils import render_many
 
         output_video = Path(load_path_dir) / "play_video.mp4"
diff --git a/scripts/visualization_env.py b/scripts/visualization_env.py
index fc1b9e04c..e0817e330 100644
--- a/scripts/visualization_env.py
+++ b/scripts/visualization_env.py
@@ -4,11 +4,10 @@
 sys.path.append(str(Path(__file__).parent.parent))
 
 import argparse
-import numpy as np
 import time
-import unilab.envs.locomotion.g1
-import unilab.envs.locomotion.go1
-import unilab.envs.locomotion.go2
+
+import numpy as np
+
 from unilab.base import registry
 
 parser = argparse.ArgumentParser(description="Visualize robot tasks")
diff --git a/unilab/__init__.py b/unilab/__init__.py
index 2efc11c85..d19618207 100644
--- a/unilab/__init__.py
+++ b/unilab/__init__.py
@@ -1,6 +1,7 @@
 """UniLab package initialization."""
 
 import os
+
 from etils import epath
 
 __version__ = "0.0.0"
diff --git a/unilab/algos/mlx/ppo/model.py b/unilab/algos/mlx/ppo/model.py
index 65089f8e7..6296c2f5d 100644
--- a/unilab/algos/mlx/ppo/model.py
+++ b/unilab/algos/mlx/ppo/model.py
@@ -8,7 +8,7 @@
 import mlx.core as mx
 import mlx.nn as nn
 
-from unilab.algos.mlx.common import EmpiricalNormalization, MLP, diag_gaussian_log_prob
+from unilab.algos.mlx.common import MLP, EmpiricalNormalization, diag_gaussian_log_prob
 
 
 class MLPActorCritic(nn.Module):
diff --git a/unilab/algos/mlx/ppo/ppo.py b/unilab/algos/mlx/ppo/ppo.py
index 05f24fb3d..ef640f3a0 100644
--- a/unilab/algos/mlx/ppo/ppo.py
+++ b/unilab/algos/mlx/ppo/ppo.py
@@ -2,8 +2,8 @@
 
 from __future__ import annotations
 
-from dataclasses import dataclass
 import math
+from dataclasses import dataclass
 from typing import Dict
 
 import mlx.core as mx
@@ -12,6 +12,7 @@
 from mlx.utils import tree_flatten, tree_map
 
 from unilab.algos.mlx.common import RolloutBuffer, diag_gaussian_entropy, diag_gaussian_log_prob
+
 from .model import MLPActorCritic
 
 
diff --git a/unilab/algos/mlx/ppo/runner.py b/unilab/algos/mlx/ppo/runner.py
index 843c5f448..1dd5b4e31 100644
--- a/unilab/algos/mlx/ppo/runner.py
+++ b/unilab/algos/mlx/ppo/runner.py
@@ -13,6 +13,7 @@
 from mlx.utils import tree_map
 
 from unilab.algos.mlx.common import EmpiricalDiscountedVariationNormalization, RolloutBuffer
+
 from .model import MLPActorCritic
 from .ppo import PPOConfig, PPOTrainer
 
diff --git a/unilab/algos/torch/appo/learner.py b/unilab/algos/torch/appo/learner.py
index 837976974..c6a81a8ad 100644
--- a/unilab/algos/torch/appo/learner.py
+++ b/unilab/algos/torch/appo/learner.py
@@ -10,15 +10,13 @@
 """
 
 import copy
+from itertools import chain
 
 import torch
 import torch.nn as nn
-import torch.optim as optim
-from tensordict import TensorDict
-
 from rsl_rl.models import MLPModel
 from rsl_rl.utils import resolve_optimizer
-from itertools import chain
+from tensordict import TensorDict
 
 
 def vtrace_advantages(
@@ -88,7 +86,6 @@ def vtrace_advantages(
     return vs, advantages
 
 
-import torch.nn as nn
 from torch.distributions import Normal
 
 
diff --git a/unilab/algos/torch/appo/runner.py b/unilab/algos/torch/appo/runner.py
index bc349d712..3427d8df8 100644
--- a/unilab/algos/torch/appo/runner.py
+++ b/unilab/algos/torch/appo/runner.py
@@ -9,16 +9,16 @@
 import multiprocessing as mp
 import os
 import time
-import statistics
+from collections import deque
+
 import torch
-from collections import defaultdict, deque
+from rsl_rl.utils import resolve_callable
 
-from unilab.ipc import AsyncRunner, SharedOnPolicyStorage, SharedWeightSync
+from unilab.algos.torch.appo.learner import APPOActorWrapper, APPOLearner
 from unilab.algos.torch.appo.worker import appo_collector_fn
-from unilab.algos.torch.appo.learner import APPOLearner, APPOActorWrapper
+from unilab.ipc import AsyncRunner, SharedOnPolicyStorage, SharedWeightSync
 from unilab.utils.offpolicy_logger import OffPolicyLogger
 from unilab.utils.rsl_rl_compat import convert_config_v3_to_v4, is_rsl_rl_v4
-from rsl_rl.utils import resolve_callable
 
 
 class APPORunner(AsyncRunner):
@@ -93,8 +93,8 @@ def _build_learner(self):
         if is_rsl_rl_v4():
             cfg = convert_config_v3_to_v4(cfg)
 
-        from tensordict import TensorDict
         import torch
+        from tensordict import TensorDict
 
         obs_example = torch.zeros((self.num_envs, self.obs_dim), device=self.device)
         td_example = TensorDict({"policy": obs_example}, batch_size=self.num_envs)
@@ -207,9 +207,8 @@ def learn(
         logger.start()
         logger.log_status("Waiting for first rollout...")
 
-        reward_history = deque(maxlen=100)
-        start_time = time.time()
-        total_steps = 0
+        deque(maxlen=100)
+        time.time()
         last_metrics_msg = {}
 
         for iteration in range(1, max_iterations + 1):
diff --git a/unilab/algos/torch/appo/worker.py b/unilab/algos/torch/appo/worker.py
index a96b994f0..7179e81f7 100644
--- a/unilab/algos/torch/appo/worker.py
+++ b/unilab/algos/torch/appo/worker.py
@@ -3,12 +3,12 @@
 Collects on-policy rollouts and writes to SharedOnPolicyStorage.
 """
 
+import statistics
 import sys
 from collections import defaultdict
-import statistics
 
-import torch
 import numpy as np
+import torch
 from rsl_rl.utils import resolve_callable
 
 from unilab.utils.algo_utils import ensure_registries
@@ -33,9 +33,10 @@ def appo_collector_fn(
 
     Creates environment + policy, collects rollouts, writes to SharedOnPolicyStorage.
     """
-    from unilab.ipc import SharedOnPolicyStorage, SharedWeightSync
-    from unilab.base import registry
     from tensordict import TensorDict
+
+    from unilab.base import registry
+    from unilab.ipc import SharedOnPolicyStorage, SharedWeightSync
     from unilab.utils.rsl_rl_compat import convert_config_v3_to_v4, is_rsl_rl_v4
 
     ensure_registries()
diff --git a/unilab/algos/torch/common/__init__.py b/unilab/algos/torch/common/__init__.py
index 0e5aa789a..b851d2614 100644
--- a/unilab/algos/torch/common/__init__.py
+++ b/unilab/algos/torch/common/__init__.py
@@ -1,8 +1,8 @@
+from unilab.algos.torch.common.networks import Critic, DistributionalQNetwork
 from unilab.algos.torch.common.normalization import EmpiricalNormalization
-from unilab.algos.torch.common.networks import DistributionalQNetwork, Critic
 from unilab.algos.torch.common.stability import check_nan_loss, clip_gradients, safe_tensor
+from unilab.utils.algo_utils import build_actor, ensure_registries
 from unilab.utils.offpolicy_logger import OffPolicyLogger
-from unilab.utils.algo_utils import ensure_registries, build_actor
 
 __all__ = [
     "EmpiricalNormalization",
diff --git a/unilab/algos/torch/common/stability.py b/unilab/algos/torch/common/stability.py
index 97447a85c..5c9aba24f 100644
--- a/unilab/algos/torch/common/stability.py
+++ b/unilab/algos/torch/common/stability.py
@@ -1,7 +1,6 @@
 """Numerical stability utilities for RL training."""
 
 import torch
-import torch.nn as nn
 
 
 def check_nan_loss(
diff --git a/unilab/algos/torch/fast_sac/learner.py b/unilab/algos/torch/fast_sac/learner.py
index a7d0745fb..2c8407ef4 100644
--- a/unilab/algos/torch/fast_sac/learner.py
+++ b/unilab/algos/torch/fast_sac/learner.py
@@ -11,14 +11,12 @@
 from __future__ import annotations
 
 import math
+from typing import Dict, Tuple
+
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
 import torch.optim as optim
-from typing import Dict, Tuple
-
-from unilab.algos.torch.common.normalization import EmpiricalNormalization
-
 
 # ---------------------------------------------------------------------------
 # Actor Network (holosoma-style: SiLU + LayerNorm + Tanh squashing)
diff --git a/unilab/algos/torch/fast_sac/runner.py b/unilab/algos/torch/fast_sac/runner.py
index d6071ec1d..1a3794ee8 100644
--- a/unilab/algos/torch/fast_sac/runner.py
+++ b/unilab/algos/torch/fast_sac/runner.py
@@ -36,9 +36,10 @@ def __init__(
         sim_backend: str = "mujoco",
         use_symmetry: bool = False,
     ):
+        import torch
+
         from unilab.base import registry
         from unilab.utils.algo_utils import ensure_registries
-        import torch
 
         ensure_registries()
         env = registry.make(env_name, num_envs=1, sim_backend=sim_backend)
diff --git a/unilab/algos/torch/fast_td3/learner.py b/unilab/algos/torch/fast_td3/learner.py
index 999b63566..5de9f4573 100644
--- a/unilab/algos/torch/fast_td3/learner.py
+++ b/unilab/algos/torch/fast_td3/learner.py
@@ -16,14 +16,15 @@
 from __future__ import annotations
 
 import math
+from typing import Dict
+
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
 import torch.optim as optim
-from typing import Dict
 
-from unilab.algos.torch.common.normalization import EmpiricalNormalization
 from unilab.algos.torch.common.networks import Critic
+from unilab.algos.torch.common.normalization import EmpiricalNormalization
 from unilab.algos.torch.common.stability import check_nan_loss, clip_gradients
 
 # ---------------------------------------------------------------------------
diff --git a/unilab/algos/torch/offpolicy/runner.py b/unilab/algos/torch/offpolicy/runner.py
index a1feea9a2..f0a83b0b5 100644
--- a/unilab/algos/torch/offpolicy/runner.py
+++ b/unilab/algos/torch/offpolicy/runner.py
@@ -2,15 +2,15 @@
 
 import os
 import time
-import torch
 from collections import deque
 
-from unilab.ipc import SharedWeightSync, SharedObsNormStats
-from unilab.ipc.replay_buffer import ReplayBuffer
-from unilab.ipc.async_runner import AsyncRunner
+import torch
+
 from unilab.algos.torch.offpolicy.worker import off_policy_collector_fn
+from unilab.ipc import SharedObsNormStats, SharedWeightSync
+from unilab.ipc.async_runner import _SPAWN_CTX, AsyncRunner
+from unilab.ipc.replay_buffer import ReplayBuffer
 from unilab.utils.offpolicy_logger import OffPolicyLogger
-from unilab.ipc.async_runner import _SPAWN_CTX
 
 
 class OffPolicyRunner(AsyncRunner):
@@ -258,7 +258,7 @@ def learn(
 
             self.learner.update_count += 1
             weight_sync.write_weights(self.learner.actor.state_dict())
-            sync_time = time.time() - sync_start
+            time.time() - sync_start
             train_time = time.time() - train_start
 
             if self.sync_collection and trainer_done_queue:
diff --git a/unilab/algos/torch/offpolicy/worker.py b/unilab/algos/torch/offpolicy/worker.py
index b70586ddb..a6142f39b 100644
--- a/unilab/algos/torch/offpolicy/worker.py
+++ b/unilab/algos/torch/offpolicy/worker.py
@@ -4,9 +4,10 @@
 actor policy. Runs in a subprocess; writes to ReplayBuffer.
 """
 
-import torch
 import numpy as np
-from unilab.utils.algo_utils import ensure_registries, build_actor
+import torch
+
+from unilab.utils.algo_utils import build_actor, ensure_registries
 
 
 def off_policy_collector_fn(
@@ -32,8 +33,8 @@ def off_policy_collector_fn(
     **kwargs,
 ):
     """Entry point for the off-policy collector subprocess."""
-    import traceback
     import sys
+    import traceback
 
     try:
         print("[Collector] Entry point called", file=sys.stderr, flush=True)
@@ -89,8 +90,8 @@ def _run_collector(
     shared_obs_normalizer_stats,
     sim_backend,
 ):
-    from unilab.ipc import SharedWeightSync
     from unilab.base import registry
+    from unilab.ipc import SharedWeightSync
 
     ensure_registries()
 
diff --git a/unilab/base/backend/__init__.py b/unilab/base/backend/__init__.py
index 2d8dab6ad..989165aba 100644
--- a/unilab/base/backend/__init__.py
+++ b/unilab/base/backend/__init__.py
@@ -1,7 +1,7 @@
 from .base import SimBackend
-from .mujoco_backend import MuJoCoBackend
-from .motrix_backend import MotrixBackend, MOTRIX_AVAILABLE
+from .motrix_backend import MOTRIX_AVAILABLE, MotrixBackend
 from .motrix_numba_backend import MotrixNumbaBackend
+from .mujoco_backend import MuJoCoBackend
 
 
 def create_backend(
diff --git a/unilab/base/backend/base.py b/unilab/base/backend/base.py
index 716a7fabe..3dff6c04b 100644
--- a/unilab/base/backend/base.py
+++ b/unilab/base/backend/base.py
@@ -1,4 +1,5 @@
 import abc
+
 import numpy as np
 
 
diff --git a/unilab/base/backend/motrix_backend.py b/unilab/base/backend/motrix_backend.py
index 86695e373..cdaeba32d 100644
--- a/unilab/base/backend/motrix_backend.py
+++ b/unilab/base/backend/motrix_backend.py
@@ -37,7 +37,7 @@ def __init__(
         self.backend_type = "motrix"
 
     def _process_rigid_body_props(self, cfg) -> None:
-        if cfg.domain_rand.randomize_base_mass == True:
+        if cfg.domain_rand.randomize_base_mass:
             mass = self._model.get_link(cfg.asset.body_name).get_mass_override(self._data)
             mass_low = cfg.domain_rand.added_mass_range[0]
             mass_high = cfg.domain_rand.added_mass_range[1]
@@ -45,7 +45,7 @@ def _process_rigid_body_props(self, cfg) -> None:
             self._model.get_link(cfg.asset.body_name).set_mass_override(self._data, random_mass)
             mass = self._model.get_link(cfg.asset.body_name).get_mass_override(self._data)
 
-        if cfg.domain_rand.random_com == True:
+        if cfg.domain_rand.random_com:
             com_offset = np.zeros(
                 (self._num_envs, 3), dtype=np.float32
             )  # [x_offset, y_offset, z_offset]
@@ -55,7 +55,7 @@ def _process_rigid_body_props(self, cfg) -> None:
             self._model.get_link(cfg.asset.body_name).set_center_of_mass_override(
                 self._data, com_offset
             )
-            com_get = self._model.get_link(cfg.asset.body_name).get_center_of_mass_override(
+            self._model.get_link(cfg.asset.body_name).get_center_of_mass_override(
                 self._data
             )
 
diff --git a/unilab/base/backend/mujoco_backend.py b/unilab/base/backend/mujoco_backend.py
index 7189c7a3e..9f0162612 100644
--- a/unilab/base/backend/mujoco_backend.py
+++ b/unilab/base/backend/mujoco_backend.py
@@ -1,10 +1,12 @@
 import os
 from multiprocessing import cpu_count
+
 import mujoco
-from mujoco import rollout, batch_forward
 import numpy as np
-from .base import SimBackend
+from mujoco import batch_forward, rollout
+
 from ..dtype_config import get_global_dtype
+from .base import SimBackend
 
 
 class MuJoCoBackend(SimBackend):
diff --git a/unilab/base/curriculum.py b/unilab/base/curriculum.py
index 1f021918c..ff2ba206d 100644
--- a/unilab/base/curriculum.py
+++ b/unilab/base/curriculum.py
@@ -1,9 +1,11 @@
 """Curriculum learning for adaptive difficulty adjustment."""
 
 from __future__ import annotations
-import numpy as np
+
 from typing import TYPE_CHECKING
 
+import numpy as np
+
 if TYPE_CHECKING:
     from unilab.envs.locomotion.g1.joystick import G1JoystickEnv
 
diff --git a/unilab/base/np_env.py b/unilab/base/np_env.py
index a9c02e1ae..6514285bd 100644
--- a/unilab/base/np_env.py
+++ b/unilab/base/np_env.py
@@ -1,13 +1,12 @@
 import abc
 import dataclasses
 from dataclasses import dataclass
-from typing import Tuple
+from typing import Optional, Tuple
+
 import numpy as np
-import gymnasium as gym
-from typing import Optional
 
-from unilab.base.base import ABEnv, EnvCfg
 from unilab.base.backend import SimBackend
+from unilab.base.base import ABEnv, EnvCfg
 from unilab.base.dtype_config import get_global_dtype
 
 
@@ -40,7 +39,7 @@ def __init__(self, cfg: EnvCfg, backend: SimBackend, num_envs: int):
         self.push_robots_flag = False
         if self._backend.backend_type == "motrix":
             self._backend._process_rigid_body_props(cfg)
-            if self._cfg.domain_rand.push_robots == True:
+            if self._cfg.domain_rand.push_robots:
                 self.push_robots_flag = True
 
     @property
@@ -140,7 +139,7 @@ def _reset_done_envs(self):
                     self._state.info[key][env_indices] = value
 
     def push_robots(self):
-        if self.push_robots_flag == True:
+        if self.push_robots_flag:
             if self.step_counter % self._cfg.domain_rand.push_interval == 0:
                 self._backend.push_robots(self._cfg.domain_rand.max_force)
 
diff --git a/unilab/config/manipulation_params.py b/unilab/config/manipulation_params.py
index 8553b609a..d77be5af2 100644
--- a/unilab/config/manipulation_params.py
+++ b/unilab/config/manipulation_params.py
@@ -1,6 +1,5 @@
 from ml_collections import config_dict
 
-
 DEFAULT_ENV_NUM_BY_TASK: dict[str, int] = {
     "AllegroInhandRotation": 16384,
 }
diff --git a/unilab/envs/locomotion/g1/__init__.py b/unilab/envs/locomotion/g1/__init__.py
index 9d0ae877b..200f0fe75 100644
--- a/unilab/envs/locomotion/g1/__init__.py
+++ b/unilab/envs/locomotion/g1/__init__.py
@@ -1,2 +1,2 @@
-from .joystick import G1JoystickPPOCfg, G1JoystickPPO
+from .joystick import G1JoystickPPO, G1JoystickPPOCfg
 from .joystick_sac import G1JoystickSACCfg, G1WalkTaskMjSAC
diff --git a/unilab/envs/locomotion/g1/base.py b/unilab/envs/locomotion/g1/base.py
index ca442c931..fbefaa4a2 100644
--- a/unilab/envs/locomotion/g1/base.py
+++ b/unilab/envs/locomotion/g1/base.py
@@ -1,13 +1,14 @@
 from __future__ import annotations
 
+from dataclasses import dataclass, field
+
 import gymnasium as gym
 import mujoco
 import numpy as np
-from dataclasses import dataclass, field
 
+from unilab.base.backend import SimBackend
 from unilab.base.base import EnvCfg
 from unilab.base.np_env import NpEnv, NpEnvState
-from unilab.base.backend import SimBackend
 
 
 @dataclass
diff --git a/unilab/envs/locomotion/g1/joystick.py b/unilab/envs/locomotion/g1/joystick.py
index 83c8fafaf..94633d2c8 100644
--- a/unilab/envs/locomotion/g1/joystick.py
+++ b/unilab/envs/locomotion/g1/joystick.py
@@ -2,18 +2,19 @@
 
 from __future__ import annotations
 
+import math
 from dataclasses import dataclass, field
-from etils import epath
+
 import gymnasium as gym
-import math
 import numpy as np
+from etils import epath
 
 from unilab.base import registry
-from unilab.base.np_env import NpEnvState
 from unilab.base.backend import create_backend
-from unilab.utils.math_utils import np_quat_mul, np_yaw_to_quat
 from unilab.base.dtype_config import get_global_dtype
+from unilab.base.np_env import NpEnvState
 from unilab.envs.locomotion.g1.base import G1BaseCfg, G1BaseEnv
+from unilab.utils.math_utils import np_quat_mul, np_yaw_to_quat
 
 
 @dataclass
@@ -109,7 +110,7 @@ def __init__(self, cfg: G1JoystickPPOCfg, num_envs=1, backend_type="mujoco"):
         )
         self._pose_weights = np.array(cfg.reward_config.pose_weights, dtype=get_global_dtype())
         if self._pose_weights.shape[0] != self._num_action:
-            raise ValueError(f"pose_weights length mismatch")
+            raise ValueError("pose_weights length mismatch")
 
         self._init_obs_space()
         self._init_reward_functions()
diff --git a/unilab/envs/locomotion/g1/joystick_sac.py b/unilab/envs/locomotion/g1/joystick_sac.py
index deb530bcf..2105a5ca1 100644
--- a/unilab/envs/locomotion/g1/joystick_sac.py
+++ b/unilab/envs/locomotion/g1/joystick_sac.py
@@ -3,15 +3,16 @@
 from __future__ import annotations
 
 from dataclasses import dataclass, field
-from etils import epath
+
 import numpy as np
+from etils import epath
 
 from unilab.base import registry
 from unilab.base.backend import create_backend
+from unilab.base.curriculum import EpisodeLengthTracker, PenaltyCurriculum
 from unilab.base.dtype_config import get_global_dtype
 from unilab.envs.locomotion.g1.base import G1BaseCfg, G1BaseEnv
 from unilab.envs.locomotion.g1.joystick import G1JoystickPPO, InitState
-from unilab.base.curriculum import EpisodeLengthTracker, PenaltyCurriculum
 
 
 @dataclass
diff --git a/unilab/envs/locomotion/go1/__init__.py b/unilab/envs/locomotion/go1/__init__.py
index c437c3616..40667c153 100644
--- a/unilab/envs/locomotion/go1/__init__.py
+++ b/unilab/envs/locomotion/go1/__init__.py
@@ -1 +1 @@
-from .joystick import Go1WalkTask, Go1JoystickCfg
+from .joystick import Go1JoystickCfg, Go1WalkTask
diff --git a/unilab/envs/locomotion/go1/base.py b/unilab/envs/locomotion/go1/base.py
index a52afce0d..431905945 100644
--- a/unilab/envs/locomotion/go1/base.py
+++ b/unilab/envs/locomotion/go1/base.py
@@ -1,14 +1,15 @@
 from __future__ import annotations
 
+from dataclasses import dataclass, field
+
 import gymnasium as gym
 import mujoco
 import numpy as np
-from dataclasses import dataclass, field
 
-from unilab.base.base import EnvCfg
-from unilab.base.np_env import NpEnv, NpEnvState
 from unilab.base.backend import SimBackend
+from unilab.base.base import EnvCfg
 from unilab.base.dtype_config import get_global_dtype
+from unilab.base.np_env import NpEnv, NpEnvState
 
 
 @dataclass
@@ -122,7 +123,7 @@ def _init_buffers(self):
             )
 
     def push_robots(self):
-        if self.push_robots_flag == True:
+        if self.push_robots_flag:
             if self.step_counter % self._cfg.domain_rand.push_interval == 0:
                 self._backend.push_robots(self._cfg.domain_rand.max_force)
 
diff --git a/unilab/envs/locomotion/go1/joystick.py b/unilab/envs/locomotion/go1/joystick.py
index cca3e2a74..d845c3bda 100644
--- a/unilab/envs/locomotion/go1/joystick.py
+++ b/unilab/envs/locomotion/go1/joystick.py
@@ -1,17 +1,17 @@
 from __future__ import annotations
 
-from etils import epath
+from dataclasses import dataclass, field
+
 import gymnasium as gym
 import numpy as np
-from dataclasses import dataclass, field
+from etils import epath
 
 from unilab.base import registry
-from unilab.base.np_env import NpEnvState
 from unilab.base.backend import create_backend
-from unilab.utils.math_utils import np_quat_mul, np_yaw_to_quat
 from unilab.base.dtype_config import get_global_dtype
-
-from unilab.envs.locomotion.go1.base import Go1BaseEnv, Go1BaseCfg
+from unilab.base.np_env import NpEnvState
+from unilab.envs.locomotion.go1.base import Go1BaseCfg, Go1BaseEnv
+from unilab.utils.math_utils import np_quat_mul, np_yaw_to_quat
 
 
 @dataclass
@@ -116,7 +116,6 @@ def observation_space(self) -> gym.spaces.Box:
         return self._observation_space
 
     def update_state(self, state: NpEnvState) -> NpEnvState:
-        period = 0.6
         self.phase = np.fmod(self.phase + self._cfg.ctrl_dt * self.gait_frequency, 1.0)
         self.feet_phase[:, 0] = self.phase
         self.feet_phase[:, 3] = self.phase
diff --git a/unilab/envs/locomotion/go2/__init__.py b/unilab/envs/locomotion/go2/__init__.py
index da32159e6..d6f14dbb3 100644
--- a/unilab/envs/locomotion/go2/__init__.py
+++ b/unilab/envs/locomotion/go2/__init__.py
@@ -1 +1 @@
-from .joystick import Go2WalkTask, Go2JoystickCfg
+from .joystick import Go2JoystickCfg, Go2WalkTask
diff --git a/unilab/envs/locomotion/go2/base.py b/unilab/envs/locomotion/go2/base.py
index e1c8e7f07..5299f1026 100644
--- a/unilab/envs/locomotion/go2/base.py
+++ b/unilab/envs/locomotion/go2/base.py
@@ -1,13 +1,14 @@
 from __future__ import annotations
 
+from dataclasses import dataclass, field
+
 import gymnasium as gym
 import mujoco
 import numpy as np
-from dataclasses import dataclass, field
 
+from unilab.base.backend import SimBackend
 from unilab.base.base import EnvCfg
 from unilab.base.np_env import NpEnv, NpEnvState
-from unilab.base.backend import SimBackend
 
 
 @dataclass
@@ -67,7 +68,7 @@ def __init__(self, cfg: Go2BaseCfg, backend: SimBackend, num_envs=1):
         self.push_robots_flag = False
         if self._backend.backend_type == "motrix":
             self._backend._process_rigid_body_props(cfg)
-            if self._cfg.domain_rand.push_robots == True:
+            if self._cfg.domain_rand.push_robots:
                 self.push_robots_flag = True
 
     def _init_action_space(self):
diff --git a/unilab/envs/locomotion/go2/joystick.py b/unilab/envs/locomotion/go2/joystick.py
index 1d9111add..a5cad9c27 100644
--- a/unilab/envs/locomotion/go2/joystick.py
+++ b/unilab/envs/locomotion/go2/joystick.py
@@ -1,17 +1,17 @@
 from __future__ import annotations
 
-from etils import epath
+from dataclasses import dataclass, field
+
 import gymnasium as gym
 import numpy as np
-from dataclasses import dataclass, field
+from etils import epath
 
 from unilab.base import registry
-from unilab.base.np_env import NpEnvState
 from unilab.base.backend import create_backend
+from unilab.base.np_env import NpEnvState
+from unilab.envs.locomotion.go2.base import Go2BaseCfg, Go2BaseEnv
 from unilab.utils.math_utils import np_quat_mul, np_yaw_to_quat
 
-from unilab.envs.locomotion.go2.base import Go2BaseEnv, Go2BaseCfg
-
 
 @dataclass
 class InitState:
diff --git a/unilab/envs/locomotion/task_registry.py b/unilab/envs/locomotion/task_registry.py
index 95c30fb33..a0e5dd1dd 100644
--- a/unilab/envs/locomotion/task_registry.py
+++ b/unilab/envs/locomotion/task_registry.py
@@ -1,4 +1,4 @@
-from typing import Dict, Tuple, Any
+from typing import Any, Dict, Tuple
 
 
 class TaskRegistry:
diff --git a/unilab/envs/manipulation/inhand_rot_allegro/base.py b/unilab/envs/manipulation/inhand_rot_allegro/base.py
index 21e083be8..22c18786d 100644
--- a/unilab/envs/manipulation/inhand_rot_allegro/base.py
+++ b/unilab/envs/manipulation/inhand_rot_allegro/base.py
@@ -1,14 +1,15 @@
 from __future__ import annotations
 
+from dataclasses import dataclass, field
+
 import gymnasium as gym
 import mujoco
 import numpy as np
-from dataclasses import dataclass, field
 
-from unilab.base.base import EnvCfg
-from unilab.base.np_env import NpEnv, NpEnvState
 from unilab.base.backend import SimBackend
+from unilab.base.base import EnvCfg
 from unilab.base.dtype_config import get_global_dtype
+from unilab.base.np_env import NpEnv, NpEnvState
 
 # ─────────────────────────── Configuration ────────────────────────────
 
diff --git a/unilab/envs/manipulation/inhand_rot_allegro/gen_grasp.py b/unilab/envs/manipulation/inhand_rot_allegro/gen_grasp.py
index a84f47f19..74e6966de 100644
--- a/unilab/envs/manipulation/inhand_rot_allegro/gen_grasp.py
+++ b/unilab/envs/manipulation/inhand_rot_allegro/gen_grasp.py
@@ -21,10 +21,10 @@
 import time
 from pathlib import Path
 
+import mediapy as media
 import mujoco
 import mujoco.viewer
 import numpy as np
-import mediapy as media
 
 # ── Path setup ──────────────────────────────────────────────────────────────
 # gen_grasp.py lives at: UniLab/unilab/envs/manipulation/inhand_rot_allegro/
@@ -52,12 +52,12 @@ def ensure_registries():
 ensure_registries()
 
 from unilab.base import registry  # noqa: E402  (after sys.path setup)
-from unilab.utils import render_many  # noqa: E402
 from unilab.base.dtype_config import get_global_dtype  # noqa: E402
 
 # Explicit import to guarantee the @registry.env decorator runs,
 # since ensure_registries() silently swallows import errors.
 from unilab.envs.manipulation.inhand_rot_allegro import rotation as _rotation_register
+from unilab.utils import render_many  # noqa: E402
 
 _ = _rotation_register  # side-effect import: triggers @registry.env decorator
 
diff --git a/unilab/envs/manipulation/inhand_rot_allegro/rotation.py b/unilab/envs/manipulation/inhand_rot_allegro/rotation.py
index 29c6b20ec..62ed09847 100644
--- a/unilab/envs/manipulation/inhand_rot_allegro/rotation.py
+++ b/unilab/envs/manipulation/inhand_rot_allegro/rotation.py
@@ -6,7 +6,6 @@
 import numpy as np
 from etils import epath
 
-
 # ─────────────────────────── Quaternion helpers ───────────────────────────
 # All quaternions are w-first: [w, x, y, z], shape (N, 4)
 
@@ -54,8 +53,8 @@ def _quat_to_axis_angle(q: np.ndarray) -> np.ndarray:
 
 
 from unilab.base import registry
-from unilab.base.np_env import NpEnvState
 from unilab.base.backend import create_backend
+from unilab.base.np_env import NpEnvState
 from unilab.envs.manipulation.inhand_rot_allegro.base import AllegroBaseCfg, AllegroBaseMjEnv
 
 # ─────────────────────────── Configuration ────────────────────────────
diff --git a/unilab/ipc/__init__.py b/unilab/ipc/__init__.py
index 3501bc265..28d3eab91 100644
--- a/unilab/ipc/__init__.py
+++ b/unilab/ipc/__init__.py
@@ -1,10 +1,10 @@
 """IPC primitives for multi-process RL training."""
 
-from unilab.ipc.shared_storage import SharedOnPolicyStorage
-from unilab.ipc.weight_sync import SharedWeightSync
 from unilab.ipc.async_runner import AsyncRunner
-from unilab.ipc.shared_obs_stats import SharedObsNormStats
 from unilab.ipc.replay_buffer import ReplayBuffer
+from unilab.ipc.shared_obs_stats import SharedObsNormStats
+from unilab.ipc.shared_storage import SharedOnPolicyStorage
+from unilab.ipc.weight_sync import SharedWeightSync
 
 __all__ = [
     "SharedOnPolicyStorage",
diff --git a/unilab/ipc/async_runner.py b/unilab/ipc/async_runner.py
index 751463224..28e050c2c 100644
--- a/unilab/ipc/async_runner.py
+++ b/unilab/ipc/async_runner.py
@@ -1,6 +1,7 @@
 """Base async runner for multi-process RL training."""
 
 from __future__ import annotations
+
 import multiprocessing as mp
 from abc import ABC, abstractmethod
 from typing import Any, Callable
diff --git a/unilab/ipc/replay_buffer.py b/unilab/ipc/replay_buffer.py
index 2898af7fb..1321913fb 100644
--- a/unilab/ipc/replay_buffer.py
+++ b/unilab/ipc/replay_buffer.py
@@ -1,8 +1,9 @@
 """Minimal zero-copy replay buffer for off-policy RL."""
 
-import torch
 from typing import Dict
 
+import torch
+
 
 class ReplayBuffer:
     """Shared tensor replay buffer with device-adaptive sampling.
diff --git a/unilab/ipc/shared_storage.py b/unilab/ipc/shared_storage.py
index b7ed83cc9..f25e29af9 100644
--- a/unilab/ipc/shared_storage.py
+++ b/unilab/ipc/shared_storage.py
@@ -1,9 +1,11 @@
 """Shared on-policy storage for APPO-style algorithms."""
 
 from __future__ import annotations
+
 import multiprocessing as mp
 from multiprocessing import shared_memory
 from typing import Dict
+
 import numpy as np
 
 _SPAWN_CTX = mp.get_context("spawn")
diff --git a/unilab/ipc/weight_sync.py b/unilab/ipc/weight_sync.py
index d9c477025..9d692e3f4 100644
--- a/unilab/ipc/weight_sync.py
+++ b/unilab/ipc/weight_sync.py
@@ -1,9 +1,11 @@
 """Shared weight synchronization for actor networks."""
 
 from __future__ import annotations
+
 import multiprocessing as mp
 from multiprocessing import shared_memory
 from typing import Dict
+
 import numpy as np
 
 _SPAWN_CTX = mp.get_context("spawn")
diff --git a/unilab/utils/__init__.py b/unilab/utils/__init__.py
index d75b82cd0..5954b6b21 100644
--- a/unilab/utils/__init__.py
+++ b/unilab/utils/__init__.py
@@ -1,8 +1,8 @@
 # Utility modules for UniLab
-from unilab.utils.torch_utils import to_torch, to_numpy
+from unilab.utils.algo_utils import build_actor, ensure_registries
 from unilab.utils.offpolicy_logger import OffPolicyLogger
 from unilab.utils.onpolicy_logger import OnPolicyLogger
-from unilab.utils.algo_utils import ensure_registries, build_actor
+from unilab.utils.torch_utils import to_numpy, to_torch
 
 __all__ = [
     "to_torch",
diff --git a/unilab/utils/algo_utils.py b/unilab/utils/algo_utils.py
index cdf2fb471..d640102df 100644
--- a/unilab/utils/algo_utils.py
+++ b/unilab/utils/algo_utils.py
@@ -1,7 +1,7 @@
 """Common utilities for RL algorithms."""
 
-import pkgutil
 import importlib
+import pkgutil
 
 
 def ensure_registries():
diff --git a/unilab/utils/math_utils.py b/unilab/utils/math_utils.py
index c8309aeeb..6f0b15a34 100644
--- a/unilab/utils/math_utils.py
+++ b/unilab/utils/math_utils.py
@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 import importlib
+
 import numpy as np
 
 
diff --git a/unilab/utils/offpolicy_logger.py b/unilab/utils/offpolicy_logger.py
index 29fd9ceb6..f774885b7 100644
--- a/unilab/utils/offpolicy_logger.py
+++ b/unilab/utils/offpolicy_logger.py
@@ -31,15 +31,15 @@
 
 from __future__ import annotations
 
-import os
 import time
 from collections import deque
+
+from rich import box
 from rich.console import Console, Group
 from rich.live import Live
 from rich.panel import Panel
 from rich.table import Table
 from rich.text import Text
-from rich import box
 
 
 def _fmt_time(seconds: float) -> str:
@@ -414,16 +414,16 @@ def _build_display(self) -> Panel:
         eta = self._estimate_eta()
         header_text = Text()
         header_text.append(f" {self.algo_name}", style="bold cyan")
-        header_text.append(f"  │  ", style="dim")
+        header_text.append("  │  ", style="dim")
         header_text.append(f"{self.env_name}", style="bold white")
-        header_text.append(f"  │  ", style="dim")
+        header_text.append("  │  ", style="dim")
         header_text.append(f"iter {self._iteration}/{self.max_iterations}", style="yellow")
-        header_text.append(f"  │  ", style="dim")
+        header_text.append("  │  ", style="dim")
         header_text.append(f"⏱ {_fmt_time(elapsed)}", style="green")
         if eta:
-            header_text.append(f"  │  ETA ", style="dim")
+            header_text.append("  │  ETA ", style="dim")
             header_text.append(eta, style="bold magenta")
-        header_text.append(f"  │  ", style="dim")
+        header_text.append("  │  ", style="dim")
         header_text.append(self._status, style="dim italic")
 
         header_panel = Panel(header_text, style="dim", box=box.SIMPLE)
@@ -442,7 +442,7 @@ def _build_display(self) -> Panel:
 
         return Panel(
             main_group,
-            title=f"[bold] 🚀 UniLab Off-Policy Training [/]",
+            title="[bold] 🚀 UniLab Off-Policy Training [/]",
             border_style="bright_blue",
             padding=(0, 1),
         )
diff --git a/unilab/utils/onpolicy_logger.py b/unilab/utils/onpolicy_logger.py
index b64131a7d..05782c99f 100644
--- a/unilab/utils/onpolicy_logger.py
+++ b/unilab/utils/onpolicy_logger.py
@@ -5,12 +5,13 @@
 import os
 import time
 from collections import deque
+
+from rich import box
 from rich.console import Console, Group
 from rich.live import Live
 from rich.panel import Panel
 from rich.table import Table
 from rich.text import Text
-from rich import box
 
 
 def _fmt_time(seconds: float) -> str:
@@ -231,14 +232,14 @@ def _build_display(self) -> Panel:
 
         header_text = Text()
         header_text.append(f" {self.algo_name}", style="bold cyan")
-        header_text.append(f"  │  ", style="dim")
+        header_text.append("  │  ", style="dim")
         header_text.append(f"{self.env_name}", style="bold white")
-        header_text.append(f"  │  ", style="dim")
+        header_text.append("  │  ", style="dim")
         header_text.append(f"iter {self._iteration}/{self.max_iterations}", style="yellow")
-        header_text.append(f"  │  ", style="dim")
+        header_text.append("  │  ", style="dim")
         header_text.append(f"⏱ {_fmt_time(elapsed)}", style="green")
         if eta:
-            header_text.append(f"  │  ETA ", style="dim")
+            header_text.append("  │  ETA ", style="dim")
             header_text.append(eta, style="bold magenta")
 
         header_panel = Panel(header_text, style="dim", box=box.SIMPLE)
@@ -256,7 +257,7 @@ def _build_display(self) -> Panel:
 
         return Panel(
             main_group,
-            title=f"[bold] 🚀 UniLab On-Policy Training [/]",
+            title="[bold] 🚀 UniLab On-Policy Training [/]",
             border_style="bright_blue",
             padding=(0, 1),
         )
diff --git a/unilab/utils/render_many.py b/unilab/utils/render_many.py
index 8f8037ce3..7154f814f 100644
--- a/unilab/utils/render_many.py
+++ b/unilab/utils/render_many.py
@@ -1,9 +1,9 @@
-import os
 import math
-import numpy as np
-import mujoco
+import os
+
 import imageio
-from multiprocessing import Pool
+import mujoco
+import numpy as np
 
 # Use EGL for off-screen rendering on headless servers (no X11/display required)
 os.environ.setdefault("MUJOCO_GL", "egl")
diff --git a/unilab/utils/rsl_rl_compat.py b/unilab/utils/rsl_rl_compat.py
index 658bd9ecf..b27295d6c 100644
--- a/unilab/utils/rsl_rl_compat.py
+++ b/unilab/utils/rsl_rl_compat.py
@@ -13,6 +13,7 @@
 
 import importlib.metadata
 from copy import deepcopy
+
 from packaging.version import Version
 
 
diff --git a/uv.lock b/uv.lock
index 8f4aee8d5..9f0b0c0d4 100644
--- a/uv.lock
+++ b/uv.lock
@@ -2,10 +2,14 @@ version = 1
 revision = 3
 requires-python = ">=3.10, <3.14"
 resolution-markers = [
-    "python_full_version >= '3.13'",
-    "python_full_version == '3.12.*'",
-    "python_full_version == '3.11.*'",
-    "python_full_version < '3.11'",
+    "python_full_version >= '3.13' and sys_platform == 'linux'",
+    "python_full_version == '3.12.*' and sys_platform == 'linux'",
+    "python_full_version == '3.11.*' and sys_platform == 'linux'",
+    "python_full_version < '3.11' and sys_platform == 'linux'",
+    "python_full_version >= '3.13' and sys_platform != 'linux'",
+    "python_full_version == '3.12.*' and sys_platform != 'linux'",
+    "python_full_version == '3.11.*' and sys_platform != 'linux'",
+    "python_full_version < '3.11' and sys_platform != 'linux'",
 ]
 
 [[package]]
@@ -49,7 +53,8 @@ name = "contourpy"
 version = "1.3.2"
 source = { registry = "https://pypi.org/simple" }
 resolution-markers = [
-    "python_full_version < '3.11'",
+    "python_full_version < '3.11' and sys_platform == 'linux'",
+    "python_full_version < '3.11' and sys_platform != 'linux'",
 ]
 dependencies = [
     { name = "numpy", version = "2.2.6", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version < '3.11'" },
@@ -119,9 +124,12 @@ name = "contourpy"
 version = "1.3.3"
 source = { registry = "https://pypi.org/simple" }
 resolution-markers = [
-    "python_full_version >= '3.13'",
-    "python_full_version == '3.12.*'",
-    "python_full_version == '3.11.*'",
+    "python_full_version >= '3.13' and sys_platform == 'linux'",
+    "python_full_version == '3.12.*' and sys_platform == 'linux'",
+    "python_full_version == '3.11.*' and sys_platform == 'linux'",
+    "python_full_version >= '3.13' and sys_platform != 'linux'",
+    "python_full_version == '3.12.*' and sys_platform != 'linux'",
+    "python_full_version == '3.11.*' and sys_platform != 'linux'",
 ]
 dependencies = [
     { name = "numpy", version = "2.4.3", source = { registry = "https://pypi.org/simple" }, marker = "python_full_version >= '3.11'" },
@@ -184,7 +192,7 @@ name = "cuda-bindings"
 version = "12.9.4"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
-    { name = "cuda-pathfinder" },
+    { name = "cuda-pathfinder", marker = "sys_platform == 'linux'" },
 ]
 wheels = [
     { url = "https://files.pythonhosted.org/packages/7a/d8/b546104b8da3f562c1ff8ab36d130c8fe1dd6a045ced80b4f6ad74f7d4e1/cuda_bindings-12.9.4-cp310-cp310-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:4d3c842c2a4303b2a580fe955018e31aea30278be19795ae05226235268032e5", size = 12148218, upload-time = "2025-10-21T14:51:28.855Z" },
@@ -225,7 +233,8 @@ name = "etils"
 version = "1.13.0"
 source = { registry = "https://pypi.org/simple" }
 resolution-markers = [
-    "python_full_version < '3.11'",
+    "python_full_version < '3.11' and sys_platform == 'linux'",
+    "python_full_version < '3.11' and sys_platform != 'linux'",
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/9b/a0/522bbff0f3cdd37968f90dd7f26c7aa801ed87f5ba335f156de7f2b88a48/etils-1.13.0.tar.gz", hash = "sha256:a5b60c71f95bcd2d43d4e9fb3dc3879120c1f60472bb5ce19f7a860b1d44f607", size = 106368, upload-time = "2025-07-15T10:29:10.563Z" }
 wheels = [
@@ -245,9 +254,12 @@ name = "etils"
 version = "1.14.0"
 source = { registry = "https://pypi.org/simple" }
 resolution-markers = [
-    "python_full_version >= '3.13'",
-    "python_full_version == '3.12.*'",
-    "python_full_version == '3.11.*'",
+    "python_full_version >= '3.13' and sys_platform == 'linux'",
+    "python_full_version == '3.12.*' and sys_platform == 'linux'",
+    "python_full_version == '3.11.*' and sys_platform == 'linux'",
+    "python_full_version >= '3.13' and sys_platform != 'linux'",
+    "python_full_version == '3.12.*' and sys_platform != 'linux'",
+    "python_full_version == '3.11.*' and sys_platform != 'linux'",
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/26/ce/6e067242fde898841922ac6fc82b0bb2fe35c38e995880bdffdfbe30182a/etils-1.14.0.tar.gz", hash = "sha256:8136e7f4c4173cd0af0ca5481c4475152f0b8686192951eefa60ee8711e1ede4", size = 108127, upload-time = "2026-03-04T17:41:36.291Z" }
 wheels = [
@@ -522,7 +534,8 @@ name = "ipython"
 version = "8.38.0"
 source = { registry = "https://pypi.org/simple" }
 resolution-markers = [
-    "python_full_version < '3.11'",
+    "python_full_version < '3.11' and sys_platform == 'linux'",
+    "python_full_version < '3.11' and sys_platform != 'linux'",
 ]
 dependencies = [
     { name = "colorama", marker = "python_full_version < '3.11' and sys_platform == 'win32'" },
@@ -547,7 +560,8 @@ name = "ipython"
 version = "9.10.0"
 source = { registry = "https://pypi.org/simple" }
 resolution-markers = [
-    "python_full_version == '3.11.*'",
+    "python_full_version == '3.11.*' and sys_platform == 'linux'",
+    "python_full_version == '3.11.*' and sys_platform != 'linux'",
 ]
 dependencies = [
     { name = "colorama", marker = "python_full_version == '3.11.*' and sys_platform == 'win32'" },
@@ -572,8 +586,10 @@ name = "ipython"
 version = "9.11.0"
 source = { registry = "https://pypi.org/simple" }
 resolution-markers = [
-    "python_full_version >= '3.13'",
-    "python_full_version == '3.12.*'",
+    "python_full_version >= '3.13' and sys_platform == 'linux'",
+    "python_full_version == '3.12.*' and sys_platform == 'linux'",
+    "python_full_version >= '3.13' and sys_platform != 'linux'",
+    "python_full_version == '3.12.*' and sys_platform != 'linux'",
 ]
 dependencies = [
     { name = "colorama", marker = "python_full_version >= '3.12' and sys_platform == 'win32'" },
@@ -1156,7 +1172,8 @@ name = "networkx"
 version = "3.4.2"
 source = { registry = "https://pypi.org/simple" }
 resolution-markers = [
-    "python_full_version < '3.11'",
+    "python_full_version < '3.11' and sys_platform == 'linux'",
+    "python_full_version < '3.11' and sys_platform != 'linux'",
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/fd/1d/06475e1cd5264c0b870ea2cc6fdb3e37177c1e565c43f56ff17a10e3937f/networkx-3.4.2.tar.gz", hash = "sha256:307c3669428c5362aab27c8a1260aa8f47c4e91d3891f48be0141738d8d053e1", size = 2151368, upload-time = "2024-10-21T12:39:38.695Z" }
 wheels = [
@@ -1168,9 +1185,12 @@ name = "networkx"
 version = "3.6.1"
 source = { registry = "https://pypi.org/simple" }
 resolution-markers = [
-    "python_full_version >= '3.13'",
-    "python_full_version == '3.12.*'",
-    "python_full_version == '3.11.*'",
+    "python_full_version >= '3.13' and sys_platform == 'linux'",
+    "python_full_version == '3.12.*' and sys_platform == 'linux'",
+    "python_full_version == '3.11.*' and sys_platform == 'linux'",
+    "python_full_version >= '3.13' and sys_platform != 'linux'",
+    "python_full_version == '3.12.*' and sys_platform != 'linux'",
+    "python_full_version == '3.11.*' and sys_platform != 'linux'",
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/6a/51/63fe664f3908c97be9d2e4f1158eb633317598cfa6e1fc14af5383f17512/networkx-3.6.1.tar.gz", hash = "sha256:26b7c357accc0c8cde558ad486283728b65b6a95d85ee1cd66bafab4c8168509", size = 2517025, upload-time = "2025-12-08T17:02:39.908Z" }
 wheels = [
@@ -1182,7 +1202,8 @@ name = "numpy"
 version = "2.2.6"
 source = { registry = "https://pypi.org/simple" }
 resolution-markers = [
-    "python_full_version < '3.11'",
+    "python_full_version < '3.11' and sys_platform == 'linux'",
+    "python_full_version < '3.11' and sys_platform != 'linux'",
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/76/21/7d2a95e4bba9dc13d043ee156a356c0a8f0c6309dff6b21b4d71a073b8a8/numpy-2.2.6.tar.gz", hash = "sha256:e29554e2bef54a90aa5cc07da6ce955accb83f21ab5de01a62c8478897b264fd", size = 20276440, upload-time = "2025-05-17T22:38:04.611Z" }
 wheels = [
@@ -1247,9 +1268,12 @@ name = "numpy"
 version = "2.4.3"
 source = { registry = "https://pypi.org/simple" }
 resolution-markers = [
-    "python_full_version >= '3.13'",
-    "python_full_version == '3.12.*'",
-    "python_full_version == '3.11.*'",
+    "python_full_version >= '3.13' and sys_platform == 'linux'",
+    "python_full_version == '3.12.*' and sys_platform == 'linux'",
+    "python_full_version == '3.11.*' and sys_platform == 'linux'",
+    "python_full_version >= '3.13' and sys_platform != 'linux'",
+    "python_full_version == '3.12.*' and sys_platform != 'linux'",
+    "python_full_version == '3.11.*' and sys_platform != 'linux'",
 ]
 sdist = { url = "https://files.pythonhosted.org/packages/10/8b/c265f4823726ab832de836cdd184d0986dcf94480f81e8739692a7ac7af2/numpy-2.4.3.tar.gz", hash = "sha256:483a201202b73495f00dbc83796c6ae63137a9bdade074f7648b3e32613412dd", size = 20727743, upload-time = "2026-03-09T07:58:53.426Z" }
 wheels = [
@@ -1342,7 +1366,7 @@ name = "nvidia-cudnn-cu12"
 version = "9.10.2.21"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
-    { name = "nvidia-cublas-cu12" },
+    { name = "nvidia-cublas-cu12", marker = "sys_platform == 'linux'" },
 ]
 wheels = [
     { url = "https://files.pythonhosted.org/packages/ba/51/e123d997aa098c61d029f76663dedbfb9bc8dcf8c60cbd6adbe42f76d049/nvidia_cudnn_cu12-9.10.2.21-py3-none-manylinux_2_27_x86_64.whl", hash = "sha256:949452be657fa16687d0930933f032835951ef0892b37d2d53824d1a84dc97a8", size = 706758467, upload-time = "2025-06-06T21:54:08.597Z" },
@@ -1353,7 +1377,7 @@ name = "nvidia-cufft-cu12"
 version = "11.3.3.83"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
-    { name = "nvidia-nvjitlink-cu12" },
+    { name = "nvidia-nvjitlink-cu12", marker = "sys_platform == 'linux'" },
 ]
 wheels = [
     { url = "https://files.pythonhosted.org/packages/1f/13/ee4e00f30e676b66ae65b4f08cb5bcbb8392c03f54f2d5413ea99a5d1c80/nvidia_cufft_cu12-11.3.3.83-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:4d2dd21ec0b88cf61b62e6b43564355e5222e4a3fb394cac0db101f2dd0d4f74", size = 193118695, upload-time = "2025-03-07T01:45:27.821Z" },
@@ -1380,9 +1404,9 @@ name = "nvidia-cusolver-cu12"
 version = "11.7.3.90"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
-    { name = "nvidia-cublas-cu12" },
-    { name = "nvidia-cusparse-cu12" },
-    { name = "nvidia-nvjitlink-cu12" },
+    { name = "nvidia-cublas-cu12", marker = "sys_platform == 'linux'" },
+    { name = "nvidia-cusparse-cu12", marker = "sys_platform == 'linux'" },
+    { name = "nvidia-nvjitlink-cu12", marker = "sys_platform == 'linux'" },
 ]
 wheels = [
     { url = "https://files.pythonhosted.org/packages/85/48/9a13d2975803e8cf2777d5ed57b87a0b6ca2cc795f9a4f59796a910bfb80/nvidia_cusolver_cu12-11.7.3.90-py3-none-manylinux_2_27_x86_64.whl", hash = "sha256:4376c11ad263152bd50ea295c05370360776f8c3427b30991df774f9fb26c450", size = 267506905, upload-time = "2025-03-07T01:47:16.273Z" },
@@ -1393,7 +1417,7 @@ name = "nvidia-cusparse-cu12"
 version = "12.5.8.93"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
-    { name = "nvidia-nvjitlink-cu12" },
+    { name = "nvidia-nvjitlink-cu12", marker = "sys_platform == 'linux'" },
 ]
 wheels = [
     { url = "https://files.pythonhosted.org/packages/c2/f5/e1854cb2f2bcd4280c44736c93550cc300ff4b8c95ebe370d0aa7d2b473d/nvidia_cusparse_cu12-12.5.8.93-py3-none-manylinux2014_x86_64.manylinux_2_17_x86_64.whl", hash = "sha256:1ec05d76bbbd8b61b06a80e1eaf8cf4959c3d4ce8e711b65ebd0443bb0ebb13b", size = 288216466, upload-time = "2025-03-07T01:48:13.779Z" },
@@ -1870,7 +1894,7 @@ wheels = [
 
 [[package]]
 name = "rsl-rl-lib"
-version = "4.0.1"
+version = "5.0.1"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "gitpython" },
@@ -1882,9 +1906,9 @@ dependencies = [
     { name = "torch" },
     { name = "torchvision" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/aa/5b/5bea7c110c389f9336ebe5ad7cd9abbcbbb64d7de0d0f50c2af8fa2ce321/rsl_rl_lib-4.0.1.tar.gz", hash = "sha256:76a194e4834946716cc0e44dcd2be8b2254ac969c2452b145cd268c9cc908bf7", size = 41069, upload-time = "2026-02-09T09:15:10.927Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/d5/23/dae404688e571e73662b7754ab2a8bddcc48f625c338b7f55576ad528653/rsl_rl_lib-5.0.1.tar.gz", hash = "sha256:b6e1fce8f4481c6118d53c7b03b80c8070d0a1edd3df3efbec5e5e6ff7c92132", size = 58623, upload-time = "2026-03-04T08:19:49.671Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/31/db/8fa8832e6e02e5c4dc8321eb1596d5968b86a214fca03af361c94b8b045f/rsl_rl_lib-4.0.1-py3-none-any.whl", hash = "sha256:e840c8587acaad4ba3ab34fc7b07a203b34dd44a831ea9dc5de1769bfc14b148", size = 53803, upload-time = "2026-02-09T09:15:09.705Z" },
+    { url = "https://files.pythonhosted.org/packages/ad/25/293a8ffd929d40accab30057b962084601133bd8a6779c156c2f58eb93bc/rsl_rl_lib-5.0.1-py3-none-any.whl", hash = "sha256:2f71a4f753537faf7826d0b74815160e37c409b39ab80a4581524b7ef7bd8539", size = 84202, upload-time = "2026-03-04T08:19:48.044Z" },
 ]
 
 [[package]]
@@ -2251,7 +2275,7 @@ requires-dist = [
     { name = "packaging" },
     { name = "pytest", marker = "extra == 'dev'" },
     { name = "rich" },
-    { name = "rsl-rl-lib", specifier = ">=3.3.0,<5.0.0" },
+    { name = "rsl-rl-lib", specifier = ">=5.0.0" },
     { name = "ruff", marker = "extra == 'dev'" },
     { name = "tensorboard" },
     { name = "tensordict" },