chore(deps): bump unisim-core to 1.4.2 and unilab-rl to 1.2.1

.gitignore

@@ -83,9 +83,7 @@ src/unilab/assets/bots/
 src/unilab/assets/robots/x2/meshes/*.STL
 src/unilab/assets/robots/g1/assets/
 src/unilab/assets/robots/g1/textures/
-src/unilab/assets/robots/go1/assets/
 src/unilab/assets/robots/go2/assets/
-src/unilab/assets/robots/a2/assets/
 src/unilab/assets/robots/allegro_hand/assets/
 src/unilab/assets/robots/sharpa_wave/meshes/
 

docs/sphinx/source/api_reference/index.md

@@ -56,7 +56,7 @@ PPO / APPO / SAC / TD3 variants moved to the uni_rl package (issue #1480).
 :::{grid-item-card} 🏋 `unilab.training`
 :link: training/index
 :link-type: doc
-Runtime helpers, monitoring, reward bookkeeping, runner orchestration.
+Runtime helpers, reward bookkeeping, runner orchestration.
 :::
 
 :::{grid-item-card} 🔗 Shared-memory runtime → `uni_rl.ipc`

docs/sphinx/source/api_reference/training/index.md

@@ -2,7 +2,7 @@
 
 Glue between `algos`, `envs` and `ipc`: experiment lifecycle and the
 top-level `run` helpers invoked by the `train` / `eval` / `demo` CLI
-entrypoints. Layer-0 helpers (seeding, monitoring, reward bookkeeping,
+entrypoints. Layer-0 helpers (seeding, reward bookkeeping,
 checkpoint resolution, sim2sim contracts) live in `unilab.utils`; resolved
 env config adaptation lives in `unilab.base.config_adapter`.
 

docs/sphinx/source/api_reference/utils/index.md

@@ -1,7 +1,7 @@
 # `unilab.utils` — Utilities
 
-Device probing, tensor helpers, training seeding, hardware monitoring,
-reward bookkeeping, checkpoint resolution, sim2sim contract checks, NaN
+Device probing, tensor helpers, training seeding, reward bookkeeping,
+checkpoint resolution, sim2sim contract checks, NaN
 guards, and pure-numpy geometry/rotation helpers shared across envs and
 scripts.
 

docs/sphinx/source/en/2-user_guide/3-backends/7-newton.md

@@ -33,8 +33,10 @@ collector sim processes land on their own physical GPU with no cross-GPU
 leakage; single-GPU PPO/SAC regressions pass alongside. Newton/Warp follows
 standard CUDA device semantics, so no `CUDA_VISIBLE_DEVICES` pinning (the
 Genesis quirk) is needed; the rank-local device reaches spawn collectors as
-a `newton_device="cuda:N"` env override (uni_rl 1.0.0's collector-side
-process-binding gate only covers mjwarp), and the SAC owner raises the
+a `newton_device="cuda:N"` env override, and uni_rl's collector-side
+process binding is injection-based — UniLab injects
+`bind_backend_process_device_for_backend`, which covers both mjwarp and
+newton — while the SAC owner raises the
 collector tick-0 timeout to 180 s to cover Warp kernel compilation on the
 cold path.
 

docs/sphinx/source/en/2-user_guide/7-tooling/0-index.md

@@ -24,12 +24,6 @@ Configure run logging and experiment metadata.
 Inspect NaN guard dumps from PPO runs.
 :::
 
-:::{grid-item-card} Scene export
-:link: 4-scene_export
-:link-type: doc
-Export MuJoCo scenes and copied assets for inspection.
-:::
-
 :::{grid-item-card} Robot import
 :link: 5-robot_import
 :link-type: doc
@@ -44,6 +38,5 @@ Connect robot assets, model descriptions, control interfaces, and keyframes.
 1-onnx_export
 2-wandb
 3-nan_visualizer
-4-scene_export
 5-robot_import
 ```

docs/sphinx/source/en/3-deployment/1-sim_to_real/9-troubleshooting.md

@@ -24,8 +24,8 @@ sideways, start here.
 Almost always one of:
 
 1. **Joint order swapped.** Inspect `policy.onnx` input width and the joint
-   order in your motor driver. Use `unilab-export-scene` to dump the
-   training joint order.
+   order in your motor driver against the training scene XML
+   (`src/unilab/assets/robots/<robot>/scene_flat.xml`).
 2. **Action scale unit mismatch.** Policy outputs unscaled values; the
    driver expects rad, but you fed it normalized [-1, 1]. Apply the
    `env.actions.joint_pos.scale` / default-angle convention from the training

docs/sphinx/source/en/3-deployment/3-framework_migration/2-from_legged_gym.md

@@ -36,7 +36,7 @@ mostly mechanical.
 3. Mirror your reward terms; keep the same names so reward parity is
    diff-able.
 4. Translate command sampling — configure `UniformVelocityCommandCfg` under
-   the owner YAML's `env.commands` (see the Go1 flat owner).
+   the owner YAML's `env.commands` (see the `go2_joystick_flat` owner).
 5. Translate terrain — Legged Gym's heightfield generator has a UniLab
    counterpart at `unilab.terrains.heightfield_terrains`.
 

docs/sphinx/source/en/4-developer_guide/3-extending/4-new_terrain.md

@@ -25,8 +25,6 @@ materialization out of `step()`, `reset()`, and hot domain-randomization loops.
 - Terrain generator shape and numerical behavior:
   `tests/terrains/test_terrain_generator.py`
 - Rough locomotion height-scan and spawn behavior:
-  `tests/envs/locomotion/test_go2_rough_height_scan.py`,
-  `tests/envs/locomotion/test_go2_terrain_spawn.py`,
   `tests/envs/locomotion/test_terrain_spawn.py`
 - Backend materialization boundaries: `tests/utils/test_xml_utils.py`
 

docs/sphinx/source/zh_CN/2-user_guide/3-backends/7-newton.md

@@ -31,9 +31,10 @@ DpRankSupervisor DP=2（`training.devices=[0,1]`）训练冒烟均正常完成
 在各自 GPU、无对端泄漏；单卡 PPO/SAC 回归同步通过。Newton/Warp 遵循标准
 CUDA 设备语义，无需 Genesis 那样的 `CUDA_VISIBLE_DEVICES` 钉卡；rank
 本地设备通过 env override 以 `newton_device="cuda:N"` 传入 spawn
-collector（uni_rl 1.0.0 的 collector 进程绑定门只覆盖 mjwarp），SAC
-owner 将 collector tick-0 超时提升到 180 s 以覆盖 Warp 内核编译的冷
-路径。
+collector，且 uni_rl 的 collector 进程绑定为注入式——UniLab 注入的
+`bind_backend_process_device_for_backend` 同时覆盖 mjwarp 与
+newton；SAC owner 将 collector tick-0 超时提升到 180 s 以覆盖 Warp
+内核编译的冷路径。
 
 ## 安装
 

docs/sphinx/source/zh_CN/2-user_guide/7-tooling/0-index.md

@@ -23,12 +23,6 @@
 检查 PPO 运行中的 NaN guard dump。
 :::
 
-:::{grid-item-card} 场景导出
-:link: 4-scene_export
-:link-type: doc
-导出 MuJoCo 场景及复制的 asset 以供检查。
-:::
-
 :::{grid-item-card} 机器人导入
 :link: 5-robot_import
 :link-type: doc
@@ -43,6 +37,5 @@
 1-onnx_export
 2-wandb
 3-nan_visualizer
-4-scene_export
 5-robot_import
 ```

docs/sphinx/source/zh_CN/3-deployment/1-sim_to_real/9-troubleshooting.md

@@ -22,8 +22,8 @@
 
 几乎总是以下之一：
 
-1. **关节顺序被调换。** 检查 `policy.onnx` 的输入宽度与你电机驱动器中的关节顺序。
-   用 `unilab-export-scene` 导出训练时的关节顺序。
+1. **关节顺序被调换。** 检查 `policy.onnx` 的输入宽度，并将你电机驱动器中的关节顺序
+   与训练场景 XML（`src/unilab/assets/robots/<robot>/scene_flat.xml`）对照。
 2. **动作缩放单位不匹配。** 策略输出未缩放的值；驱动器期望的是弧度，而你喂给它的
    是归一化的 [-1, 1]。在把目标发送给驱动器之前，应用训练 owner YAML 中的
    `env.actions.joint_pos.scale` / 默认角度约定，并原样复现该 owner 解析后的

docs/sphinx/source/zh_CN/3-deployment/3-framework_migration/2-from_legged_gym.md

@@ -32,7 +32,7 @@ Legged Gym 曾是那套 GPU 常驻的 PPO 模板，教会了整个领域如何
 2. 在 `src/unilab/tasks/locomotion/<robot>/` 下创建一个任务模块。
 3. 镜像你的 reward 项；保持名称相同，以便 reward 一致性可被 diff。
 4. 翻译命令采样 —— 在 owner YAML 的 `env.commands` 下配置
-   `UniformVelocityCommandCfg`，参考 Go1 flat owner。
+   `UniformVelocityCommandCfg`，参考 `go2_joystick_flat` owner。
 5. 翻译地形 —— Legged Gym 的高度场生成器在 UniLab 中有一个对应物，位于
    `unilab.terrains.heightfield_terrains`。
 

docs/sphinx/source/zh_CN/4-developer_guide/3-extending/4-new_terrain.md

@@ -25,8 +25,6 @@
 - 地形生成器形状与数值行为：
   `tests/terrains/test_terrain_generator.py`
 - Rough locomotion 高度扫描与 spawn 行为：
-  `tests/envs/locomotion/test_go2_rough_height_scan.py`、
-  `tests/envs/locomotion/test_go2_terrain_spawn.py`、
   `tests/envs/locomotion/test_terrain_spawn.py`
 - 后端 materialization 边界：`tests/utils/test_xml_utils.py`
 

pyproject.rocm.toml

@@ -24,13 +24,13 @@ requires-python = ">=3.10,<3.14"
 dependencies = [
     "numpy",
     # Physics implementations are provided by the independently released
-    # unisim-core package. The 1.4.1 release carries the fixed model variant,
+    # unisim-core package. The 1.4.2 release carries the fixed model variant,
     # per-world reset-default, per-env gravity, substep body-wrench, and
     # IsaacGym fixed-variant contracts.
-    "unisim-core>=1.4.1",
+    "unisim-core>=1.4.2",
     # RL algorithms and async runtimes live in the independently released
     # uni-rl package (distribution name ``unilab-rl``); see pyproject.toml.
-    "unilab-rl==1.2.0",
+    "unilab-rl==1.2.1",
     "torch==2.11.0",
     "triton-rocm==3.6.0 ; sys_platform == 'linux' and platform_machine == 'x86_64'",
     "gymnasium",
@@ -61,7 +61,6 @@ eval = "unilab.cli:eval_main"
 demo = "unilab.cli:demo_main"
 unilab-complete = "unilab.cli_completion:main"
 unilab-viz-nan = "unilab.utils.nan_viz:main"
-unilab-export-scene = "unisim.backend.mujoco.export_scene:main"
 unilab-render-teaser = "unilab.visualization.teaser:main"
 unilab-pull-assets = "unilab.assets.pull:main"
 
@@ -163,7 +162,6 @@ venv = ".venv"
 include = ["src/unilab"]
 exclude = [
     "src/unilab/envs/",                     # lazy optional `mujoco` import; not installed in the pyright env
-    "src/unilab/utils/monitoring.py",       # optional pynvml/psutil deps
     "src/unilab/visualization/",            # direct mujoco C bindings + optional viser deps
 ]
 reportMissingImports = "warning"

pyproject.toml

@@ -39,15 +39,15 @@ requires-python = ">=3.10,<3.14"
 dependencies = [
     "numpy",
     # Physics implementations are provided by the independently released
-    # unisim-core package. The 1.4.1 release carries the fixed model variant,
+    # unisim-core package. The 1.4.2 release carries the fixed model variant,
     # per-world reset-default, per-env gravity, substep body-wrench, and
     # IsaacGym fixed-variant contracts.
-    "unisim-core>=1.4.1",
+    "unisim-core>=1.4.2",
     # RL algorithms and async runtimes (PPO/APPO/SAC/TD3 runners,
     # collectors, IPC, logging) live in the independently released uni-rl
     # package (distribution name ``unilab-rl``), consumed via the injected
     # env contract (uni_rl.env_contract.EnvFactory). Published on PyPI.
-    "unilab-rl==1.2.0",
+    "unilab-rl==1.2.1",
     "numba>=0.67",
     "prettytable>=3.10",
     # torch is a range (not an exact pin) so that published PyPI metadata lets
@@ -89,7 +89,6 @@ eval = "unilab.cli:eval_main"
 demo = "unilab.cli:demo_main"
 unilab-complete = "unilab.cli_completion:main"
 unilab-viz-nan = "unilab.utils.nan_viz:main"
-unilab-export-scene = "unisim.backend.mujoco.export_scene:main"
 unilab-render-teaser = "unilab.visualization.teaser:main"
 unilab-pull-assets = "unilab.assets.pull:main"
 
@@ -171,7 +170,7 @@ viser = ["viser>=1.0.26", "trimesh>=3.21.7"]
 # required-environments; elsewhere the extra is empty and the CLI reports a
 # targeted runtime diagnostic.
 superdex = [
-    "unisim-core[superdex]>=1.4.0 ; python_version >= '3.12' and sys_platform == 'linux' and platform_machine == 'x86_64'",
+    "unisim-core[superdex]>=1.4.2 ; python_version >= '3.12' and sys_platform == 'linux' and platform_machine == 'x86_64'",
 ]
 
 [dependency-groups]
@@ -250,7 +249,6 @@ venv = ".venv"
 include = ["src/unilab"]
 exclude = [
     "src/unilab/envs/",                     # lazy optional `mujoco` import; not installed in the pyright env
-    "src/unilab/utils/monitoring.py",       # optional pynvml/psutil deps
     "src/unilab/visualization/",            # direct mujoco C bindings + optional viser deps
 ]
 reportMissingImports = "warning"