后端支持矩阵¶
本页是后端参考页,放生成矩阵和需要精确查证的 backend 规则。它不承担首次阅读职责。
适合谁看¶
想按 task owner / algorithm / backend 精确查支持状态
想知道
Registered、Configured、Tested的证据差异想确认 playback 和 owner compose 的 backend 规则
Backend 选择规则¶
默认后端是
mujoco切到 Motrix 用统一 CLI 的
--sim motrix--algo、--task、--sim共同选择 owner YAML不要把
training.sim_backend当独立 backend switch
Playback Differences¶
mujoco:--render-mode auto会导出play_video.mp4motrix:--render-mode auto会打开交互式 renderer 窗口,不录制视频,不受play_steps限制--render-mode record: 两个后端都只录制视频--render-mode none: 不回放
Support Matrix¶
下面的矩阵由 registry、owner YAML 和测试清单自动汇总;不要手工编辑表格内容。需要刷新时运行:
uv run scripts/generate_support_matrix.py --write
Evidence Grades¶
等级 |
仓库事实来源 |
|---|---|
|
|
|
存在对应的 owner YAML: |
|
|
|
存在与该组合绑定的已提交 benchmark manifest。 |
|
仓库中存在显式 recommendation 元数据。 |
Tested 只描述仓库中已有自动化覆盖,不代表该组合具备同名 MuJoCo owner 的全部 backend capability;例如 phase-1 Motrix owner 可能只覆盖训练 smoke 和明确启用的 DR 子集。
未检测到与这些组合绑定的已提交 benchmark manifest,因此当前不会自动提升到 Benchmarked。
仓库中目前也没有单独的 recommendation 元数据,因此当前不会自动提升到 Recommended。
Entrypoint x Task Owner¶
Entrypoint |
Task owner |
MuJoCo |
Motrix |
|---|---|---|---|
PPO (torch) |
|
Tested |
Tested |
PPO (torch) |
|
Tested |
Tested |
PPO (torch) |
|
Tested |
Tested |
PPO (torch) |
|
Tested |
Tested |
PPO (torch) |
|
Tested |
Tested |
PPO (torch) |
|
Tested |
Tested |
PPO (torch) |
|
Tested |
Tested |
PPO (torch) |
|
Tested |
Tested |
PPO (torch) |
|
Tested |
Tested |
PPO (torch) |
|
Tested |
Tested |
PPO (torch) |
|
Tested |
Tested |
PPO (torch) |
|
Tested |
Tested |
PPO (torch) |
|
Tested |
Tested |
PPO (torch) |
|
Tested |
Tested |
PPO (torch) |
|
Tested |
Tested |
PPO (torch) |
|
Tested |
Tested |
PPO (torch) |
|
Tested |
- |
PPO (torch) |
|
Tested |
Tested |
PPO (torch) |
|
Tested |
Tested |
PPO (torch) |
|
Tested |
Tested |
PPO (mlx) |
|
Tested |
Tested |
PPO (mlx) |
|
Tested |
Tested |
PPO (mlx) |
|
Configured |
Configured |
PPO (mlx) |
|
Tested |
Tested |
PPO (mlx) |
|
Configured |
Configured |
PPO (mlx) |
|
Configured |
Configured |
PPO (mlx) |
|
Configured |
Configured |
PPO (mlx) |
|
Configured |
Configured |
PPO (mlx) |
|
Configured |
Configured |
PPO (mlx) |
|
Configured |
Configured |
PPO (mlx) |
|
Configured |
Configured |
PPO (mlx) |
|
Configured |
Configured |
PPO (mlx) |
|
Configured |
Configured |
PPO (mlx) |
|
Configured |
Configured |
PPO (mlx) |
|
Configured |
Configured |
PPO (mlx) |
|
Configured |
Configured |
PPO (mlx) |
|
Configured |
- |
PPO (mlx) |
|
Configured |
Configured |
PPO (mlx) |
|
Configured |
Configured |
PPO (mlx) |
|
Configured |
Configured |
APPO (torch) |
|
Tested |
Registered |
APPO (torch) |
|
Tested |
Tested |
APPO (torch) |
|
Tested |
Registered |
APPO (torch) |
|
Tested |
Tested |
APPO (torch) |
|
Tested |
Tested |
APPO (torch) |
|
Tested |
Tested |
APPO (torch) |
|
Tested |
Tested |
APPO (torch) |
|
Tested |
Tested |
APPO (torch) |
|
Tested |
Tested |
SAC (torch) |
|
Tested |
Tested |
SAC (torch) |
|
Tested |
Tested |
SAC (torch) |
|
Tested |
Tested |
SAC (torch) |
|
Tested |
Registered |
SAC (torch) |
|
Tested |
Registered |
SAC (torch) |
|
Tested |
Registered |
TD3 (torch) |
|
Registered |
Tested |
TD3 (torch) |
|
Registered |
Tested |
TD3 (torch) |
|
Tested |
Registered |
FlashSAC (torch) |
|
Tested |
Registered |
FlashSAC (torch) |
|
Tested |
Registered |
Source Index¶
Registry bootstrap:
src/unilab/envs/**decorators viaunilab.base.registry.ensure_registries().Owner YAML scan:
conf/ppo/task/**,conf/appo/task/**,conf/offpolicy/task/**.Generic compose coverage:
tests/config/test_config_system.py::test_supported_task_composes.MLX-specific compose coverage only upgrades task owners listed in
tests/config/test_config_system.py::_PPO_MLX_TASKS:go1_joystick_flat,go2_joystick_flat,g1_walk_flat.MLX runtime smoke:
tests/algos/test_mlx_ppo.py::test_mlx_ppo_one_iteration_real_envcurrently exercisesgo2_joystick_flat/mujoco.