mdp-components
Use mjlab built-in actions, actuators, observations, rewards, terminations, events, commands, curricula, metrics, and task-specific MDP terms.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- mdp-components
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Use mjlab built-in actions, actuators, observations, rewards, terminations, events, commands, curricula, metrics, and task-specific MDP terms.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/actions-actuators.md3.1 KB
- 参考文档references/mdp-term-catalog.md2.9 KB
- 参考文档references/task-specific-mdp.md2.6 KB
- 参考文档references/troubleshooting.md2.7 KB
- 脚本scripts/inspect_mdp_terms.py2.0 KB
03 SKILL.md 正文
mdp-components
Use this sub-skill when a task asks which mjlab MDP term, action config, actuator config, command generator, reward, termination, curriculum, or built-in task-family component to use.
Start here
- Read actions-actuators.md for policy action terms, actuator configs, scaling/offset/clip, delay, and differential IK.
- Read mdp-term-catalog.md for built-in observation, reward, termination, event, command, curriculum, and metric functions.
- Read task-specific-mdp.md for velocity, tracking, manipulation, and cartpole term patterns.
- Read troubleshooting.md when actions have the wrong dimension, rewards have the wrong sign, commands are missing, or target regexes fail.
- Run scripts/inspect_mdp_terms.py to list public callables/signatures from installed mjlab MDP modules.
Quick choices
| Need | Typical mjlab surface |
|---|---|
| Drive joints from policy output | JointPositionActionCfg, JointVelocityActionCfg, JointEffortActionCfg, or RelativeJointPositionActionCfg |
| Reuse XML actuators | XmlActuatorCfg on EntityArticulationInfoCfg |
| Stable implicit PD in MuJoCo | BuiltinPositionActuatorCfg, BuiltinPdActuatorCfg, or BuiltinDcMotorActuatorCfg |
| Explicit torque/PD/DC dynamics | IdealPdActuatorCfg, DcMotorActuatorCfg, LearnedMlpActuatorCfg |
| Robot target lookup | SceneEntityCfg with joint/body/geom/site/actuator names |
| Standard observations | mjlab.envs.mdp.observations functions |
| Standard penalties/rewards | mjlab.envs.mdp.rewards plus task-family rewards |
| Episode stop logic | mjlab.envs.mdp.terminations and task-family terminations |
| Reset or perturb states | EventTermCfg with event functions |
Boundary routing
- Manager dictionary shape and lifecycle: environment-configuration.
- Entity/spec attachment and scene assembly: scene-simulation-assets.
- Sensor, terrain, and domain-randomization details: perception-terrain-randomization.
- Task registry and CLI training/play execution: training-evaluation-cli.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).