scene-simulation-assets
Compose mjlab scenes, entities, variants, simulation configs, asset-zoo robots, and exportable MuJoCo packages.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- scene-simulation-assets
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Compose mjlab scenes, entities, variants, simulation configs, asset-zoo robots, and exportable MuJoCo packages.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/assets-export-variants.md2.9 KB
- 参考文档references/scene-entity-simulation.md3.4 KB
- 参考文档references/spec-editors.md2.5 KB
- 参考文档references/troubleshooting.md2.4 KB
- 脚本scripts/export_scene_smoke.py5.1 KB
03 SKILL.md 正文
Scene, Simulation, and Assets
Use this sub-skill when the task is about mjlab scene assembly or the MuJoCo
asset layer: EntityCfg, VariantEntityCfg, SceneCfg, TerrainEntityCfg,
entity-level actuator attachment, SimulationCfg / MujocoCfg, asset-zoo
robots, scene export, or direct mujoco.MjSpec / MuJoCo Warp model-data access.
Start Here
- For the scene/entity/simulation object model, lifecycle order, namespacing, data access, and graph constraints, read scene-entity-simulation.md.
- For asset-zoo robot factories, export targets,
Scene.write(...), and heterogeneous mesh variants, read assets-export-variants.md. - For texture, material, mesh, geom, collision, light, camera, and cross-entity spec editing, read spec-editors.md.
- For common MJCF, keyframe, mocap, namespacing, variant, export, and CUDA graph failures, read troubleshooting.md.
- For a safe scene-export smoke check, use scripts/export_scene_smoke.py.
Core Runtime Contract
When manually constructing the runtime, follow the same order used by mjlab's manager-based environment:
from mjlab.scene import Scene
from mjlab.sim import Simulation, SimulationCfg
scene = Scene(scene_cfg, device="cpu")
sim = Simulation(
scene.num_envs,
SimulationCfg(),
spec=scene.spec,
variant_info=scene.collect_variant_info(),
device="cpu",
)
scene.initialize(sim.mj_model, sim.model, sim.data)
if scene.sensor_context is not None:
sim.set_sensor_context(scene.sensor_context)
Use scene.write(output_dir, zip=False) or scene.write(output_dir, zip=True)
for exportable packages. For direct runtime reads, prefer entity.data and use
sim.model / sim.data only when a task needs global MuJoCo IDs or fields.
Boundary Routing
- Route action-term selection, policy-output scaling, reward/action wiring, and detailed actuator type tradeoffs to ../mdp-components/. This sub-skill only covers how actuator configs attach to an entity spec.
- Route camera/raycast/contact sensor configuration, terrain generation presets,
flat-patch sampling, and domain-randomization recipes to
../perception-terrain-randomization/.
This sub-skill covers only how those objects connect to
SceneCfgandSimulation. - Route installed
export-sceneCLI usage, task-registry command workflows, training, playback, and debug CLI flags to ../training-evaluation-cli/. This sub-skill is the API-internals reference for what that export command builds.
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).