observations-actions-rewards
Configure HighwayEnv observations, actions, rewards, and info outputs.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- observations-actions-rewards
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Configure HighwayEnv observations, actions, rewards, and info outputs.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/actions-rewards.md7.3 KB
- 参考文档references/configuration-reference.md5.9 KB
- 参考文档references/observations.md6.7 KB
- 参考文档references/troubleshooting.md6.6 KB
- 脚本scripts/inspect_spaces.py9.4 KB
03 SKILL.md 正文
Observations, actions, and rewards for HighwayEnv
Use this sub-skill when a task is about selecting or debugging HighwayEnv
observation spaces, action spaces, reward component interpretation, goal-style
parking observations, info outputs, or safe config validation.
Read this first
- Read references/configuration-reference.md
when creating or modifying a
config={...}dictionary, updating config at reset time, validating nested config shape, or choosing complete snippets. - Read references/observations.md when selecting
between
Kinematics,OccupancyGrid,TimeToCollision,LidarObservation,KinematicsGoal, image, tuple, attributes, exit, or multi-agent observations. - Read references/actions-rewards.md when
choosing
DiscreteMetaAction,ContinuousAction,DiscreteAction, orMultiAgentAction; mapping discrete labels; usingget_available_actions(); or explaining scalar rewards,info["rewards"], andinfo["is_success"]. - Read references/troubleshooting.md when an
observation/action config fails, an action has the wrong shape or unavailable
label,
infois missing a key, or a space shape is surprising. - Run scripts/inspect_spaces.py for a safe one-step JSON inspection of an env id plus optional JSON config before giving an agent a final space/reward interpretation.
Boundaries and routing
This sub-skill owns observation/action/reward configuration and interpretation. Route these adjacent tasks elsewhere:
- Environment creation, reset/step loops, vectorization, render modes, video wrappers, and display/headless issues: use simulation-environments, and for training video workflows use training-and-evaluation.
- Custom roads, lanes, vehicles, new environment classes, and road geometry that changes what observations see: use road-vehicle-dynamics.
- RL algorithm choice, Stable-Baselines3/Torch setup, long training loops, evaluation budgets, and benchmark reporting: use training-and-evaluation.
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).