navigation-environment
Operate and validate the GazeboEnv observation, action, reset, timing, and reward contract for the ROS/Gazebo Velodyne navigation environment without requiring a live simulator.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- navigation-environment
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Operate and validate the GazeboEnv observation, action, reset, timing, and reward contract for the ROS/Gazebo Velodyne navigation environment without requiring a live simulator.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/api-reference.md5.3 KB
- 参考文档references/state-action-reward.md4.5 KB
- 参考文档references/troubleshooting.md4.9 KB
- 脚本scripts/validate_navigation_contract.py10.6 KB
03 SKILL.md 正文
Navigation environment
Use this skill when a downstream agent needs to reason about the mobile-robot
GazeboEnv interface, prepare a compatible observation/action, diagnose a
sensor or ROS integration failure, or check a synthetic transition. This skill
is a distilled contract, not a simulator launcher and not a TD3 training guide.
Operating contract
- The repository's training configuration uses
environment_dim=20and four robot features, so every observation is a flat numeric vector of length 24. - The first 20 values are Velodyne distance bins, initialized to 10 and
updated by the point-cloud callback. The final four values are, in order,
distance to goal, relative goal heading, linear action, and angular action.
resetreturns the two action slots as 0;stepplaces the action used for that transition in them. There is no separate previous-action history in the implementation, so do not invent one. - The environment-facing action is
[linear_x, angular_z]: linear is in[0, 1], angular is in[-1, 1].steppublishes these values directly; it does not rescale them. A TD3 actor output in[-1, 1]therefore needs its first component mapped with(a_linear + 1) / 2before callingstep. donebecomes true on a laser collision or a goal hit. A goal hit is reported separately astarget=True; collision and target reward priority is target first, then collision.- Use
scripts/validate_navigation_contract.pyfor offline numeric checks. It imports no ROS package, starts no process, and is the preferred check for state length, action range, point-bin fixtures, and threshold behavior.
Safe reasoning workflow
- Confirm the caller has ROS Noetic/Ubuntu 20.04-compatible ROS message and
Gazebo dependencies, plus Python dependencies used by the environment
(
numpyandsquaternion). A live run additionally needsroscore,roslaunch, Gazebo, and a built catkin workspace with the scenario and Velodyne plugin packages. This host-independent skill does not install or launch any of them. - Treat a relative launch name as relative to the environment module's
assetsdirectory, not the caller's current directory. The repository'smulti_robot_scenario.launchincludes the empty TD3 world and spawns modelr1. - Validate an observation as 20 finite sensor values followed by four finite robot values. Reject 23- or 25-value vectors rather than silently padding or truncating them.
- Validate action-space provenance. If an actor produced
[-1, 1]for both dimensions, transform only the linear component before using this environment contract. - For a synthetic sensor case, reduce points into the same 20 angular bins, compute the minimum distance per bin, and then apply the strict collision threshold. Keep simulator verification explicitly unavailable when ROS or Gazebo is absent.
- Interpret
resetandstepas ROS-dependent calls. A valid offline vector does not prove that odometry, PointCloud2, publishers, or physics services are connected.
Scope boundaries
This skill covers constructor prerequisites and launch resolution, point-cloud reduction, state/action/reward semantics, reset randomization, ROS interfaces, and predictable numeric/integration failures. Do not use it to explain TD3 network architecture, replay-buffer behavior, optimizer settings, training loops, model checkpoints, or installation procedures beyond naming prerequisites. See the bundled references for the detailed contract and troubleshooting.
Numeric and evidence notes
The source callback has no explicit finite-value or zero-horizontal-range
guard. Use the validator's stricter finite-input policy for synthetic data;
never turn NaN or an acos domain error into a claimed sensor reading.
Evidence and verification status
The contract was distilled from the pinned repository's environment module,
launch file, scenario xacro/URDF, Velodyne description and Gazebo plugin, and
training caller. The offline validator is checked with its help/parser path,
self-test, a valid 24-value fixture, invalid 23/25-value fixtures, action-range
mismatches, a synthetic point cloud, and strict threshold cases. No ROS,
Gazebo, roscore, roslaunch, or catkin runtime is claimed or required here.
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).