L’interface est localisée ; les présentations et documents des Skills sont traduits automatiquement par IA dans votre langue. Le code, les commandes, chemins et termes techniques restent en version originale, et le document de l’auteur fait foi.
Explorer les Skills
Développement · SKILL.md

simulation-environments

simulation-environments

The simulation-environments skill guides users through creating, stepping, inspecting, rendering, resetting, and closing isolated IR-SIM environments in 2D, 3D, headless, and externally controlled workflows.

Independent GitHub snapshot · Automated classification · Original documentation governs

Documentation de la Skill

GitHub

name: simulation-environments description: "The simulation-environments skill guides users through creating, stepping, inspecting, rendering, resetting, and closing isolated IR-SIM environments in 2D, 3D, headless, and externally controlled workflows." disable-model-invocation: true metadata: disco-role: operating license: MIT

Simulation environments

Use this sub-skill when the task is to run IR-SIM, choose a YAML world, control its lifecycle, use internal or external state advancement, render safely, or run more than one environment. Keep scene YAML/object schemas in scene-configuration, sensor and map payloads in sensing-and-mapping, planners and behaviors in navigation-and-planning, and custom controllers/registries or live GUI input in extension-and-control.

Fast route

  1. Install the base package and verify import irsim; add pynput only for live keyboard input and imageio[ffmpeg] only for video output. No GPU is required for the covered workflows.
  2. Start from a self-contained YAML path: env = irsim.make("world.yaml", display=False, seed=7) for batch/headless work, or omit display=False only when a supported desktop backend is intended.
  3. Run env.step() then env.render() in internal mode, check env.done(), and always call env.close(ending_time=0) in a finally block. Use scripts/render_smoke.py --help for a safe, tiny, arbitrary-cwd check.
  4. Select step_mode="external" when another system owns state. Mutate every relevant object's state and velocity, call env.step() without an action, and use env.refresh() when a direct mutation needs derived data immediately.
  5. Read the linked references for exact signatures, reset/reload semantics, action alignment, multi-environment/RNG caveats, and optional rendering.

Scope boundary

This skill covers irsim.make, EnvBase/EnvBase3D, clocks and statuses, headless rendering and figure output, lifecycle cleanup, query/draw helpers, seed handling, and isolated environment instances. It does not define the full YAML object schema, sensor array formats, planner contracts, or extension registration APIs; follow the cross-links above instead.

Evidence and limits

The operating guidance is based on the IR-SIM 2.10.2 public entry point and environment/plot implementations, the English quick-start, environment, multiple-environment, and animation documentation, and focused lifecycle and plot test evidence. Interactive keyboard events, desktop windows, MP4/ffmpeg, and long visual demos are intentionally not bundled or claimed as verified. The helper adapts only a tiny headless render and does not depend on an original checkout file.

Skills enregistrées