---

name: scene-configuration description: "This skill guides authors through validated IR-SIM YAML scenes, object geometry, kinematics, placement, collision policy, and state APIs." disable-model-invocation: true metadata: disco-role: operating license: MIT

Scene configuration

Use this sub-skill when a task needs an IR-SIM YAML scene, a robot or obstacle, object geometry, a kinematic control vector, placement distributions, collision policy, or object state/goal inspection. Begin with the smallest explicit scene and run the bundled checker before constructing an environment:

python sub-skills/scene-configuration/scripts/validate_scene.py --help
python sub-skills/scene-configuration/scripts/validate_scene.py scene.yaml

Run those commands from the generated IR-SIM skill root, or resolve the script path from the installed skill location. The checker uses only yaml.safe_load; it does not import IR-SIM or open a GUI. It returns non-zero for malformed YAML, unknown schema keys, unsupported shape/kinematics/behavior combinations, ambiguous Ackermann geometry, invalid control dimensions, and mismatched per-object lists. It is a strict preflight, not a replacement for irsim.make().

Route by intent

• Author or diagnose world, robot, obstacle, or gui YAML with the schema reference.
• Choose a footprint, understand exact collision geometry, or map a control vector to state updates with geometry and kinematics.
• Need repeatable placement, circle formation, random shapes/goals, or central RNG behavior? Use distributions and randomness.
• Diagnose install, parser, factory, dimension, collision, path, and optional dependency failures with troubleshooting.

Minimal construction loop

1. Install the base distribution (python -m pip install ir-sim) and verify import irsim; use display=False for headless work. pynput is optional keyboard input, pyrvo is optional ORCA, and imageio[ffmpeg]/ffmpeg is only needed for video output.
2. Define world extents/timing and collision policy, then one robot with an explicit shape, kinematics, state, goal, and compatible behavior when it should move. Objects without kinematics are static.
3. Add obstacles only after checking that initial Shapely footprints do not overlap. Keep explicit object names unique across both sections; group and group_name organize objects but do not replace names.
4. Validate a tiny fixture, then use irsim.make("scene.yaml", display=False, seed=7). The factory accepts one mapping or a list of object-group mappings; number expands a group and the runtime repeats short per-object lists, while the bundled checker asks for exact lists to prevent accidental duplication.
5. Query obj.state, obj.velocity, obj.goal, obj.geometry, obj.get_info(), obj.get_obstacle_info(), obj.arrive, and obj.collision. Use set_state, set_velocity, and set_goal; after direct state changes, let the environment refresh/rebuild its spatial data before relying on collision or sensor queries.

Keep this route focused on scene/object contracts. Route environment lifecycle, step/render/reset, and external stepping to simulation-environments; sensor payloads and maps to sensing-and-mapping; behavior selection and planners to navigation-and-planning; and registry or custom-controller work to extension-and-control. Runtime files are self-contained: the original checkout, usage files, private environments, and review artifacts are not dependencies.