协作头脑风暴
brainstorming
在实现前澄清需求、探索设计选择,并形成清晰的开发方向。
simulation-environments【开发与工程】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.
独立 GitHub 快照 · 自动分类 · 以原作者文档为准
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.
import irsim; add pynput only for
live keyboard input and imageio[ffmpeg] only for video output. No GPU is
required for the covered workflows.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.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.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.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.
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.