界面已本地化;技能介绍与说明文档按所选语言由 AI 自动翻译,代码、命令、路径和专业术语保留原文,以原作者文档为准。
探索技能
开发与工程 · SKILL.md

task-development

task-development

【开发与工程】Guides RoboVerse task authoring, registration, observation and reward contracts, reset logic, callback composition, and parity-safe task tests.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准

技能说明

GitHub

name: task-development description: "Guides RoboVerse task authoring, registration, observation and reward contracts, reset logic, callback composition, and parity-safe task tests." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

Task Development

Use this route when adding or changing a RoboVerse task, task config, reward, observation, reset, callback, success checker, or task registration. Decide first whether the change belongs in RoboVerse (content and task behavior) or MetaSim (core abstractions, registry, handlers, or simulator backend).

Route

  1. Read task-authoring.md for the smallest config/registration workflow and state-field checklist.
  2. Compose existing MetaSim and RoboVerse config objects with @configclass; do not fork core types or work around a missing MetaSim capability in the downstream package.
  3. Define the scenario, task class, action/observation contract, reset, reward, termination, and success behavior. Register the task under a stable name and import it through the package discovery path.
  4. Add a focused regression test for every behavior fix. Run a lightweight reset/shape test first, then backend-specific parity checks only when the selected environment supports them.
  5. Use troubleshooting.md for registration, state/extras, shape, seed, and reward-order failures.

Key boundaries

  • Core simulator APIs, registry semantics, and backend bugs route to MetaSim.
  • Task/robot/scene/reward/observation content routes here.
  • External benchmark passthroughs route to benchmark-integrations.
  • Measured cross-simulator comparisons route to parity-and-tooling.
我的收藏