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
效率与协作 · SKILL.md

rendering

rendering

【效率与协作】Configure robosuite rendering, camera observations, transforms, offscreen capture, domain randomization, and optional USD/Isaac export paths.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准

技能说明

GitHub

name: rendering description: "Configure robosuite rendering, camera observations, transforms, offscreen capture, domain randomization, and optional USD/Isaac export paths." disable-model-invocation: true metadata: disco-role: operating license: NOASSERTION

Rendering

Use this sub-skill for rendering and camera-adjacent tasks in robosuite:

  • choose has_renderer vs has_offscreen_renderer
  • select renderer="mujoco" or renderer="mjviewer"
  • request camera RGB, depth, and segmentation observations
  • read and back-project camera transforms
  • record short offscreen videos
  • randomize cameras, lighting, textures, dynamics, and sensor noise
  • follow optional USD / Isaac Sim export paths

Start with:

Bundled helpers:

Do not handle here:

  • ../environments for task setup, horizons, reward flow, and general env creation
  • ../teleoperation for keyboard/SpaceMouse/DualSense capture and teleop video/USD collection
  • ../modeling for camera XML tuning and MJCF camera placement

Core rules:

  1. use_camera_obs=True requires has_offscreen_renderer=True.
  2. Depth observations are normalized; convert them with get_real_depth_map before metric use.
  3. Segmentation observations appear only when camera_segmentations is enabled, with keys like frontview_segmentation_instance.
  4. Set robosuite.macros.IMAGE_CONVENTION = "opencv" before env creation when writing video with imageio or OpenCV-style consumers.
  5. Treat USD/Isaac export as optional and environment-dependent, not as a core verified runtime path.
我的收藏