인터페이스는 현지화되어 있으며, Skill 소개와 설명 문서는 선택한 언어로 AI가 자동 번역합니다. 코드·명령어·경로·전문 용어는 원문을 유지하며 원작자 문서가 우선합니다.
Skills 둘러보기
데이터 및 연구 · SKILL.md

real-robot-operations

real-robot-operations

Safety-gated real Push-T robot collection, evaluation, sensing, shared-memory IO, and real dataset conversion.

Independent GitHub snapshot · Automated classification · Original documentation governs

Skill 문서

GitHub

name: real-robot-operations description: "Safety-gated real Push-T robot collection, evaluation, sensing, shared-memory IO, and real dataset conversion." disable-model-invocation: true metadata: disco-role: operating license: MIT

Real Robot Operations

Use this sub-skill for UR5 + RealSense + SpaceMouse workflows around real Push-T demo collection, checkpoint evaluation, shared-memory motion/camera IO, timestamp alignment, and raw real-data conversion.

Route here when

  • a task touches live robot preflight, demo capture, evaluation rollout, or real-robot data layout
  • you need to reason about RealEnv, RTDEInterpolationController, SingleRealsense, MultiRealsense, Spacemouse, SharedMemoryRingBuffer, or SharedMemoryQueue
  • you need to convert recorded robot episodes into a training-ready replay buffer or compute real Push-T metrics
  • you need to diagnose timestamp, latency, camera, SpaceMouse, or RTDE failures

Route elsewhere when

  • the task is simulation-only training, evaluation, or Hydra workspace composition: use ../training-and-evaluation/
  • the task is generic ReplayBuffer, dataset schema, or sampler work: use ../data-and-replay-buffers/
  • the task is model, policy, checkpoint, or prediction-shape reasoning: use ../policies-and-models/

Bundled references

Operating contract

  • Never start the robot, cameras, or spacenavd from this skill.
  • Treat emergency-stop reachability, camera connection, and SpaceMouse presence as manual safety confirmations.
  • Use the bundled prereq checker only for import/executable/service inspection and optional RTDE socket reachability.
  • Keep all real-robot commands bounded by the target robot IP, a known output directory, and the exact dataset/checkpoint path.
저장한 Skills