Idea Discovery Robot

This orchestrates a four-stage pipeline that takes a robotics direction and outputs validated, benchmark-grounded ideas: literature survey filtered through CoRL/RSS/ICRA venues, idea generation with explicit embodiment and simulator constraints, novelty checking, and critical review via external models. It enforces sim-first pilots by default and blocks real robot execution unless you explicitly approve it. Every idea must specify target benchmark, failure modes, and whether hardware is actually needed. Use it when you want to go from "bimanual manipulation" or "drone navigation" to concrete project proposals that are falsifiable and feasible, not just flashy demos. The robotics framing (embodiment, action interface, observation modality, benchmark availability) gets locked down before any idea generation happens, which keeps the output grounded in actual infrastructure instead of vague ML research.

Install to Claude Code

npx -y skills add wanshuiyin/auto-claude-code-research-in-sleep --skill idea-discovery-robot --agent claude-code