LEMO agent pipeline runs on real robots, completing supervised tennis-ball-in-cup task
du_yilun · x · 2026-10-17
ShaotingZ38103 reports deploying the LEMO agent pipeline on their own robot with excellent results, calling it a robotics revolution. A screenshot shows the agent workspace: the agent reads the task "put the tennis ball in the cup," consults a real-robot adaptation doc, runs git diff --check, then executes a Code-as-Policy grasp — ball in cup, gripper raised and withdrawn. Wrist camera frames and code/docs are logged, and the policy is saved into a skill library (10 baseline checks passed; supervised success, autonomous replay not yet verified).
More from coding & agent
- AI agents exceeded their scope — what actually enforces the boundary before execution? — docybo · 2026-10-17
- IBM AI engineer maps 5 levels of agentic coding — and why he stays at Level 2 — TejasKumar_ · 2026-10-17
- Data viz principles are reshaping everyday UI design in the AI era — enjalot · 2026-10-17
- 7 open-source repos that turn Claude into a knowledge-graph-powered memory system — blaizedsouza · 2026-10-17
- No drone needed: Claude + Fish Creative MCP generates realistic FPV cliff-flight video — socialwithaayan · 2026-10-17
- Modular launches LLM Inference Handbook, a free one-stop reference for inference engineering — techNmak · 2026-10-17