NTU's COAP: code replaces LLMs in the robot control loop, hitting 75% on 42 bimanual tasks
liuziwei7 · x · 2026-10-14
Ziwei Liu's NTU team proposes Embodied Turing Machines and Code-Only-as-Policy (COAP): coding agents build and debug a shared Python library offline, while the robot runs plain code online — no VLA, no VLM in the loop. Three RGB cameras plus proprioception feed a NumPy/OpenCV state tracker that stores explicit object poses, grasp state and progress; the policy follows measure → act → verify → recover. On RoboDojo's 42 bimanual tasks the library reaches 75.45% success vs 33.95% for VLA and 36.27% for agentic harnesses, and parallel code branches make recursive self-improvement explicit and fast.
More from Embodied
- NTU team has Claude write robot control code, hitting 70.24% success in sim — aigclink · 2026-10-14
- Rabbit R1 called 'ahead of its time' as modders plan to hack the once-mocked AI device — SimonBalmain · 2026-10-14
- RSI-Nav: recursive self-improvement for navigation agents hits 68.6% SR on R2R-CE — OpenGVLab · 2026-10-14
- Cepheid decouples scale from structure for cross-family sparse-to-dense motion completion — NYCU · 2026-10-14
- Visor: open-source Android app replaces the Meta AI app on Ray-Ban Meta glasses with your own model — jedisct1 · 2026-10-14
- 304 robotics papers hit arXiv in a single day as the field explodes — DJiafei · 2026-10-14