Code-only-as-policy beats VLA and agentic harness on RoboDojo, no VLM in the control loop
liuziwei7 · x · 2026-10-14
Liu Ziwei's team proposes the Embodied Turing Machine and Code-Only-As-Policy (COAP): the robot is the tape, environment state and rules are the policy — exploring how far robot policies can go without a VLM/VLA in the runtime control loop.
- Three paradigms compared: VLA (learned weights), Agentic Harness (model + context + tools), COAP (code + state → actions)
- Recursive self-improvement loop: observe → debug policy.py → validate with tests → save, reuse, refactor
- Results: on RoboDojo, COAP scores 75.45 progress score, beating Agentic (70.24) and VLA (33.95)
Related event: Embodied Turing Machines: Code-Only Robot Policies(3 posts)→
More from Embodied
- China reportedly shipped 97% of the world's humanoid robots in H1 2026 — Polymarket · 2026-10-14
- Physical Intelligence opens Pi API beta, one model adapting robots from strawberry harvesting to server assembly — svlevine · 2026-10-14
- China's Tiangong Ultra robot beat Bolt's 100m record — and exposes America's robotics supply-chain dilemma — mattsheehan88 · 2026-10-14
- GoDrift shows off realtime sim2real robot demo; 2D plane eases reaction demands — jonstephens85 · 2026-10-14
- Microsoft science president Peter Lee leaves to become Mercedes F1 team's AI advisor — lazowska · 2026-10-14
- NVIDIA releases GR00T N1.7 AppleToPlate checkpoint-15000 on Hugging Face — nvidia · 2026-10-14