AI

Embodied Turing Machines: Stateful Code for Robot Recursive Self-Improvement

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 09/10/2026, 05:18:24 UTC.

Briefings

This article proposes Code-Only-as-Policy (COAP) as a paradigm for embodied tasks, where robot decision-making is entirely encoded in code rather than relying on vision-language models. On RoboDojo's 42 bimanual tasks, the resulting library achieved a 70.24% success rate without a model at test time. The method offers explicit state tracking, controllable execution, and task extensibility, serving as a foundation for recursive self-improvement.

Key points

• The proposed COAP paradigm replaces vision-language models with code-only decision-making for embodied tasks.
• On RoboDojo's 42 bimanual tasks, the COAP library achieved a 70.24% success rate without a model at test time.

Context

Embodied AI systems typically use vision-language models (VLMs) or vision-language-action models (VLAs) to bridge perception and action. This research shifts focus to a stateful code-based approach, offering a new paradigm for embodied tasks.

Limitations

The study is limited to RoboDojo's 42 bimanual tasks. Success rates are measured at test time without a model, and the upper bound of COAP depends on state representation accuracy and code logic robustness.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 14/11/2026, 13:12:28 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeicvhzfn245gevzzy2snjozft7f7r75cyrlnwrbnacm6zdovsltj5m