Embodied Turing Machines: Stateful Code for Robot Recursive Self-Improvement
Code-Only-as-Policy makes robot decisions in code, reaching 70.24% success on 42 RoboDojo tasks.
The paper frames robots as Embodied Turing Machines and proposes Code-Only-as-Policy (COAP), where code tracks robot, environment, and task state from camera images and proprioception and makes every decision. The same library is reused across episodes and tasks without a VLM or VLA in the loop. On RoboDojo's 42 bimanual tasks the library reaches 70.24% success with no model at test time, and the authors position it as a medium for recursive self-improvement and as a data engine for VLAs.
- COAP tracks robot, environment, and task state in code.
- No VLM or VLA runs in the control loop at test time.
- Shared library hits 70.24% success on 42 RoboDojo bimanual tasks.
- Coding agents can extend the library for recursive self-improvement.
- Products
- Code-Only-as-PolicyRoboDojo
From the original article
1 min read at huggingface.co
Most robot policies keep a model in the control loop: a VLA maps observations to actions, and an Agent Harness, such as Agent-as-Policy or Harness VLA queries a VLM for decision making at run time. We propose a different view: the embodied world is an Embodied Turing Machine, whose tape is the robot and environment state and rules are the policy.
By Kairui Hu, Siyuan Hu, Fangzhou Hong · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.