Generalist AI controls robots better than specialist policies in manipulation tasks
On Agentic Manipulation Policies
RoboticsArtificial IntelligenceComputer Vision and Pattern RecognitionMachine Learning
Summary
Robot hands need plans to pick up and move things, and usually they learn from examples made for each task. This paper shows that a general-purpose AI model can directly control robot arms better than specialized methods without needing any examples or training for specific tasks. The AI learns its own ways to solve problems like finding hidden objects or using tools, and it even improves from experience when given memory. This approach works well on many different robots and tasks, making robot control more flexible and reliable.
What this means in practice
- •For robotics engineers: Control various robot arms and mobile manipulators without per-task training by using a generalist AI policy that adapts actions during execution.
- •For warehouse automation teams: Deploy robot manipulators capable of handling diverse tasks and unexpected scenarios more reliably without requiring custom demonstration data.
Authors
Chen Chen, Omid Bagherieh, Naveen Nikamanth Appiah Balaji, Jesse T. Gonzalez, Negin Heravi, Fahad Islam, Xueting Li, Jean-Baptiste Regli, Kejia Ren, Petr Švec, David J. Yoon, Matthias Müller, Vladlen Koltun
Abstract
Recent work on robot manipulation has focused on specialist policies trained on demonstrations. We show that generalist AI models can control manipulators more reliably than specialist policies such as vision-language-action models (VLAs) and world-action models (WAMs). We systematically analyze agentic manipulation policies (AMPs), in which a generalist AI model commands the manipulator directly, adapting its actions as the task unfolds. We present AMP1, a reference harness that runs unchanged across single-arm, bimanual, and mobile manipulation. AMP1 achieves state-of-the-art results on the RoboLab-120 and CaP-Gym benchmarks, with no demonstrations or task-specific training. On RoboLab-120, AMP1 reaches an 85.6\% success rate zero-shot, compared to 49.2\% and 42.9\% for leading VLA and WAM policies. The agent succeeds zero-shot in 104 of 108 real-world trials spanning twelve tasks and three robot instances, reducing failure rates by 20x and 15x relative to strong VLA and WAM baselines. The basic harness supplies no grasping, pushing, or placing skills; the agent develops its own strategies, such as searching for hidden objects, using objects as tools, and calibrating its cameras. With the addition of a persistent memory store, the agent can also learn from experience: an augmented harness with a memory representation that is carried across episodes lowers the failure rate on RoboDojo bimanual manipulation tasks by 4.6x relative to zero-shot deployment. Our analysis suggests that AMPs are substantially more general and robust than specialist manipulation policies, providing a promising foundation for robot manipulation.