On Agentic Manipulation Policies
On Agentic Manipulation Policies
This paper presents AMP1, a generalist AI-based manipulation policy that outperforms specialist vision-language-action models (VLAs) and world-action models (WAMs) on RoboLab-120 and CaP-Gym benchmarks. AMP1 achieves an 85.6% zero-shot success rate on RoboLab-120 compared to 49.2% and 42.9% for leading baselines, succeeding in 104 of 108 real-world trials across twelve tasks and three robot instances without demonstrations or task-specific training. The system adapts its actions autonomously, developing strategies such as object search and tool use, and can further improve with a persistent memory store that reduces failure rates on RoboDojo bimanual tasks by 4.6x relative to zero-shot deployment.
The authors introduce AMP1, a generalist agentic manipulation policy harness that achieves state-of-the-art zero-shot performance on RoboLab-120 and CaP-Gym benchmarks without task-specific training or demonstrations, demonstrating robustness across single-arm, bimanual, and mobile configurations.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
failure rate reduction factor
20
View original evidence
LA 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 ad
Open source S4failure rate reduction factor
15
View original evidence
LA 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 ad
Open source S4failure rate reduction factor
4.6
View original evidence
dition 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 rob
Open source S5
- paper: not_reported
What remains unknown
Not established in the collected evidence: Environment, Control, Data origin.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
LA 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 ad
Open source S4
arXiv:2611.09892v1 Announce Type: new 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 comman
Open source S2
ds 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 V
Open source S3
dition 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 rob
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org