A Controlled Study of Models, Skills, Feedback, and Curriculum in Agentic Robotics
Abstract
A controlled simulation study of agentic robotics finds that successful self-learning depends less on information access than on how well agents interpret feedback, attribute failures, reason about spatial and geometric structure, and revise code to improve robot policies.
Agentic robotics asks coding agents to implement and iteratively improve robot policies from simulated and real-world physical interactions. We study the self-learning loop of agentic robotics through a controlled simulation study spanning nine manipulation tasks, 631 different experiment configurations, and 28K coding agent rollouts, systematically varying model capability, skill priors, feedback structure, and curriculum. The experiment data suggests that the bottleneck is often not the access to sufficient information, but interpreting it into the right policy update. Across these factors, a successful self-learning loop depends on how effectively agents attribute failures, reason task-relevant spatial and geometric structure, and selectively restructure code to improve policies.
Community
Thank you for sharing your work. Your controlled study varies model capability, skill priors, feedback, and curriculum in the agentic self-learning loop. Its analysis emphasizes that obtaining feedback and turning that feedback into an effective policy revision are distinct challenges.
We discuss your work in Section 7.2 (“Sources of improvement and comparable evidence”), p. 69 of the updated manuscript of our survey, In-Context Learning for Robots: Methods and Applications. We discuss these controlled factors when explaining how to attribute gains to contextual information, existing execution resources, or the revision procedure itself. Read the updated discussion.
If you find our discussion of evaluating agentic robot learning useful for your own work, we would be grateful if you would consider citing the survey.
Get this paper in your agent:
hf papers read 2611.09030 Don't have the latest CLI?
curl -LsSf https://hf.co/cli/install.sh | bash Models citing this paper 0
No model linking this paper
Datasets citing this paper 1
Spaces citing this paper 0
No Space linking this paper
Collections including this paper 0
No Collection including this paper