Paper Overview
Field: Machine Learning Authors: Zihang Rui, Renhao Wang, Haoxu Huang, Yang Gao Published: 2026-09-29 arXiv: 2609.38178
Summary
Robots deployed in the physical world must be able to improve beyond their initial training as they encounter new situations and failures. For this improvement to scale across tasks, it must make effective use of experience without requiring human demonstration of each correction. Recent agentic systems offer a way to reduce this reliance on human effort by using foundation models to autonomously compose learned behaviors to complete tasks. Yet completing tasks this way does not itself teach a task policy to overcome its own failures; that requires turning these behaviors into learnable corrections for the policy.
The authors' insight is that many such corrections are familiar short behaviors, or skills: they recur across tasks and describe actions that foundation models can reason about from a scene.
Method
The proposed skill-space shooting approach:
- Uses foundation models to guide exploration of policy corrections through a space of reusable skills
- Converts successful trial corrections into learnable updates for the task policy
- Enables skills to be shared across tasks, reducing the teaching required to improve on new tasks
Results
Real-world experiments demonstrate that policies achieve repeated improvement during autonomous operation. By using reusable skills as a source of corrective supervision, skill-space shooting delivers scalable, generalizable policy improvement both within tasks and across tasks.
More results and videos are available on the project page.
---
*Auto-collected on 2026-10-01*