Skill-Space Shooting for Autonomous Robot Policy Improvement
Paper introduces skill-space shooting, using foundation model guidance over reusable skills to turn robot failure corrections into autonomous policy improvement.
Researchers propose skill-space shooting, a method that uses foundation model guidance to explore corrections as reusable short behaviors (skills) and converts successful trials into learnable policy improvements. This reduces reliance on human demonstration for correcting robot policy failures. Real-world experiments show repeated improvement in policies acting autonomously, and skills can be shared across tasks to reduce teaching effort on new tasks.
- Corrections framed as recurring short skills that foundation models can reason about
- Successful trials become learnable corrections for the task policy
- Real-world experiments show autonomous policy improvement within and across tasks
From the original article
1 min read at arxiv.orgRobots 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.
By Zihang Rui · arxiv.org
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.