Robot software reuse boosts control success on new tasks
Learning and Transferring Closed-Loop Robot Software
RoboticsArtificial Intelligence
Summary
Designing robot control programs that rely on constant feedback is usually hard and time-consuming. The authors show that software improved by running on one set of tasks can be stored and reused to help robots learn how to handle new tasks. By combining past improved code with few example demonstrations and feedback from simulations, the robot policies get better results across various tasks. This approach helps robots perform new activities more successfully without extra computational complexity when running.
What this means in practice
- •For robotics software developers: Reuse and improve robot control programs built for previous tasks to speed up adapting policies for new robot challenges using simulation feedback.
- •For industrial automation engineers: Deploy closed-loop robot controllers that leverage archived policy improvements to reduce manual tuning when configuring robots for different manufacturing jobs.
Authors
So Kuroki, Yujin Tang
Abstract
Closed-loop robot policies require observation processing, state management, and situation-dependent branching, making them costly to design and tune manually. Although coding agents increasingly support control-code generation and optimization, it remains unclear whether implementations improved on source tasks also support policy acquisition for new tasks. We study this question by treating complete closed-loop implementations as reusable execution experience. For each source task, a coding agent generates policy code from a few successful demonstrations and iteratively improves it using simulation feedback. The validation-selected implementations are retained in a software archive. For new tasks, the agent generates and improves policies using archived implementations, target demonstrations, and execution feedback. The resulting policy is then frozen and executes without further model calls. Across four source tasks in RoboCasa, iterative optimization increases mean success from 28.3% to 64.2%. Across nine target tasks and three independent runs, mean success is 45.2% without references, 41.5% with initial source code, and 57.0% with optimized source code. Optimized references outperform initial references in all three runs on the nine-task average, with a mean gain of 15.6 percentage points. These results demonstrate the value of execution-improved software as a resource for acquiring new policies in this setting, although initial references remain better on two target tasks when averaged across runs.