Sustainable Transfer Learning for Adaptive Robot Skills

TL;DR

This paper explores policy transfer in robotic learning, demonstrating that adaptation techniques like fine-tuning enhance efficiency and generalization across different robots for peg-in-hole tasks.

Contribution

It introduces a transfer learning approach for robot skills that improves sample efficiency and sustainability through policy adaptation across platforms.

Findings

Zero-shot transfer has lower success rates and longer execution times.

Fine-tuning significantly improves performance with fewer training steps.

Policy transfer with adaptation enhances generalization and reduces retraining.

Abstract

Learning robot skills from scratch is often time-consuming, while reusing data promotes sustainability and improves sample efficiency. This study investigates policy transfer across different robotic platforms, focusing on peg-in-hole task using reinforcement learning (RL). Policy training is carried out on two different robots. Their policies are transferred and evaluated for zero-shot, fine-tuning, and training from scratch. Results indicate that zero-shot transfer leads to lower success rates and relatively longer task execution times, while fine-tuning significantly improves performance with fewer training time-steps. These findings highlight that policy transfer with adaptation techniques improves sample efficiency and generalization, reducing the need for extensive retraining and supporting sustainable robotic learning.