Mutual Alignment Transfer Learning

TL;DR

This paper introduces a mutual alignment transfer learning method that leverages simulation and real-world training to reduce sample complexity and improve robot policy performance by using reciprocal auxiliary rewards.

Contribution

It proposes a novel reciprocal alignment approach that enhances transfer learning between simulation and real robots, improving sample efficiency and policy robustness.

Findings

Reciprocal alignment improves real robot policy performance.

Simulation-guided exploration accelerates learning.

Reduced sample requirements for real robot training.

Abstract

Training robots for operation in the real world is a complex, time consuming and potentially expensive task. Despite significant success of reinforcement learning in games and simulations, research in real robot applications has not been able to match similar progress. While sample complexity can be reduced by training policies in simulation, such policies can perform sub-optimally on the real platform given imperfect calibration of model dynamics. We present an approach -- supplemental to fine tuning on the real robot -- to further benefit from parallel access to a simulator during training and reduce sample requirements on the real robot. The developed approach harnesses auxiliary rewards to guide the exploration for the real world agent based on the proficiency of the agent in simulation and vice versa. In this context, we demonstrate empirically that the reciprocal alignment for both agents provides further benefit as the agent in simulation can adjust to optimize its behaviour for states commonly visited by the real-world agent.