Control-Tutored Reinforcement Learning: an application to the Herding Problem

TL;DR

This paper introduces a control-tutored Q-learning method designed for safe, model-based reinforcement learning in continuous spaces, demonstrated through a multi-agent herding control problem.

Contribution

It presents a novel control-tutored Q-learning approach (CTQL) for safe reinforcement learning in continuous state spaces, applied to multi-agent herding.

Findings

Effective in multi-agent herding scenarios

Advances safe RL with control-tutoring in continuous spaces

Demonstrates potential for complex multi-agent control

Abstract

In this extended abstract we introduce a novel control-tutored Q-learning approach (CTQL) as part of the ongoing effort in developing model-based and safe RL for continuous state spaces. We validate our approach by applying it to a challenging multi-agent herding control problem.