ROSARL: Reward-Only Safe Reinforcement Learning

TL;DR

This paper introduces a reward-only safe reinforcement learning method that determines an upper reward bound to minimize unsafe state reachability, using environment controllability and diameter, enabling safe policy learning without explicit penalty design.

Contribution

It proposes the concept of Minmax penalty for safe RL, providing a model-free algorithm to learn this bound during task learning, improving safety in high-dimensional environments.

Findings

The Minmax penalty effectively bounds unsafe state reachability.

The algorithm learns safe policies in complex continuous control tasks.

Using Minmax penalty accelerates safe policy convergence.

Abstract

An important problem in reinforcement learning is designing agents that learn to solve tasks safely in an environment. A common solution is for a human expert to define either a penalty in the reward function or a cost to be minimised when reaching unsafe states. However, this is non-trivial, since too small a penalty may lead to agents that reach unsafe states, while too large a penalty increases the time to convergence. Additionally, the difficulty in designing reward or cost functions can increase with the complexity of the problem. Hence, for a given environment with a given set of unsafe states, we are interested in finding the upper bound of rewards at unsafe states whose optimal policies minimise the probability of reaching those unsafe states, irrespective of task rewards. We refer to this exact upper bound as the "Minmax penalty", and show that it can be obtained by taking into account both the controllability and diameter of an environment. We provide a simple practical model-free algorithm for an agent to learn this Minmax penalty while learning the task policy, and demonstrate that using it leads to agents that learn safe policies in high-dimensional continuous control environments.