Safe Exploration Method for Reinforcement Learning under Existence of Disturbance

TL;DR

This paper presents a safe exploration method for reinforcement learning that guarantees constraint satisfaction with high probability even under stochastic disturbances, using partial prior knowledge and theoretical safety guarantees.

Contribution

It introduces a novel safe exploration approach that explicitly accounts for disturbances and provides theoretical conditions for conservative input construction to ensure safety.

Findings

Guarantees safety with high probability under normal disturbance distributions

Validated the method through simulations on inverted pendulum and robot manipulator

Provides theoretical conditions for conservative input design

Abstract

Recent rapid developments in reinforcement learning algorithms have been giving us novel possibilities in many fields. However, due to their exploring property, we have to take the risk into consideration when we apply those algorithms to safety-critical problems especially in real environments. In this study, we deal with a safe exploration problem in reinforcement learning under the existence of disturbance. We define the safety during learning as satisfaction of the constraint conditions explicitly defined in terms of the state and propose a safe exploration method that uses partial prior knowledge of a controlled object and disturbance. The proposed method assures the satisfaction of the explicit state constraints with a pre-specified probability even if the controlled object is exposed to a stochastic disturbance following a normal distribution. As theoretical results, we introduce sufficient conditions to construct conservative inputs not containing an exploring aspect used in the proposed method and prove that the safety in the above explained sense is guaranteed with the proposed method. Furthermore, we illustrate the validity and effectiveness of the proposed method through numerical simulations of an inverted pendulum and a four-bar parallel link robot manipulator.