DeepSynth: Automata Synthesis for Automatic Task Segmentation in Deep Reinforcement Learning

TL;DR

DeepSynth introduces an automata synthesis method to automatically uncover task structure, significantly improving deep RL training efficiency in environments with sparse, non-Markovian rewards.

Contribution

It presents a novel automata synthesis algorithm that guides deep RL with interpretable high-level task structures, enhancing learning in complex environments.

Findings

Reduces policy training iterations by two orders of magnitude.

Improves scalability in high-dimensional, sparse reward settings.

Successfully applied to Atari's Montezuma's Revenge.

Abstract

This paper proposes DeepSynth, a method for effective training of deep Reinforcement Learning (RL) agents when the reward is sparse and non-Markovian, but at the same time progress towards the reward requires achieving an unknown sequence of high-level objectives. Our method employs a novel algorithm for synthesis of compact automata to uncover this sequential structure automatically. We synthesise a human-interpretable automaton from trace data collected by exploring the environment. The state space of the environment is then enriched with the synthesised automaton so that the generation of a control policy by deep RL is guided by the discovered structure encoded in the automaton. The proposed approach is able to cope with both high-dimensional, low-level features and unknown sparse non-Markovian rewards. We have evaluated DeepSynth's performance in a set of experiments that includes the Atari game Montezuma's Revenge. Compared to existing approaches, we obtain a reduction of two orders of magnitude in the number of iterations required for policy synthesis, and also a significant improvement in scalability.