Variational Delayed Policy Optimization

TL;DR

This paper introduces Variational Delayed Policy Optimization (VDPO), a new RL framework that improves learning efficiency in environments with delayed observations by reformulating the problem as variational inference, reducing sample complexity.

Contribution

VDPO reformulates delayed RL as a variational inference problem, enabling more efficient learning through a two-step iterative process combining TD learning and behavior cloning.

Findings

Achieves comparable performance to SOTA methods.

Reduces sample complexity by approximately 50%.

Demonstrates effectiveness on MuJoCo benchmarks.

Abstract

In environments with delayed observation, state augmentation by including actions within the delay window is adopted to retrieve Markovian property to enable reinforcement learning (RL). However, state-of-the-art (SOTA) RL techniques with Temporal-Difference (TD) learning frameworks often suffer from learning inefficiency, due to the significant expansion of the augmented state space with the delay. To improve learning efficiency without sacrificing performance, this work introduces a novel framework called Variational Delayed Policy Optimization (VDPO), which reformulates delayed RL as a variational inference problem. This problem is further modelled as a two-step iterative optimization problem, where the first step is TD learning in the delay-free environment with a small state space, and the second step is behaviour cloning which can be addressed much more efficiently than TD learning. We not only provide a theoretical analysis of VDPO in terms of sample complexity and performance, but also empirically demonstrate that VDPO can achieve consistent performance with SOTA methods, with a significant enhancement of sample efficiency (approximately 50\% less amount of samples) in the MuJoCo benchmark.