Neural ODE and SDE Models for Adaptation and Planning in Model-Based Reinforcement Learning

TL;DR

This paper explores neural ODEs and SDEs for modeling stochastic dynamics in reinforcement learning, demonstrating improved policy performance and sample efficiency in complex environments, especially under partial observability.

Contribution

It introduces neural SDEs for better stochastic modeling and a latent SDE approach for partial observability, advancing model-based RL methods.

Findings

Neural SDEs better capture stochastic transition dynamics.

Latent SDE models improve policy performance in partial observability.

Proposed methods outperform or match existing approaches on benchmarks.

Abstract

We investigate neural ordinary and stochastic differential equations (neural ODEs and SDEs) to model stochastic dynamics in fully and partially observed environments within a model-based reinforcement learning (RL) framework. Through a sequence of simulations, we show that neural SDEs more effectively capture the inherent stochasticity of transition dynamics, enabling high-performing policies with improved sample efficiency in challenging scenarios. We leverage neural ODEs and SDEs for efficient policy adaptation to changes in environment dynamics via inverse models, requiring only limited interactions with the new environment. To address partial observability, we introduce a latent SDE model that combines an ODE with a GAN-trained stochastic component in latent space. Policies derived from this model provide a strong baseline, outperforming or matching general model-based and model-free approaches across stochastic continuous-control benchmarks. This work demonstrates the applicability of action-conditional latent SDEs for RL planning in environments with stochastic transitions. Our code is available at: https://github.com/ChaoHan-UoS/NeuralRL