Generative-Model Predictive Planning for Navigation in Partially Observable Environments

2026-06-17

AI Summary

This work addresses the challenge of autonomous navigation in partially observable environments, where limited perception and multimodal belief distributions complicate decision-making. The authors propose BeliefDiffusion, a novel framework that leverages diffusion models to explicitly represent multimodal belief distributions by generating diverse plausible environment configurations conditioned on observation history. These configurations are then aggregated to enable long-horizon path planning via model predictive control (MPC). By directly modeling belief uncertainty in a generative manner, BeliefDiffusion overcomes key limitations of conventional approaches in handling multimodal uncertainty and long-term planning. Experimental results in synthetic map environments demonstrate that BeliefDiffusion significantly outperforms model-free reinforcement learning and other generative methods, achieving state-of-the-art performance in both navigation success rate and path efficiency.

Abstract

Navigation in partially observable environments presents a significant challenge for autonomous agents, requiring effective decision-making with limited sensory information in unknown environments. Belief-based methods, particularly those using neural networks to approximate the belief space, often fail to capture the inherent multimodality of belief spaces, especially in high-dimensional cases with perceptual aliasing. While generative models present a compelling alternative, they typically require substantial data or expert demonstrations and lack explicit mechanisms for long-term planning. In this paper, we introduce BeliefDiffusion, a novel framework that combines the benefits of both generation and planning. BeliefDiffusion leverages diffusion models to explicitly characterize multimodal belief distributions and utilizes Model Predictive Control (MPC) to simultaneously plan ahead. It consists of two steps: (1) Imagining plausible environment configurations based on observation history and (2) Planning efficient navigation strategies across an aggregated configurations. Through extensive experiments in synthetic map environments, we demonstrate that BeliefDiffusion significantly outperforms both model-free reinforcement learning baselines and other generative approaches in navigation success rate and path efficiency. Our results validate that explicitly incorporating multimodal belief representations into planning enables more robust navigation in partially observable settings.

Problem

• partially observable environments
• multimodal belief
• navigation
• autonomous agents
• belief representation

Innovation

• diffusion models
• belief representation
• model predictive control
• partially observable environments
• multimodal belief