RoamFlow: Reinforcement-Aligned One-Step Action MeanFlow Policy for Image-Goal Navigation

2026-06-29

AI Summary

This work addresses the challenge of suboptimal trajectories in image-based object navigation caused by difficulties in modeling long-term dependencies. To this end, we propose RoamFlow, a novel framework that introduces MeanFlow to this task for the first time, enabling efficient trajectory synthesis and single-step action prediction through the generation of an average velocity field. The method employs a two-stage training strategy: an initial policy is first acquired via expert imitation learning and subsequently refined through reinforcement learning for task-specific optimization. Evaluated on both the Habitat simulation platform and real robotic systems, RoamFlow demonstrates low-latency inference and strong real-time navigation performance, significantly enhancing policy stability and overall navigation efficacy.

Abstract

Image-goal navigation is a key challenge in embodied robotics, where an agent must reach a target specified solely by a goal image. While existing reinforcement learning approaches map perceptual observations directly to actions, they struggle to model long-horizon dependencies, often leading to suboptimal trajectories. To address this limitation, we propose RoamFlow, a generative navigation framework that leverages MeanFlow to predict the average velocity field for trajectory synthesis, enabling efficient few-step generation and reducing inference latency. We further adopt a two-stage training strategy that combines expert imitation for stable initialization with reinforcement learning for task-specific policy refinement. Extensive experiments in both Habitat simulation and real-world robotic platforms demonstrate that RoamFlow achieves efficient inference while maintaining strong navigation performance under real-time constraints.

Problem

• image-goal navigation
• embodied robotics
• long-horizon dependencies
• trajectory optimization
• real-time navigation

Innovation

• MeanFlow
• image-goal navigation
• reinforcement learning
• trajectory synthesis
• two-stage training