Abstract:Autonomous navigation for intelligent mobile robots has gained significant attention, with a focus on enabling robots to generate reliable policies based on maintenance of spatial memory. In this paper, we propose a learning-based visual navigation pipeline that uses topological maps as memory configurations. We introduce a unique online topology construction approach that fuses odometry pose estimation and perceptual similarity estimation. This tackles the issues of topological node redundancy and incorrect edge connections, which stem from the distribution gap between the spatial and perceptual domains. Furthermore, we propose a differentiable graph extraction structure, the topology multi-factor transformer (TMFT). This structure utilizes graph neural networks to integrate global memory and incorporates a multi-factor attention mechanism to underscore elements closely related to relevant target cues for policy generation. Results from photorealistic simulations on image-goal navigation tasks highlight the superior navigation performance of our proposed pipeline compared to existing memory structures. Comprehensive validation through behavior visualization, interpretability tests, and real-world deployment further underscore the adaptability and efficacy of our method.

Cognitive Navigation for Intelligent Mobile Robots: A Learning-Based Approach with Topological Memory Configuration