Visual Navigation

Visual navigation studies how robots and embodied agents use images, video, depth, or event-camera data to locate themselves, understand surroundings, and move toward goals while avoiding obstacles. Approaches range from mapping and geometric planning to vision-language policies, traversability estimation, and world models that predict future views and actions; systems are tested in settings such as Habitat and R2R as well as on real robots. Current work explores long-horizon planning, interaction with people and objects, adaptation across platforms, and efficient deployment on limited hardware. A central challenge is making success reliable under visual change and distribution shift, while evaluating safety, collisions, and real-world transfer rather than goal-reaching alone.