Unifying Map and Landmark Based Representations for Visual Navigation
This addresses the problem of robust visual navigation for autonomous systems, though it appears incremental as it builds on existing map and landmark approaches.
The paper tackles visual navigation by unifying map-based spatial reasoning with landmark-based plan execution, enabling efficient navigation in novel environments using only sparse registered images. It reports performance gains over competitive baselines in simulated real-world environments.
This works presents a formulation for visual navigation that unifies map based spatial reasoning and path planning, with landmark based robust plan execution in noisy environments. Our proposed formulation is learned from data and is thus able to leverage statistical regularities of the world. This allows it to efficiently navigate in novel environments given only a sparse set of registered images as input for building representations for space. Our formulation is based on three key ideas: a learned path planner that outputs path plans to reach the goal, a feature synthesis engine that predicts features for locations along the planned path, and a learned goal-driven closed loop controller that can follow plans given these synthesized features. We test our approach for goal-driven navigation in simulated real world environments and report performance gains over competitive baseline approaches.