Bayesian Hilbert Maps for Continuous Occupancy Mapping in Dynamic Environments

Building accurate occupancy maps is crucial for autonomous vehicles to make path planning safer. Hilbert maps (HMs) are used for building such occupancy maps in a continuous fashion from depth sensors such as LiDAR in static environments. However, HMs are highly dependent on coefficients of the regularization term of the objective function which needs to be tuned heuristically. In this paper, we take a Bayesian approach, thus getting rid of the regularization term. Further, we extend the proposed model, Bayesian Hilbert maps (BHMs), to learn longterm occupancy maps in dynamic environments. Comparing with the state-of-the art techniques, experiments are conducted in environments with moving vehicles to demonstrate the robustness against occlusions as well as various aspects of building long-term occupancy maps. Video and code: https://goo.gl/5yvHT4