Effective Map Merging for Long-Term Autonomous Navigation

Abstract

An algorithm is presented that intelligently merges multiple 3D point clouds used for localization based on when the point cloud was recorded to create an updated map that is more similar to the current environment. The algorithm was implemented on a Boston Dynamics Spot robot and was used to upgrade Spot’s autonomous navigation algorithm called Autowalk by adding the capability for long-term navigation in semi-static environments. The proposed algorithm was validated by having Spot navigate both indoor and outdoor environments over multiple months traveling over 43 km autonomously without losing localization. The proposed method extends the life of programmed autonomous missions to ensure a robot can be used over extended periods of time without the need to re-teach these autonomous missions due to changes in the environment.