Cloud-based Mapping and Localization in Dynamic Warehouse Environments

Reliable, life-long mapping and localization is an essential component for mobile robotics in continuously changing warehouse environments. We present a system based on Cartographer in which robots run finite-history SLAM for low-latency localization and continuously stream local map updates to a cloud service. The cloud component assembles and optimizes a globally consistent pose graph out of the streaming data of the agents. Magazino piloted cloud-based Cartographer in a customer warehouse using its fleet of mobile picking robots. By sharing the local map changes among each other, the robots were able to maintain their localization accuracy while dealing with dynamic environments.

• Machine Perception

• Robotics