ABSTRACT
The University of Western Australia's multi-robot system (MRS) comprises seven Pioneer 3AT-based outdoor robots. This chapter presents a decentralized multi-robot system for simultaneous localization and mapping (SLAM) in urban outdoor environments together with visual odometry and semantic segmentation techniques. Local SLAM is performed independently on each unmanned ground vehicle (UGV) whereby a single-robot SLAM algorithm builds its own submap by processing its sensor data, which is then broadcasted across the mesh network. The incorporation of semantic segmentation into the MRS enables navigation to be supplemented with scene understanding and object classification. Like visual odometry, semantic segmentation was also implemented in a decentralized approach onto individual UGVs. Static objects are stationary, while dynamic objects are moving. It is important for an multi-robot SLAM system to differentiate static and dynamic objects to devise proper navigational reactions to the environment. For an MRS to properly navigate an environment and perform cooperative tasks, these localization estimates need to be robust.
