A Modular Multi-Robot Graph SLAM system for Indoor Environments

Abstract: Robots are playing an increasingly essential role in modern industry. SLAM (Simultaneous Localization and Mapping) is one of the critical technologies for robot to understand the environment and improve the automation level. In addition, multi-robot cooperation is becoming an interesting aspect to boost the efficiency of using robots. This thesis aims to study multi-agent SLAM, which is a cross application of SLAM and multi-robot system. Two methods, namely general grid map merging method and multi-agent graph SLAM method, are developed to achieve multi-agent SLAM. For the general grid map merging method, Cartographer is used to generate local map for a single robot and then the map merging is employed to put the local maps generated by individual robots together for achieving a global map. For the multi-agent graph SLAM method, robots will send all the sensor data to the center server and the server employs the GTSAM algorithm to generate and optimize the pose graph, based on which the global map is then generated. The two algorithms are implemented in ROS and are tested in Gazebo. Experiments are carried out to evaluate the performance of the two methods. The pose estimation accuracy obtained by the general grid map merging method is 2.45 cm on average, while that obtained by the multi-agent graph SLAM method is 11.86 cm on average with the application of GTSAM backend optimization. It has been proved that both solutions are feasible to achieve multi-agent SLAM.