Narrow Results

Climbing robot, flying robot or human wearable devices usually execute daily tasks in a pre-defined workspace sharing with humans. Long-term operation for these robots posts three challenge: 3D pose estimation, cost limitation and unexpected low dynamics. To address these challenges, we propose a solution for performing multi-session SLAM using a RGBD sensor. The main model is a multi-session pose graph, which evolves over the multiple visits of the workspace. When the robot explores the new areas, its poses will be added to the graph. The poses in the graph will be pruned if their corresponding 3D point scans are out of date. Thus the scans corresponding to the poses kept in the current graph will always give a map of the latest environment. To detect the changes of the environment, an out-of-dated scans identification module is proposed. Pruning of the poses also decreases the computational burden in graph optimization. Experimental results using real world data acquired by a Kinect sensor show that the proposed framework is able to manage the map in date for low dynamic environments with a reduction in complexity and an acceptable error level compared to the method saving all poses.