A3RGB-D SLAM: Active RGB-D SLAM with Active Exploration, Adaptive TEB and Active Loop Closure

In this paper, we present an active RGB-D simultaneous localization and mapping (SLAM) method for mobile robots to obtain a collision-free trajectory with good localization performance in fast exploration. Specifically, a reinforcement learning based motion planning algorithm is proposed to dynamically reflect influences of the waypoint difference, obstacle avoidance, non-holonomic constraint, and total time in the Timed-Elastic-Band (TEB) algorithm in real-time. Then, a new active RGB-D SLAM method is presented. The octomap is generated from the visual depth point cloud and used for environmental representation. The selection strategy of frontiers considers both the information gain of exploration and reprojection errors of map points, and thus can guide the robot to explore the unknown environment efficiently with high localization accuracy. The effectiveness of the proposed A3RGB-D SLAM is verified by simulation comparisons.