Multimodal Perception based Autonomous Exploration with Active Camera Control in Unknown Environments

Autonomous exploration in the unknown environment is a fundamental problem of robot autonomy, however, it is always difficult to improve the efficiency of a robot to reconstruct its surroundings. In this paper, based on multimodal sensory fusion and active camera control, we proposed an active exploration algorithm that allows a robot highly efficiently reconstruct a complete map. First, we present a multimodal perception based map fusion algorithm to integrate 3D spatial information into 2D maps for better computational performance. Then, an active camera control algorithm is designed to take full advantage of the camera's mobility during exploration so that the robot can reduce the cost of exploration time and path length. Moreover, we establish an active exploration framework, which further leads to a significant improvement in map coverage and reconstruction efficiency. Compared with the traditional exploration algorithm and the passive fixed camera platform, simulation results show that the map coverage and the exploration efficiency are improved by 10-20% and 33% respectively, which demonstrates the advantages of our method.