Distributed cooperative visual odometry for planetary exploration rovers

Navigation is a basic skill for planetary exploration robots. In the last years, planetary robotics navigation has become an important research field that includes all the robot capabilities such as perception, localisation, and mapping. This paper provides a novel algorithm to cooperatively localise a distributed team of robots on another planet when the initial pose of the robots is unknown. To perform this, a single Visual Odometry (VO) algorithm based on the conventional KLT feature tracker is employed to localise each single exploration rover in a planetary environment while detecting loop closures. The trajectory of the robots is described by a pose graph form in the designed Cooperative VO (CVO) algorithm. Detecting loop closure from the previously observed areas by the robots leads to triggering the optimisation process and improving the accuracy of the localisation. Accordingly, a planetary analogue real dataset is used to investigate the accuracy of the proposed algorithm. The superiority of the proposed distributed CVO is proved by comparing the obtained results with the single VO algorithm.