A tree-based planner for active localisation: Applications to autonomous underwater vehicles

Autonomous Underwater Vehicle (AUV) are moving to a new phase with the development of light intervention systems. New vehicles will be equipped with lightweight manipulators and operate around subsea infrastructures. One of the key capabilities to safely perform such mission is robust and accurate autonomous localisation, i.e. the ability for the AUV to estimate correctly its position and orientation in the environment. Most of the current approaches to localisation are "passive", i.e., with no active control of the vehicle to improve localisation performances based on the current knowledge of the environment and the current estimate of the vehicle position. The "active" localization framework aims at incorporating the control of the robot motion in the localisation process by finding the best path to follow in order to reduce the uncertainty in the position state estimation. This paper aims at presenting a novel approach to the active localisation problem underwater using a priori maps of the environment or maps previously built using SLAM or mosaicing techniques. This is very relevant to the Trident project which aims at developing and demonstrating technologies for light intervention using an AUV. In the proposed framework, the position of the vehicle is estimated using Monte Carlo localization techniques (the state of the vehicle is represented by particles) and the motion of the vehicle is optmised to reach a single cluster of the particles (the vehicle knows where it is) by minimizing the expected entropy of the move. Both simulation results and tank trials showing the advantages of using this technique in realistic environments are presented here.