Prof Tomonari Furukawa School of Mechanical and Manufacturing University of New South Wales Sydney 2052, New South Wales, Australia [email protected] Tomonari Furukawa is a Senior Lecturer at University of New South Wales (UNSW), Sydney, Australia. He received the B.Eng. in Mechanical Engineering from Waseda University, Japan, in 1990, the M.Eng. (Research) in Mechatronic Engineering from University of Sydney, Australia, in 1993 and Ph.D in Quantum Engineering and Systems Science from University of Tokyo, Japan, in 1996. He was an Assistant Professor (1995-1997) and Lecturer (1997-2000) at the University of Tokyo, and Research Fellow (2000-2002) at the University of Sydney before joining UNSW. His research work focuses on inverse analysis and optimisation methods in computational mechanics and robotics. He has published over 160 technical papers and won various early career research awards and paper awards including the most prestigious computational mechanics young investigator award from International Association for Computational Mechanics. Coordination of MAVs and UGVs for Information-theoretic Urban Search and Rescue

This talk presents an information-theoretic control (ITC) technique that coordinates Micro Aerial Vehicles (MAVs) and Unmanned Ground Vehicles (UGVs) for Urban Search and Rescue (USAR). USAR missions are concerned with the state estimation of various static and dynamic targets such as (i) victims to search for & rescue and (ii) enemies to capture or escape where their information is often partially available. The technique, unlike the traditional area coverage and tracking techniques, can utilize any available information including prior knowledge and empirical knowledge. ITC technique effectively estimates the target states in the form of probability density function (PDF). The use of a nonlinear recursive Bayesian estimator further enables the estimation of a non-Gaussian PDF of a nonlinear system. Thus the search, results in a highly non-Gaussian PDF due to the use of the negative observation likelihood, as well as the tracking. The independent nodewise computation in the nonlinear recursive Bayesian estimation (RBE) also allows its implementation into a parallel computer including the graphical processing unit, making the real-time RBE possible irrespective of the number of vehicles to coordinate. The preliminary numerical investigations show successful implementation through validation and verification as well as real-time performance even when the number of nodes used for RBE exceeds one million. The proposed technique was further used for the RBE by a team of rotary-wing MAV and UGVs each equipped with a GPS and a compass to identify its global state, a camera to detect a target and a wireless module to communicate with the ground station. Although the ground to search for a target was vast and thus made the number of nodes considerably large, the proposed technique could execute real-time RBE while the MAVs and UGVs were cooperatively observing the ground.