Distributed Asynchronous Algorithm for Collaborative Multi-UAV Multi-Target Tracking

The use of mobile sensor networks for multi-target tracking is an active research field. This paper describes the application of a novel two-layer relaxation for high-level control and coordination of UAVs to optimally track a set of moving targets considering a limited resources scenario. Using a singlebest tracking criterion, the scheme deploys distributed algorithms constrained to ensure full target coverage at all times. Quadrotor and fixed-wing UAV dynamics are considered, using respectively a point-mass and a Dubins car dynamical model. Our novel twolayer relaxation divides the problem into a distributed assignment problem and many local single-agent tracking problems. The proposed distributed algorithm requires low bandwidth messages and solves the assignment problem with an asynchronous-unreliable communication protocol between only neighbour agents. The tracking problems are formulated as a SDP problem for a quadrotor, while for a non-holonomic fixed-wing dynamics, a sequential convex programming algorithm is added to solve the linearised SDP problem iteratively. The proposed two-layer algorithms allow the use of heterogeneous teams and emergency manoeuvres to deal with collision avoidance and agents’ refuelling are implemented. Analysis proves that our relaxation optimises the same functional as the initial centralised formulation at steady-state, thus they have the same global optimum in a static targets scenario. The proposed distributed algorithms have linear time complexity with the number of targets and their complexity does not depend on the number of agents. Results show that our algorithms perform reasonable well in a wide range of scenarios. Keywords—Multi-target tracking, distributed robotics, multirobot collaboration, semi-definite programming.