Multi-agent autonomous target tracking using distance-based formations

Abstract: In this th​esis, the problem of cooperative control of a system of autonomous agents for target tracking is studied, with special focus on autonomous surface vehicles and minimizing the energy usage in the system. By separating the control problem into two different parts, we independently calculate the coordination scheme of the system using virtual agents with single integrator dynamics and design the control of the actual agents to follow the virtual.

Two different control strategies are introduced for the coordination scheme with the goal of making the agents form a regular polygon formation around the target. The first strategy is based on gradient descent for three agents. A proof of almost global stability for the system is provided, both with and without the target included. As long as the initial positions of the agents and target are not collinear, the system will asymptotically converge to the desired formation. The second strategy uses a rigid body formation for the virtual agents, where the most cost efficient pose is found using assignment and least squares minimization.

The viability of the control strategies are shown through simulations and with implementation in a real system of small autonomous surface vehicles. Successful experiments are presented from a setting where the target can be both stationary and moving. The results from the thesis will be used and further developed in a project centered around autonomous tracking of underwater targets, carried out in collaboration between KTH, NTNU Trondheim and LSTS Porto.