Abstract: A fundamental requirement of many multi—agent systems is the ability to intelligently coordinate so as to achieve some desired set of objectives. For instance, in mobile sensor networks featuring a restricted power budget, the ability to obtain some favorable configuration while minimizing expended energy, perhaps while actively tracking targets, is highly desirable. To endow such systems with these capabilities requires the team to plan, either collaboratively or individually, a time–based sequence of trajectories for each member based upon its current belief of the world state — ultimately allowing the group to obtain its goals while safely avoiding any environmental hazards.
Accordingly, the focus of this talk will be on developing optimal motion planning strategies and supporting technologies for enabling such systems. Towards this end, a suite of convex optimization strategies to facilitate both optimal formation changes and collaborative target tracking subject to performance guarantees will be presented. Regarding the former, the capabilities of second–order cone programming (SOCP) are exploited to offer computationally efficient centralized, distributed, and hierarchically decentralized solutions. Regarding the latter, a discrete—time, semi—definite programming (SDP) framework will be presented that guarantees each target is tracked by at least k team members while ensuring network connectivity remains intact across the robot formation. A relaxation for large—scale tracking scenarios is also considered. Supporting simulation and experimental results will be presented that highlight the utility of the optimization framework for real-time motion planning with systems comprised of 100’s of vehicles.