Distributed Receding Horizon Control of Multiagent Systems

Citation

Dunbar, William Bruce (2004) Distributed Receding Horizon Control of Multiagent Systems. Dissertation (Ph.D.), California Institute of Technology. doi:10.7907/1N74-MZ62. https://resolver.caltech.edu/CaltechETD:etd-05282004-170123

Abstract

Multiagent systems arise in several domains of engineering. Examples include arrays of mobile sensor networks for aggregate imagery, autonomous highways, and formations of unmanned aerial vehicles. In these contexts, agents are governed by vehicle dynamics and often constraints, and the control objective is achieved by cooperation. Cooperation refers to the agreement of the agents to 1) have a common objective with neighboring agents, with the objective typically decided offline, and 2) share information online to realize the objective. To be viable, the control approach for multiagent systems should be distributed, for autonomy of the individual agents and for scalability and improved tractability over centralized approaches.

Optimization-based techniques are suited to multiagent problems, in that such techniques can admit very general objectives. Receding horizon control is an optimization-based approach that is applicable when dynamics and constraints on the system are present. Several researchers have recently explored the use of receding horizon control to achieve multi-vehicle objectives. In most cases, the common objective is formulated, and the resulting control law implemented, in a centralized way.

This dissertation provides a distributed implementation of receding horizon control with guaranteed convergence and performance comparable to a centralized implementation. To begin with, agents are presumed to be individually governed by heterogeneous dynamics, modelled by a nonlinear ordinary differential equation. Coupling between agents occurs in a generic quadratic cost function of a single optimal control problem. The distributed implementation is generated by decomposition of the single optimal control problem into local problems, and the inclusion of local compatibility constraints in each local problem. The coordination requirements are globally synchronous timing and local information exchanges between neighboring agents. For sufficiently fast update times, the distributed implementation is proven to be asymptotically stabilizing. Extensions for handling inter-agent coupling constraints and partially synchronous timing are also explored. The venue of multi-vehicle formation stabilization demonstrates the efficacy of the implementation in numerical experiments. Given the generality of the receding horizon control mechanism, there is great potential for the implementation presented here in dynamic and constrained distributed systems.

Thesis Files