Vision-based control of multi-agent systems

Abstract

Scope and Methodology of Study: Creating systems with multiple autonomous vehicles places severe demands on the design of decision-making supervisors, cooperative control schemes, and communication strategies. In last years, several approaches have been developed in the literature. Most of them solve the vehicle coordination problem assuming some kind of communications between team members. However, communications make the group sensitive to failure and restrict the applicability of the controllers to teams of friendly robots. This dissertation deals with the problem of designing decentralized controllers that use just local sensor information to achieve some group goals.

Findings and Conclusions: This dissertation presents a decentralized architecture for vision-based stabilization of unmanned vehicles moving in formation. The architecture consists of two main components: (i) a vision system, and (ii) vision-based control algorithms. The vision system is capable of recognizing and localizing robots. It is a model-based scheme composed of three main components: image acquisition and processing, robot identification, and pose estimation.

Using vision information, we address the problem of stabilizing groups of mobile robots in leader- or two leader-follower formations. The strategies use relative pose between a robot and its designated leader or leaders to achieve formation objectives. Several leader-follower formation control algorithms, which ensure asymptotic coordinated motion, are described and compared. Lyapunov's stability theory-based analysis and numerical simulations in a realistic tridimensional environment show the stability properties of the control approaches.