Adaptive Distributed Cluster Flow Control for a Group of Autonomous Robots*

Abstract

Conventional multiagent control strategies rely either on pre-defined operational patterns, or complex inter-agent communications. While the former provides simple yet compelling results, the latter allows to achieve goals in complex environments with unpredictable disturbances. However, designing local networked control solutions for large-scale robotic systems is extremely challenging. A novel way to control complex systems by manipulating system clusters as sub-systems is proposed. The relation between sparsity and cluster flows is manifested in the new developed approach to estimate cluster structure of a group of robots based only on a few local observations of their states. Aiming to perform necessary measurements of all N agent states, the compressed sensing methodology is used to obtain ∼ logN randomized aggregated measurements, followed by the corresponding series of local voting protocols for the network to reach a common aggregated state with randomized weights. Thus, the cluster control synthesis by these compressed measurements is computationally efficient, and yet precise. Effectiveness of the proposed method is illustrated in numeric simulations, where adaptive observation feedback cluster control outperforms conventional strategies.