Network-Decentralized Control with Collision Avoidance for Multi-Agent Systems

Abstract

An often preferred method to control a multi-agent system is by a network-decentralised controller. Network-decentralised means that each agent only has knowledge about its own state and the state of its neighbouring agents. In this thesis the multi-agent system consists of holonomic robots moving in a 2-dimensional configuration space. Each agent is equipped with a collision avoidance algorithm which does not need unmeasurable information from other agents, hence it is completely decentralised. The collision avoidance algorithm gen- erates a reference velocity which is always away and (counter-clockwise) around a possible collision. Cooperation of the agents is reached by estimating their position and moving in an assigned formation by only communicating with neighbouring agents, hence the strategy is network-decentralised. The communication network of the agents is modelled as a graph and described by the associated incidence matrix. This thesis combines for the first time the network-decentralised estimation method proposed by Giordano et al. [1] together with the network-decentralized control method proposed by Blanchini et al. [2] and a collision avoidance algorithm. When anonymity in coordination is possible the agents are able to switch target location. Each agent solves a local optimisation problem to minimise the total distance travelled by itself and its neighbouring agents. The agent with the highest cost sav- ings is allowed to reallocate the target locations. The results obtained in the simulations are very promising and suggest that the proposed algorithms can be successfully implemented to efficiently coordinate agents and avoid collisions in real-world applications.