Midterm colloquium Daan van Wijk
25 June 2018 14:00 till 14:30 - Location: lecture room F, 3me - By: DCSC
"Collision Avoidance with Network Decentralized Control for a Multi-Agent system"
Recently the use of multiple agents to perform certain tasks is getting more and more popular but as the number of agents in a network increases, the time for a centralized controller to calculate all the trajectories of each agent increases drastically therefore network-decentralized control algorithms are needed.
This thesis show how the use of a network-decentralized controller can keep agents in a formation, prevent possible situations where a collision can occur and how this method prioritizes the team performance above the individual performance while steering the agent to its goal. To show the strengths of this decentralized controller a second order model of a homogeneous agent with a collision avoidance algorithm is introduced. The algorithm only needs the distance to the closest agent to perform a collision avoidance manoeuvre, the computation time for performing a collision avoidance manoeuvre does therefore not increase when the number of agents increase making it an interesting method when the system consists of many agents.
The agents only have local information and run a network-decentralized observer to estimate their absolute position and orientation in a two dimensional space. With several simulations it is shown that the collision avoidance algorithm along with network decentralized control is an effective method to steer the agents to its target location without collision and deadlocks while keeping formation. Finally a local optimization method is proposed to allocate the target locations of the agents so the sum of all the distances travelled is minimal.
Dr.Ing. G. Giordano