Distributed Multi-Robot Learning using Particle Swarm Optimization

摘要: This thesis studies the automatic design and optimization of high-performing robust controllers for mobile robots using exclusively on-board resources. Due to often large parameter space noisy performance metrics, this constitutes an expensive problem. Population-based learning techniques have been proven be effective in dealing with noise are thus promising tools approach We focus research on Particle Swarm Optimization (PSO) algorithm, which, addition noise, allows a distributed implementation, speeding up process adding robustness failure individual agents. In thesis, we systematically analyze different variables that affect multi-robot obstacle avoidance benchmark. These include algorithmic parameters, controller architecture, testing environments. The analysis is performed experimental setups increasing evaluation time complexity: numerical benchmark functions, high-fidelity simulations, experiments real robots. Based analysis, apply PSO framework learn more complex, collaborative task: flocking. attempt task manner is, our knowledge, first such kind. addition, address problem evaluations encountered these robotic tasks present %new algorithm suitable resource-constrained due its low requirements terms memory limited local communication.