Optimizing Univector Field Navigation Parameters Using CMA-ES

A path planning algorithm that fits well in the context of IEEE Very Small Size Soccer league, a mobile robotics competition where cubic differential robots play a soccer match autonomously, is the univector field navigation. The implementation of this path planning method involves determining a series of adjustable parameters. Thus, this paper discusses the usage of Covariance Matrix Adaptation Evolution Strategy (CMA-ES) to optimize univector parameters. The contribution of this work is a methodology where optimization tasks are defined to determine a set of univector parameters, taking into account that two robots are trying to attack the ball simultaneously. At the same time, teammates, field walls, and opponents are obstacles that we must avoid. The results show a substantial improvement in the performance of a simulated team, which went from not scoring any goals in a 12-second long game simulation to scoring goals in an average time frame of 3.57 seconds. Also, the ITAndroids VSSS team won the Latin American Robotics Competition (LARC) 2020 using the obtained parameters.