Multi-Objective Optimization for Swarm Tracking Problem Using the CMA-ES Algorithm

This paper studies the optimization method for tuning model parameters in the swarm-tracking task. First, a discretetime and linearized agent dynamics model is given, and a swarm control model is designed according to the swarm trajectory tracking task. Second, the order parameters reflecting the tracking performance are designed according to the multiple goals of the task. A series of normalization functions are used on the given order parameters, and then a multi-objective unified fitness function is obtained. Third, the unified fitness function is optimized using the Covariance Matrix Adaptation Evolution Strategy (CMA-ES). The result of numerical simulation shows that the swarm can track the trajectory consistently without collision. Compared to the model with manually designed parameters, the optimized swarm model achieves higher velocity correlation and lower tracking error.