Evolutionary Search of Optimal Hyperparameters for Learning Various Robot Manipulation Tasks

This paper presents a comprehensive study of robotic manipulation tasks, focusing on the movement planning and task-handling capabilities of robots. We introduce a novel approach that employs Dynamic Movement Primitives (DMP) for movement planning, coupled with an evolutionary algorithm, specifically the Genetic Algorithm (GA), for hyperparameter tuning of the DMP. Our method significantly enhances the system's precision and control, thereby facilitating a more accurate output. Furthermore, we conduct a comparative analysis of two optimization techniques - user-based and GA-based. Our findings indicate that the GA-based technique offers superior precision, underscoring its potential in advancing robotic manipulation tasks.