OPTIMIZATION OF ROBOT MOTION PLANNING THROUGH MACHINE LEARNING-BASED HUMAN TRAJECTORY PREDICTION IN COLLABORATIVE ENVIRONMENTS

The integration of collaborative robots (cobots) into environments shared with humans has introduced a change in the field of industrial automation. Within this context, motion prediction and trajectory planning are fundamental to enable cooperation and compute potential collision risk during shared tasks. The aim of this thesis is to combine predictive models of human movement and strategies for the adaptation of the robot trajectory. To this end, the thesis exploits deep learning techniques, particularly neural networks trained on human joint trajectory data, to forecast short-term human motion. These predictions are then incorporated into a framework that minimizes the robot's trajectory execution time, while also including a penalty term based on the Artificial Potential Field in the objective function to account for proximity to predicted human motion.