To enhance the ability of robots to effectively sharing their workspace with humans, this research activity focuses on proactively re-planning the robot trajectory in real time, also using predictions of the human motion, while guaranteeing safety. The implemented motion planning algorithms are based on recent developments in reinforcement learning and model predictive control (including learning-based and stochastic formulations), so as to tackle the inherent uncertainty of the human motion, at the same time guaranteeing the satisfaction of safety standards and taking care of the workers’ psychological well-being. The developed algorithms will be tested on actual robot manipulators Kinova Gen3 and Universal Robots UR5, available in our labs. As a related topic, we are working on robot teleoperation algorithms with obstacle avoidance also based on model predictive control and deep reinforcement learning.
This research activity is funded by Nazarbayev University collaborative research project "Stochastic and Learning-Based Predictive Control Methods for Physical Human-Robot Interaction'' (2020-2022), and is in collaboration with the NU Alaris Lab, the NU HRI Lab, National Laboratory Astana, and Baishev University of Aktobe.