Safe and efficient learning control for robotic manipulators in human-robot collaboration

Autonomous robotic manipulators that work and live alongside humans have the potential to transform our society and economy fundamentally. However, despite significant progress in robotics over the last few decades, such systems that can execute various complicated tasks with high performance while being near humans are still a long way from mainstream deployment. The main objective of this project is to investigate whether I can develop data-efficient learning-based constrained control algorithms for safe and fast complex Human-Robot Collaboration (HRC). I will develop a Learning-based Explicit Reference Governor (L-ERG) to guarantee safety and achieve high robot performance under model uncertainties, external disturbances, dynamic non-prehensile object transportation, and dynamic obstacles. I will study synergies with Learning from Demonstration (LfD) techniques to teach a robotic manipulator new skills by gathering trajectories from only a few demonstrations while improving the perceived robot safety with human-like repulsive obstacle behavior demonstrations, which will result in the LfD–ERG. I will combine the developed approaches in the LfD–L-ERG to teach a robotic manipulator to efficiently help a human with a collaborative task by physical interaction while guaranteeing safety. The project results will show its relevance for industrial and service applications and raise new research questions regarding safe learning control for physical HRC