This project wil investigate cutting-edge reinforcement learning algorithms, in which the focus concentrates on exploring a general framework for joint consideration of safety and efficiency, to advance the research development in the field of learning and control for both China and Germany.

The project aims to resolve the central and deep scientific challenges of deciding how to select valuable data and generating augmented data for efficient model training, how to optimally construct a risk-averse reward functions and how to address safety constraints in generating the control policies. The framework can then be applied to study the multi-agent coordination problem, that agents must cooperate in the context of achieving a global objective in an unknown environment, while respecting constraints on operating environment, information sensing, energy, and general mechanical limitations.
Another key goal is to experimentally validate the work using robotics platforms at both University of Science and Technology of China (USTC) and Technical University of Munich (TUM), which wil advance practical applications to end-users in industry.