Safe motion-planning with ​learning in the loop

This project continues our work within optimal motion planning where we combine methods from AI and optimal control into new more efficient methods compared to the individual ones being combined. The primary target tools to be considered here are reinforcement learning and generative AI/foundation models. The main research challenge is that these types of tools are not known to combine well with the desire to guarantee safety in safety-critical applications. There are several candidate approaches to approach this challenge, where one can, e.g., explicitly consider the uncertainty in the learned components and/or consider various forms of supervisory functionality.