Safe motion-planning with ​learning in the loop

This project advances our work in optimal motion planning by integrating methods from AI and optimal control into more efficient and powerful approaches than those achieved individually. The primary tools to be integrated with the current state-of-the-art optimal control and motion planning algorithms are reinforcement learning and generative AI/foundation models. Although these AI tools often yield impressive results, they typically do not align well with the need to ensure safety in safety-critical applications. Achieving such impressive results while simultaneously ensuring some level of guaranteed performance and safety is the main research challenge of this project. Several candidate approaches exist to address this challenge, such as explicitly accounting for the uncertainty in the learned components and/or incorporating various forms of supervisory functionality. The main industrial partner in this project is Scania, with special focus on their autonomous heavy-vehicle application. However, the research is of significant general importance with numerous applications. For example, motion planning for drones performing advanced maneuvers is another interesting target application that might be considered in the project.