PI: Björn Olofsson, Lund University
co-PI: Erik Frisk, Linköping University
The research objective of this project is to advance proactive decision-making in multi-agent autonomous systems by leveraging and combining deep-learning based motion predictions and optimization-based motion planning. The project is motivated by the multi-modal uncertainties inherent in multi-agent scenarios, which are present when such systems operate in unstructured environments. The state of the art today does not perform at the level of human drivers or operators, e.g., in negotiating complex maneuvering situations. Complementing methodological developments for learning-based motion planning and decision-making with context aware- ness, a key question of the project is generality and transferability across different scenarios of datasets used for training of motion-prediction models. Generalization will also be essential when designing small, real-time capable, predictive models since the right dataset can be chosen to cover the scenarios of interest. The project builds on an established ELLIIT collaboration with joint supervision and publications over 15 years.
Project number: F10