PI: Daniel Axehill (LiU). Co-PI: Ass. Prof. Johan Löfberg (LiU).
The proposed research considers motion planning problems in unstructured environments where there are non-negligible model uncertainties and disturbances present. There is a fundamental trade-off between performance and robustness of a motion plan. If the performance of a system is pushed to its limits in terms of input, state, and collision avoidance constraints, any discrepancy in the a priori knowledge could result in dangerous situations. To avoid this, a conservative plan is often computed. The proposed research aims at, in different ways, to incorporate more precise and up-to-date information about disturbances and to decrease uncertainties in the plan that often traditionally are handled by introducing conservativeness. The proposed research presents four different directions to provide a better combination of performance and safety: disturbance-parameterized motion primitives to explicitly take into account disturbances, realtime improvement to essentially convert an open-loop plan to a closed-loop policy, learning to eliminate systematic errors over time, and High Performance Computing (HPC) in the cloud to be able to consider advanced approaches at reasonable computation times. The proposed research is strongly connected to what is described under the ELLIIT focus theme 1 “Autonomous vehicles and robots” and is suitable to be investigated by a post-doc during the duration of two years.
Project number: A6
