Efficient Multiagent Trajectory Optimization via Model-Algorithm Co-Design

Abstract

Jointly optimizing the trajectories of multiple agents is essential for the coordination and integration of complementary capabilities. However, multiagent trajectory optimization poses unique challenges. In addition to the increased dimensionality relative to single-agent problems, it introduces nonconvex inter-agent constraints arising from shared space and resources. These constraints fundamentally limit the efficiency in multiagent trajectory optimization. To address these limitations, we propose a model-algorithm co-design approach where we use novel problem formulation to improve algorithmic efficiency. In particular, by leveraging the duality between optimal control models and optimal estimation models, we simultaneously accelerate convergence and improve solution quality in optimizing multiagent trajectories within shared spaces. In addition, by leveraging the equivalence between disjunctive constraints and nonsmooth constraints, we improve the efficiency in optimizing the trajectories of UAV-UGV systems with energy-sharing constraints. Together, these results demonstrate how aligning modeling choices with algorithm development can significantly improve the efficiency and scalability of multiagent trajectory optimization.

Biography

Yue Yu is an assistant professor in the Department of Aerospace Engineering & Mechanics at University of Minnesota, Twin Cities. He obtained his PhD in Aeronautics & Astronautics from University of Washington in 2021. His research interests include optimal control, numerical optimization, and multiagent systems.