GPU Accelerated Convex Approximations for Fast Multi-Agent Trajectory Optimization

In this letter, we present a computationally efficient trajectory optimizer that can exploit GPUs to jointly compute trajectories of tens agents in under second. At the heart our is novel reformulation non-convex collision avoidance constraints reduces core computation each iteration large scale, convex, unconstrained Quadratic Program (QP). Importantly, QP structure requires us associated matrix factorization/inverse only once for fixed number agents. Moreover, do it offline and then use same different problem instances. This further simplifies solution process, effectively reducing few matrix-vector products. For agents, be trivially accelerated on using existing off-the-shelf libraries. We validate optimizer's performance challenging benchmarks show substantial improvement over state art time quality.