A Parallel Iterative Linear Quadratic Controller for Autonomous Trajectory Optimization

In this article, we introduce a novel trajectory optimization approach for autonomous robots, which is essentially a parallelized constrained Iterative Linear Quadratic Regulator (ILQR). Specifically, we utilize parallel computing techniques to break down the sequential dynamic programming procedure of ILQR into parallel computing units, which improves the temporal complexity of the algorithm from $\mathcal{O}(N)$ to $\mathcal{O}(\log N)$. Simulation and real world experiments are conducted to validate the effectiveness and efficiency of our approach. Compared with the constrained ILQR without parallel computing techniques, we demonstrate remarkable computation efficiency improvements using the proposed approach, with different problem settings.