DEMO-PAST: A Decentralized Multi-MAV Online Navigation System using Parallel Strategy Acceleration

In this paper, a decentralized and asynchronous navigation system framework for multi Micro Aerial Vehicle (MAV) in unknown obstacle-rich scenarios is proposed. The developed framework strives to improve the performance of real-time by using parallel strategy on graphics processing units (GPU). In this way, we investigate parallel algorithms of map updating and trajectory planning. Specifically, map updating method based on spherical coordinate is initially presented. Then a hierarchical trajectory planning scheme is designed. Wherein the state lattice planning module as front-end is solved quickly in closed form, the trajectory optimization module based on model predictive path integral (MPPI) as back-end is utilized to refine trajectory. The optimization program is independent of gradient that might not exist when the cost function is composed of discontinuous components. Finally, the efficiency of the proposed method is validated by extensive numerical simulations and real-world experiments.