RAPA-Planner: Robust and Efficient Motion Planning for Quadrotors Based on Parallel RA-MPPI

Motion planning for quadrotors in cluttered environments is an open problem, due to the limited onboard computational power and lack of risk assessment. To this end, an urgent need exists for an efficient and robust strategy that can generate safe motions. In this article, we develop a risk-aware parallel planner that is capable of quickly finding a safe trajectory with high smoothness and aggressiveness. Specifically, kinodynamic feasible motions are sampled in state space and connected to generate the guiding trajectory, which keeps topological consistency with execution trajectories simultaneously and can be solved quickly in closed-form. Then, a gradient-free and robust optimization scheme is presented to refine the trajectory based on risk-aware model predictive path integral (RA-MPPI). It incorporates the conditional value-at-risk (CVaR) as a metric to measure the risk level, which can estimate potential risks (such as collision and exceeding dynamic limits) automatically according to specified CVaR. Furthermore, the RA-MPPI can be solved using Monte-Carlo sampling, which avoids gradient calculation and can be deployed on GPUs in parallel, thereby improving efficiency. Benchmark results in simulation demonstrate that the proposed method outperforms others in terms of trajectory length and achieves a favorable performance with respect to efficiency and trajectory smoothness. Moreover, the effectiveness of the presented approach is further validated through real-world experiments.