Multi-Objective Trajectory Inference for Quadrotor Slung Payload System

Abstract

This paper presents a trajectory-planning method based on probabilistic inference for the quadrotor unmanned aerial vehicle (UAV) slung payload system. The trajectories of both the quadrotor and the payload are modeled as Gaussian processes (GPs), thereby suppressing redundant motion. An implicit dual-trajectory model ensures the smoothness of both the quadrotor and payload trajectories. The trajectory planning task is formulated as a probabilistic inference problem. Several likelihood factors are designed and integrated to address multiple aerial-transportation objectives. A real-time trajectory inference method is developed and validated through comprehensive simulations and real-world flight experiments. Experiments demonstrate real-time generation of safe and feasible trajectories with local perception and limited onboard computation. Comparative analysis reveals that the generated trajectories achieve greater smoothness, lower control effort, and better transport stability than existing approaches. Note to Practitioners—Quadrotor slung payload systems have considerable potential for diverse applications including logistics transport, disaster rescue, and agricultural protection. These complex scenarios require the system to satisfy multiple objectives, such as energy conservation, smooth transport, safe flight, and real-time planning. Current approaches often fail to simultaneously meet these diverse requirements. This work proposes a probabilistic trajectory planning method that effectively reduces control efforts and redundant motion while improving transport safety and stability at the planning level. Furthermore, practical experiments demonstrate its real-time planning capability under onboard computational constraints. This planning framework could be enhanced in the future by further incorporating likelihood functions or extended to multi-UAV systems for more challenging missions.