A Hyperplane-Based Real-Time Whole-Body Planner for Multirotor Transportation Systems in Unknown Dynamic Environments

Abstract

Autonomous aerial transportation systems are increasingly deployed in real-world scenarios, where multirotors with suspended payloads offer mechanical simplicity and high agility while remaining challenging for motion planning. Existing whole-body planning approaches are typically computationally demanding, making it difficult to achieve safe, online trajectory generation in unknown environments populated with dynamic obstacles. Moreover, the suspended payload yields a time-varying geometry that is difficult to capture with compact collision avoidance constraints. To overcome this issue, a real-time whole-body planning framework is developed for safe navigation in unknown dynamic environments. The framework first employs a computationally efficient initial search scheme that explicitly accounts for whole-body collision avoidance and, when coupled with a lightweight trajectory prediction module for dynamic obstacles, produces initial trajectory candidates. On this basis, an optimization problem is formulated that jointly incorporates perception, collision avoidance, and dynamic feasibility, thereby enabling the online refinement of smooth, safe, and dynamically feasible trajectories. Extensive simulations and hardware experiments demonstrate that the proposed framework attains high success rates while generating high-quality feasible trajectories in such environments. Note to Practitioners—This paper addresses the challenge of safe motion planning for multirotor transportation systems in practical applications like aerial logistics and urban delivery. During actual flight operations among moving obstacles, the swinging payload creates a constantly changing physical geometry. Currently, the application of existing whole-body collision avoidance algorithms is often limited by high computational demands and the difficulty of dynamic obstacle prediction. Consequently, whole-body safety of the system is not easily guaranteed in dynamic environments. To resolve these practical issues, a real-time motion planning framework is proposed for onboard computation. The framework assumes that the cable remains taut throughout flight. First, an efficient search algorithm is utilized to generate initial trajectories, explicitly ensuring whole-body collision avoidance. Subsequently, a joint optimization formulation is applied to refine these trajectories by incorporating perception, collision avoidance, and dynamic feasibility constraints. For onboard real-time execution, the framework runs on an Intel NUC 13 Pro, using an Intel RealSense D455 depth camera for obstacle perception and a Livox Mid-360 LiDAR for state estimation, sustaining a replanning rate above $10{\,}\mathrm {Hz}$ . Extensive simulations and hardware experiments demonstrate the practical reliability of this method.