Real-time convex-geometric obstacle avoidance for UAVs via sub-target guidance

초록

This paper presents a hierarchical geometric obstacle avoidance framework based on convex safe half-spaces for unmanned aerial vehicles (UAVs) that aims to improve real-time feasibility and computational predictability in locally cluttered environments. Unlike traditional optimization-based methods that directly embed nonconvex obstacle-distance constraints into trajectory optimization, the proposed approach separates obstacle avoidance into a boundary-based geometric sub-target generation stage and an MPC-based desired profile generation stage. A spherical safety region is defined around the UAV, and hyperplanes derived from the relative positions of surrounding obstacles define convex safe half-spaces. On the boundary of the safety region, intermediate target candidates are geometrically extracted, and a sub-target is selected based on a cost that prioritizes alignment with the global mission direction while maintaining safe separation. The boundary-based candidate extraction involves a nonconvex spherical equality constraint. However, this nonconvexity is confined to the lowdimensional sub-target generation stage rather than being directly imposed on the MPC trajectory optimization problem. The selected sub-target is then tracked using an MPC-based planner with obstacle-related penalty terms in the cost function. Since the number of detected obstacles does not directly increase the MPC decisionvector dimension or prediction model structure, while the preprocessing and penalty-evaluation steps may still depend on the number of detected obstacles, the framework maintains predictable computation times in the tested static-obstacle scenarios. Simulation results show that the proposed method achieves computation times below the MPC sampling requirement and smooth obstacle avoidance, indicating its potential for UAV navigation in locally cluttered environments with multiple static obstacles.

키워드