Lightweight external safety enhancement strategy for UAV local trajectory planning in cluttered mapping environments
Abstract
Unmanned aerial vehicles (UAVs) are increasingly used in low-altitude surveying, mapping, and remote-sensing missions, where safe autonomous flight in cluttered environments is essential for reliable data acquisition. Gradient-based local planners such as EGO-Planner provide real-time obstacle avoidance, but in dense obstacles, narrow passages, and confined goal regions they may exhibit near-obstacle flight, insufficient safety margin, and local stagnation. Improving such planners usually requires modifying their cost functions or solvers, which is costly, planner-specific, and may introduce new failure modes. This paper proposes a lightweight external safety enhancement layer that improves flight safety without modifying the planner’s internal trajectory-optimization logic. The layer monitors the UAV state and obstacle proximity online, derives a normalized risk signal, and applies three risk-triggered interventions: a risk-aware safe-distance constraint with a proactive warning band, a risk-adaptive velocity modulation for high-risk regions, and a stall-recovery mechanism that issues temporary sub-goals to escape entrapment. Implemented as an independent Robot Operating System node, it interacts with the planner only through runtime parameter updates and goal re-publication, making it planner-agnostic and easy to deploy. Evaluated on identically evolved high-risk simulation scenarios with EGO-Planner as the planner under test, the enhanced planner raises the minimum obstacle clearance on the most challenging scenario from about 0.03 m to about 0.30 m and reduces the peak trajectory-point risk while preserving real-time performance. The results show that decoupling safety from the planner with an external, planner-agnostic layer effectively and transferably mitigates safetycritical behavior of local planners in cluttered mapping environments.