GOEN: Guided Obstacle Endpoint Navigation for Real-Time Collision-Free Path Planning in Unstructured Environments

Abstract

We present GOEN, an advanced navigation and path planning framework specifically engineered to tackle the complexities of dynamic and unstructured environments through real-time 3D pointcloud processing. Our approach integrates pointcloud downsampling, collision risk assessment, and obstacle endpoint extraction to generate intermediate waypoints. These waypoints are refined iteratively through multi-stage safety validation and cubic-spline interpolation, resulting in kinematically feasible and collision-free trajectories. The system exhibits minimal computational overhead, achieving a planning latency below 10 milliseconds, thereby demonstrating suitability for deployment in resource-constrained scenarios. Extensive empirical evaluations in simulated environments and on quadruped robots demonstrate the framework’s robustness in dynamically identifying optimal navigation paths across unstructured terrains. Compared with baseline navigation methods, it has significant advantages in indicators representing navigation performance. Experimental validation confirms GOEN’s capability to balance trajectory optimality, safety constraints, and real-time responsiveness, providing an innovative solution for autonomous navigation in complex environments.