A Real-Time local path planning method based on SVM for UGV

Path planning is one of essentials of unmanned ground vehicle (UGV). For the case of poor lighting and weather, traditional vision based methods can not extract effective route boundaries to generate reasonable path stably in unstructured road. By taking advantage of distance-sensing technology (e.g. 64-beam LiDAR), th is paper proposes an efficient real-time path planning approach. In this approach, given grid map from 64-beam LiDAR, obstacles on both sides of the road are regarded as two classes fed to Support Vector Machine (SVM) to generate an initial safe path. During driving, a time weight based leas t square fitting is adopted to refine path from multiple safe paths which will be described by quartic polynomial, providing stable driving route. Combined with UGV's state, controls points from the refined path are adopted to generate the final path through Bezier curve fitting. Experiments on real UGV under different road scenario are conducted, showing that the proposed method can obtain stable and reasonable path with promising performance.