基于可通行度估计的无人履带车辆路径规划

In response to the insufficient consideration of terrain features in the existing path planning methods, a path planning method based on the estimation of terrain traversability for unmanned tracked vehicles is proposed. In the proposed method, a ConvLSTM-based deep neural network is used to extract the spatial and temporal correlation features of LiDAR point clouds from continuous trajectories, fuse the vehicle motion state features, and estimate the terrain traversability. Based on the terrain traversability, the node expansion mode and cost function of A^* algorithm are improved, and the discrete waypoints that meet the collision-free constraints and the low traversability cost are output. A gradient-free iterative smoothing algorithm is used to reduce the cost of path relaxation and traversability. Then a cubic B-spline curve is used to generate a smooth reference path which is used to establish the Frenet coordinate system. A safety corridor based on the traversability is constructed in the Frenet coordinate system. On the premise of meeting the collision-free constraints and the low traversability cost, a smooth path meeting the vehicle kinematics constraints is generated in the corridor. The experimental results indicate that the proposed method can fully consider the terrain features and improve the stability of path planning results and the traversability of path.