针对水面无人艇在复杂海况下的动态避障问题,引入基于LSTM的航迹预测方法,结合速度障碍法,实现复杂海况下的反应式动态避障。首先,构建航迹预测训练集,以船舶历史时刻的经度、纬度、航速、艏向作为模型的输入,下一时刻的经度和纬度作为模型的输出,构建基于LSTM的船舶航迹实时预测模型。基于航迹预测模型的输出,获得障碍船舶下一时刻的预测位置,作为速度障碍法的输入,构建扩展的速度障碍区域,根据代价函数计算最佳避障速度,从而实现无人艇的实时动态避障。实验结果表明,该方法使无人艇能够及时地避开动态障碍物并完成局部巡航功能,满足在水上复杂交通环境条件下水面无人艇的动态避障要求。

Aiming to solve the problem of dynamic collision avoidance of unmanned surface vehicle in complex sea conditions, a route prediction method based on RNN-LSTM is proposed, combining with the speed obstacle method, the reactive dynamic collision avoidance in complex sea conditions is realized. First,construct a training set for trajectory prediction, using the longitude, latitude, speed, and heading angle of the ship at historical moments as the input of the model, and the longitude and latitude at the next moment as the output of the model to construct a real-time prediction of the ship's trajectory based on LSTM model. Based on the output of the trajectory prediction model, the predicted position of the obstacle ship at the next moment is obtained, which is used as the input of the speed obstacle method to construct the expanded speed obstacle area, and the optimal collision avoidance speed is calculated according to the cost function to realize the dynamic avoidance function of the unmanned surface vehicle. Experiments on collision avoidance on real ships show that the dynamic collision avoidance method in this paper ensures that the unmanned boat can avoid dynamic obstacles in time and complete the local cruise function, meeting the dynamic collision avoidance requirements of unmanned surface vehicle under complicated water traffic conditions.