Research On Trajectory Planning Of Unmanned Off-road Vehicles

Off-road unmanned vehicles have significant advantages in military,mining and other application scenarios.Most of the existing unmanned driving research is aimed at structured traffic environment.There is a big difference between trajectory planning in off-road environment and structured road trajectory planning.In the unstructured off-road environment,vehicles need to consider road quality and material,and need to predict the risk of different road areas.Based on this information,an optimized driving trajectory is obtained in the global region,and local obstacle avoidance is performed on the dynamic obstacles encountered during driving.The research contents of this paper are as follows:(1)Environmental image recognition and classification based on neural network.By establishing a classification network model,different risk coefficients are given to areas with different attributes of the environment,and finally pooled to form a grid map.In this paper,the elevation element is added to the local planning map,and the lateral slope and longitudinal slope information in the vehicle coordinate system are extracted by using the elevation element.On the one hand,it is used to improve the comfort of vehicle operation,on the other hand,it is used to predict whether the vehicle will be dangerous.The danger of driving in the off-road environment comes from collision and vehicle rollover.(2)Design global planning algorithm.In the trajectory planning,the trajectory generation method of model prediction is used,that is,based on the optimized trajectory generation method,the BSGF solution method is used for the optimization problem.In the global planning,in order to avoid the planning falling into local optimum in the narrow area,the sampling method of random state points is used,and finally the global trajectory is searched by combining the optimization principle of A~*.The cost function in the A~*algorithm is evaluated by projecting the sampling trajectory onto a two-dimensional global risk map.(3)Design local planning algorithm.The local trajectory is based on the global trajectory,which ensures that the local trajectory generation has a guiding direction.Therefore,a fixed area sampling method is used to improve the stability of the planning algorithm.In the local planning,the trajectory is mapped to the spatial surface to consider the influence of road slope,vehicle geometric model and vehicle kinematics.(4)In order to ensure the practicability of the algorithm,the experimental design is carried out.Using ROS(Robot Operating System)as the software development platform,the algorithm function packages such as perception,positioning,and control that are not designed in this paper are referenced,and finally transplanted to the computing platform of Xavier ARM64.The unstructured plane experimental site and off-road experimental site were designed.The final experiment proves that the environment recognition through neural network can be used for trajectory planning.The method based on model prediction trajectory generation can obtain a feasible trajectory that meets the constraints of the vehicle model,and local trajectory planning can achieve dynamic and static obstacle avoidance and meet safety and real-time requirements.