Multi-sensor based attitude prediction for agricultural vehicles

Agricultural vehicles need active attitude prediction to prevent rollover and maintain safe operations on unstructured terrains. This paper proposes a multi-sensor based approach to the attitude prediction of agricultural vehicles. Firstly, the unstructured terrain information in front of agricultural vehicle is obtained by the embedded multi-sensor system such as LIDAR, IMU and encoders. Secondly, a combination of Genetic Algorithm and BP neural network (GA-BP) is proposed to predict the future position of the agricultural vehicle. Both the unstructured terrain information and the predicted vehicle position are used to calculate the tire grounding points of agricultural vehicle on the unstructured terrain. Finally, the tire grounding points are combined with vehicle geometry model to predict the vehicle attitude (pitch and roll angles) for the stability control of the vehicle. A prototype vehicle is deployed to conduct experiments in order to demonstrate the feasibility and performance of the proposed approach.