Deep Longitudinal Dynamic Model for Electric Semi-Trailer Trucks with Varying Load

For the autonomous driving of semi-trailer trucks, the main challenge is the model uncertainty caused by the vehicle structure complexity and working condition variation. This paper concentrates on the longitudinal dynamic modeling problem, which describes the longitudinal motion characteristics of a varying load semi-trailer truck. A deep neural network based method is proposed to deal with the underfitting problem of conventional empirical modeling methods. The model is designed according to the physical structure, including driving and braking subsystems, using engine and braking throttle signals as input. The output of subsystems are integrated to predict the acceleration and velocity. This model is time-varying w.r.t. vehicle states, which are encoded as the network features to describe the coupling effect between lateral and longitudinal dynamics. Different working load is considered explicitly by the mass ratio in the model and is simultaneously identified in the training and inference process. Experimental results are provided to show the effectiveness of the proposed model using real driving data.