Prediction of surrounding vehicles accurately is an essential prerequisite for safe autonomous driving. Trajectory prediction methods can be classified into physics-, maneuver-, or learning-based methods. Learning-based methods have been studied extensively in recent years because it effectively exploits the road information and interactions among vehicles. However, learning-based methods perform poorly in unseen environments that were not considered during training and provide unreasonable results such as inconsistent trajectories according to road geometry. In this paper, to address this problem, a hybrid model that combines a learning-based model with physics- and maneuver-based models according to their uncertainties is proposed. The deep ensemble technique is also used to estimate the uncertainty of the learning-based method. Because the deep ensemble tends to show a large variance in unseen environments, this method is used to determine whether to use a hybrid model. The proposed method is trained and validated using the Lyft l5 dataset, the real environment vehicle driving data containing several types of intersections.

Accurately estimating the road slope and mass of a tractor-trailer is crucial from a control perspective as it significantly impacts the dynamic behavior. However, achieving accurate estimation of these variables poses a challenge, especially in scenarios involving mass variations due to loading on the semi-trailer. This difficulty arises from the need to appropriately decouple among the longitudinal acceleration, the acceleration due to road slope and acceleration sensor bias due to pitch angle. To solve this difficulty, this study proposes a road slope and mass estimation framework that accounts for changes in pitch angle. Firstly, a kinematic model is introduced including the longitudinal acceleration and the pitch angle. Secondly, a physically feasible neural network-based pitch angle estimator is designed to determine the pitching effect in the longitudinal acceleration sensor data. The input features are derived based on the pitch dynamics, and pitch bias is additionally introduced to consider the mass variation during weight loading/unloading. Then, a Kalman Filter (KF) is designed to estimate the longitudinal acceleration and slope angle with considering the estimated pitch angle. Finally, a Recursive Least Squares (RLS)-based mass estimation method is presented utilizing the estimated longitudinal acceleration and the estimated road slope. Outlier data are filtered using the Random Sample Consensus (RANSAC) algorithm to improve the performance of the RLS. The proposed estimation method is validated through simulations.

In this paper, a novel steering control algorithm for semi-trailer trucks is proposed, which effectively incorporates the vehicle’s dynamic characteristics and compensates for lateral motion delays. To model the tractor and trailer, both kinematic and dynamic models are derived and utilized to design a state estimator and a lateral control system. To achieve accurate trailer motion estimation, a novel Velocity-Dependent Kalman Filter (VDKF) is developed. Unlike the standard Kalman Filter, the VDKF adaptively adjusts measurement confidence according to vehicle speed, thereby improving hitch angle estimation across a wide range of operating conditions. Furthermore, an advanced steering controller is proposed that integrates trailer dynamics and a prediction-based compensation mechanism for lateral motion delay. By predicting the future states of the tractor and trailer, the controller mitigates the delay effect by incorporating these future states into the control system. The overall objective of this work is to enhance lateral stability and robust path tracking of semi-trailer trucks under varying payloads and steering delays. Simulation results demonstrate that the proposed controller reduces lateral deviation by up to 86.4% compared to Pure Pursuit (PP) and 53.0% compared to the Linear Quadratic Regulator (LQR), while maintaining robustness under payload variation and a 500 ms steering delay.

Driving on boulder field terrain induces severe wheel slip and rapidly varying, highly uneven vertical tire loads, making accurate force estimation essential for maintaining traction and mobility of off-road vehicles. However, direct measurement of tire vertical forces typically requires specialized sensors that are costly and difficult to integrate into practical vehicle systems, while conventional model-based estimation approaches often struggle to capture the complex tire–terrain interactions and nonlinear suspension dynamics encountered in off-road environments. This paper proposes a Pre-Layer-Normalization (Pre-LN) Transformer-based vertical tire force estimation method for off-road vehicles operating on boulder terrain. The proposed approach utilizes five vehicle chassis signals to learn the nonlinear relationship between vehicle motion and tire vertical forces, where a learnable positional encoding is employed to capture the irregular, impulsive temporal dependencies induced by boulder contact. A tire-specific decoupled output architecture with four independent MLP heads is adopted to handle the asymmetric loading conditions characteristic of discontinuous boulder contact, and a tire-weighted MSE loss is employed to promote balanced learning across all four tire force estimators. The proposed estimator is trained and validated using IPG CarMaker simulations with a BMW X5 model traversing boulder terrain at low speed. Simulation results demonstrate that the proposed method achieves an average RMSE of 41.2 N and an average MAPE of 0.76%, reducing the average RMSE by 91.3%, 89.2%, and 86.4% compared to the LSTM, TCN, and DARNN baselines, respectively, and offering accurate and robust vertical tire force estimation for traction management and torque distribution in off-road vehicles.