Answers are grounded in this paper and cite their sources.
Accurate tire-force estimation is essential for chassis stability control and motion planning in intelligent vehicles, especially under transient load transfer, combined-slip operation, and variations in tire-road friction. This paper presents a physics-guided tire-force estimator based on a Physics-Informed Neural Network with a Differentiable Friction Ellipse Projection (PINN-DFEP) layer, termed the PINN-DFEP estimator. The estimator predicts longitudinal, lateral, and vertical tire forces from vehicle and suspension-related signals. In the proposed architecture, the Magic Formula is used as a differentiable nominal shape-prior term to regularize the learned force-slip relationship, while the DFEP layer projects the raw longitudinal-lateral force estimates onto the admissible friction domain at the network output. This design embeds tire-mechanics priors into the learning process and enforces friction feasibility with respect to the supplied road-friction coefficient and boundary-defining vertical load. The estimator is trained using CarSim-MATLAB/Simulink co-simulation data from sinusoidal steering maneuvers and evaluated under unseen fishhook and J-turn maneuvers with prescribed Gaussian noise added to the corresponding input or measurement channels. Compared with the learning-based baselines and a physics-based UKF observer under the same test maneuvers and force references, the PINN-DFEP estimator achieves lower tire-force estimation errors while enforcing longitudinal-lateral feasibility with respect to the supplied friction boundary in the tested co-simulation scenarios.
No takes yet. Share an insight, caveat, or question.
Li et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: