Experimental assessment of neural networks for tire-road force estimation

Modern vehicle dynamics control systems may leverage the knowledge of tire-road forces to improve safety and performance. However, the lack of commercial sensors entails the need for suitable real-time estimation techniques. Beside classical model-based estimation, neural networks represent a promising solution in terms of achievable accuracy and computational burden. In this paper, we propose an experimental assessment of such data-driven modeling tools for longitudinal, lateral, and vertical forces estimation. A sensor layout typically available on high-end vehicles is considered, while the dataset for training and validation is composed of a few challenging maneuvers. As a key contribution, we provide some practical guidelines on how to select the set of sensors for each estimation problem.