High-Precision Data-Mechanism-Driven Lateral Velocity Estimation Using Transfer Learning in Distributed Vehicles

Accurate lateral velocity estimation is vital for vehicle stability and motion control. Traditional methods based on kinematic or dynamic models often suffer from drift and parameter mismatch, while neural network (NN) approaches demand extensive, high-quality datasets. This article presents a high-precision, data-mechanism-driven method for lateral velocity estimation on limited labeled datasets, leveraging transfer learning within a distributed vehicle experimental platform (DVEP). A long short-term memory (LSTM)-based network integrates time-series vehicle state inputs with mechanism-based estimations, effectively compensating for model errors and improving accuracy. To ensure high precision across different steering modes in distributed vehicles, the network is strategically designed with a feature extraction layer for fine-tuning and a regression output layer for retraining. This structure enables fast adaptation of the estimation algorithm using minimal data. Validation on a DVEP demonstrates that the proposed method reduces RMSE by 58.57% (to 0.49 km/h) in four-wheel-steering mode compared with existing algorithms. Furthermore, with just 5 min of labeled data, the model successfully transfers from four-wheel to two-wheel steering with minimal accuracy loss. This approach not only enhances estimation accuracy but also significantly reduces data requirements and improves deployment efficiency across varied driving conditions. For the data collection, refer to <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://drive.google.com/file/d/12G-RfclCuqvmw34WZcza8_SF8BluZXLR/view?usp=sharing</uri>