Experimental study demonstrates high-precision vehicle tracking in autonomous platoons, indicating enhanced real-world safety and efficiency through physics-informed learning.
Key Points
Develop and evaluate a physics-informed, learning-augmented hierarchical control framework to overcome sensor noise, model uncertainties, and actuator nonlinearities in autonomous vehicle platooning.
Integrated a prediction-correction multi-sensor fusion module combining wheel odometry, inertial measurement unit (IMU), and GPS data for continuous state estimation.
Coupled longitudinal Model Predictive Control (MPC) and lateral feedforward-feedback control (LFFC) with an offline Conservative Q-Learning (CQL) parameter tuner and a neural residual network for error compensation.
Tested the framework on a physical vehicle platooning platform under real-world operating conditions.
Achieved a position root mean square error (RMSE) of 0.0781 m on the real-world platooning testbed.
Demonstrated a 30.66% improvement in linear velocity tracking compared to baseline controllers.