Weight Adaptive Path Tracking Control for Autonomous Vehicles Based on PSO-BP Neural Network.

Area of Science:

• Autonomous Systems
• Control Theory
• Artificial Intelligence

Background:

• Autonomous vehicles require robust tracking control adaptable to diverse driving conditions like varying speeds and road curvatures.
• Classical Model Predictive Control (MPC) offers high-precision tracking but can struggle with dynamic environmental changes.

Purpose of the Study:

• To develop an adaptive model predictive control (AMPC) system for autonomous vehicles.
• To improve tracking adaptability under different vehicle speeds and road curvatures.
• To enhance the real-time control and lateral stability of autonomous vehicles.

Main Methods:

• Developed an adaptive model predictive control (AMPC) system integrating a dynamics-based MPC with an optimal weight adaptive regulator.
• Utilized a Particle Swarm Optimization-Backpropagation (PSO-BP) neural network trained offline with optimal weights from simulations.
• Implemented online adjustment of MPC weights based on trained PSO-BP neural network outputs.

Main Results:

• The AMPC system demonstrated superior tracking adaptation capability compared to classical MPC in joint simulations (Prescan-Carsim-Simulink).
• Real-world autonomous vehicle testing confirmed improved tracking accuracy.
• The adaptive control strategy successfully met real-time control and lateral stability requirements.

Conclusions:

• The proposed AMPC system significantly enhances the tracking adaptability of autonomous vehicles.
• The PSO-BP neural network effectively enables online weight adjustment for improved control performance.
• This adaptive strategy offers a promising solution for robust autonomous vehicle navigation.