Adaptive Model Predictive Control with Vehicle Mass Estimation for Intelligent Vehicle Control

Tian, Kai Wen (2025) Adaptive Model Predictive Control with Vehicle Mass Estimation for Intelligent Vehicle Control. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text Tian Kai Wen - FULL TEXT.pdf Restricted to Registered users only Download (2MB)

Abstract

This project develops an advanced intelligent vehicle control system by integrating Adaptive Model Predictive Control (AMPC) and a real-time vehicle estimation framework, enhanced through Reinforcement Learning (RL). The system dynamically optimizes control strategies and stability by combining Recursive Least Squares with Multiple Forgetting Factors (RLSMFF) for mass estimation and a Q-learning algorithm to autonomously adjust the RLSMFF parameters to enhance real-time vehicle mass estimation accuracy. The RL agent observes states derived from composite errors (torque, velocity, acceleration) and selects actions for parameter adjustments. By leveraging RLSMFF, the system demonstrates a significant improvement in vehicle mass estimation accuracy compared to the Extended Kalman Filter (EKF). Simulation results highlight the superiority of RLSMFF in this context, reinforcing its potential in enhancing vehicle control systems. Moreover, results confirm that the RL-enhanced RLSMFF outperforms the standard RLSMFF by further reducing estimation errors and improving adaptability to dynamic conditions. The unified AMPC-ACC framework leverages these adaptive estimations to compute predictive control inputs, ensuring fuel efficiency, collision avoidance, and robust tracking across diverse scenarios. This hybrid approach merging RLSMFF, RL, and AMPC demonstrates superior performance over conventional static parameter methods, achieving enhanced safety, responsiveness, and adaptability under uncertainty.