Does model predictive control outperform tuned PID for high-speed path tracking on a differential-drive platform?

Abstract

This study investigates whether Model Predictive Control (MPC) outperforms a tuned Proportional-Integral-Derivative (PID) controller for high-speed path tracking on a differential-drive robotic platform. As autonomous mobile robots increasingly operate at higher speeds, the choice of control strategy becomes critical for maintaining accuracy and stability while managing computational constraints. Using raw robotics data from ROS bags, including odometry, IMU, and wheel encoder readings, along with repeatable waypoint logs, both controllers were implemented and evaluated under identical experimental conditions. The PID controller was tuned using Ziegler–Nichols and relay methods, while the MPC utilized a linearized kinematic model of the robot. Performance was assessed using defensible metrics such as Root Mean Square (RMS) lateral error, settling time, control effort and energy consumption, CPU load, and missed-deadline rate across varying speed levels. The findings highlight tradeoffs between control accuracy and real-time feasibility, showing that while MPC generally provides superior path-tracking accuracy, it demands significantly higher computational resources, especially at high speeds. This raises practical considerations for real-world deployment, particularly when hardware limitations exist. The results provide guidance on identifying speed and latency thresholds at which MPC’s added complexity justifies its adoption over PID control.