Optimized self-adaptive PID speed control for autonomous vehicles

Abstract: The main control tasks in autonomous vehicles are steering (lateral) and speed (longitudinal) control. PID controllers are widely used in the industry because of their simplicity and good performance, but they are difficult to tune and need additional adaptation to control nonlinear systems with varying parameters. In this paper, the longitudinal control task is addressed by implementing adaptive PID control using two different approaches: Genetic Algorithms (GA-PID) and then Neural Networks (NN-PID) respectively. The vehicle nonlinear longitudinal dynamics are modeled using Powertrain blockset library. Finally, simulations are performed to assess and compare the performance of the two controllers subject to external disturbances. Code can be found here: https://github.com/yassinekebbati/Self-adaptive-PID

• The paper introduces optimized self-adaptive PID approaches using genetic algorithms (GA-PID) and neural networks (NN-PID) to improve longitudinal speed control under diverse conditions.
• It models vehicle dynamics with MATLAB, integrating nonlinear engine, transmission, and propulsion factors to accurately simulate real-world disturbances.
• Simulation results highlight GA-PID's precision in stable scenarios and NN-PID's superior adaptability for handling wind, road slope, and other external challenges.

Introduction

The paper "Optimized self-adaptive PID speed control for autonomous vehicles" explores advanced PID control techniques specifically designed for managing the longitudinal dynamics in autonomous vehicles. Two self-adaptive PID approaches are analyzed: one employs Genetic Algorithms (GA-PID) for PID optimization, and another utilizes Neural Networks (NN-PID) for continuous PID adaptation. The focus is on enhancing longitudinal speed control, addressing real-world disturbances, and maintaining adaptability to varying operational conditions and external environmental impacts.

Vehicle Dynamics Modeling

The study implements a comprehensive model of vehicle longitudinal dynamics using the MATLAB Powertrain blockset. By modeling the nonlinear aspects of engine, transmission, and the entirety of the electric propulsion system, the dynamics are accurately simulated. This modeling approach encompasses key forces such as aerodynamic drag, gravity, rolling resistance, and external influences like wind speed and road gradient, facilitating precise control system evaluations.

PID Controller Design

The foundation of the control system is a PID controller that processes error signals—typically the difference between desired and actual vehicle speeds—to adjust engine output for speed regulation. While PID controllers are prevailing due to their straightforward design and effectiveness, the challenge lies in selecting optimal PID gains, especially under variable and nonlinear conditions.

Genetic Algorithm-Based PID Optimization

Genetic Algorithms are leveraged to optimize PID gains. The GA-PID method initiates by establishing a population of potential solutions. These are assessed using a fitness function based on the Mean Squared Error (MSE) between target and actual performance. GA leverages operations such as mutation and crossover to refine the population, seeking optimal PID parameter sets tailored for distinct environmental conditions, enhancing control precision and response stability.

Neural Network PID Adaptation

The NN-PID method offers adaptive online tuning capabilities. Utilizing a MULTI-LAYER PERCEPTRON (MLP) architecture, the PID gains are continuously adjusted through back-propagation of error signals. This approach predicts optimal gains dynamically in response to fluctuating disturbances and driving scenarios, ensuring robust adaptability and immediate response to control errors.

Simulation Results

Simulation scenarios reveal distinct performances of GA-PID and NN-PID strategies across varied disturbance profiles:

1. GA-PID demonstrates smooth and accurate performance in scenarios devoid of disturbances, showcasing its optimization strength under stable conditions.
2. In scenarios involving wind speed and road slope variations, the NN-PID showcases superior adaptability and reactive speed due to its online learning mechanism.
3. The comparative analysis highlights GA-PID's precision and NN-PID's robustness and agility in adjusting to changing road and environmental conditions.

Comparative Performance Analysis

Quantitative assessments, including MSE and transient response metrics such as rise time, settling time, and overshoot, underscore the efficacies of both approaches. GA-PID exhibits efficient optimization for targeted scenarios, while NN-PID enhances generalization and adaptability for diverse driving profiles and external challenges.

Conclusion

The study rigorously evaluates the robustness and adaptability of self-adjusting PID controllers for autonomous vehicles. GA-PID and NN-PID each offer unique advantages in control precision and disturbance adaptability. Future directions are suggested to investigate alternative optimization algorithms, improve transition smoothness among PID states, and explore novel control strategies to further refine longitudinal vehicle dynamics control, enhancing autonomous vehicle performance under diverse operational conditions.