Gaussian Process Based Model Predictive Control for Overtaking in Autonomous Driving.

Area of Science:

• Robotics and Autonomous Systems
• Control Theory
• Machine Learning

Background:

• Autonomous vehicles require robust systems for safe navigation, especially during complex maneuvers like overtaking.
• Traditional control strategies often struggle with model uncertainties and computational demands.
• Ensuring vehicle safety under dynamic conditions is a critical challenge in autonomous driving research.

Purpose of the Study:

• To develop a novel framework for autonomous overtaking and obstacle avoidance.
• To integrate path planning directly into Gaussian Process-based Model Predictive Control (GPMPC).
• To enhance safety and reduce computational burden compared to conventional methods.

Main Methods:

• Incorporating Gaussian Process (GP) regression with a nominal model to learn from uncertainties and unmodeled dynamics.
• Formulating the geometric relationships between vehicles as constraints within the GPMPC framework.
• Transforming state constraints into soft constraints using a relaxed barrier function in the cost function for improved optimizer efficiency.

Main Results:

• The proposed GPMPC framework successfully enables autonomous overtaking maneuvers.
• Vehicle safety is consistently maintained throughout the simulated scenarios.
• The approach demonstrates improved performance by learning from model mismatch and unmodeled dynamics.
• Reduced computational complexity is achieved by eliminating the need for a separate high-level path planner.

Conclusions:

• The novel GPMPC framework provides an effective solution for safe autonomous overtaking and obstacle avoidance.
• The integration of GP regression enhances control system robustness and adaptability.
• This method offers a computationally efficient and safe approach for complex autonomous driving tasks.