Data-Driven Steering Control for Heterogeneous Autonomous Vehicle Platoons via Interval Excitation-Based Learning...

Area of Science:

• Robotics and Control Systems
• Autonomous Vehicle Navigation
• Machine Learning for Control

Background:

• Autonomous ground vehicle platoons (AGVPs) face challenges in path following due to unknown dynamics.
• Accurate system models are often unavailable, hindering traditional control approaches.
• Minimizing lateral offset and heading error is crucial for safe and efficient platooning.

Purpose of the Study:

• To develop a novel data-driven control scheme for heterogeneous AGVPs.
• To address the path following problem for AGVPs with unknown dynamics.
• To formulate the control objective as an inhomogeneous linear quadratic tracking (LQT) problem.

Main Methods:

• A model-based iterative learning algorithm using matrix decomposition was proposed.
• The algorithm was extended to a data-driven implementation with a double-layer integral structure.
• The persistence-of-excitation (PE) condition was relaxed to the interval excitation (IE) condition.

Main Results:

• The proposed method avoids the need for an initial stabilizing control policy and guarantees convergence speed.
• It circumvents numerical issues associated with small discount factors in LQT problems.
• The algorithm iteratively solves the optimal LQT controller with reduced computational complexity and no historical data storage.

Conclusions:

• The novel data-driven control scheme effectively addresses the path following problem for AGVPs with unknown dynamics.
• The method demonstrates feasibility and superiority over existing approaches through simulations.
• This research contributes to safer and more efficient autonomous vehicle coordination.