Multi-ASV Coordinated Tracking With Unknown Dynamics and Input Underactuation via Model-Reference Reinforcement...

Area of Science:

• Robotics
• Control Systems
• Artificial Intelligence

Background:

• Coordinated tracking of autonomous surface vehicles (ASVs) faces challenges due to model uncertainties and underactuation.
• Existing control methods may not adequately address adaptive communication needs in multi-ASV systems.

Purpose of the Study:

• To develop and evaluate a model-reference reinforcement learning control strategy for coordinated tracking of uncertain ASVs.
• To investigate the integration of model-reference control with reinforcement learning for improved performance and adaptive communication.

Main Methods:

• Implementation of a model-reference reinforcement learning algorithm tailored for ASV control.
• Comparative analysis against baseline control methods to assess performance.
• Evaluation of training efficiency improvements offered by the proposed reinforcement learning approach.

Main Results:

• The proposed model-reference reinforcement learning control demonstrates superior performance compared to baseline methods.
• The algorithm effectively addresses challenges posed by model uncertainties and input underactuation in ASVs.
• Significant improvements in training efficiency were observed for the reinforcement learning process.

Conclusions:

• Model-reference reinforcement learning offers a robust solution for coordinated tracking of underactuated and uncertain ASVs.
• The developed approach enhances system performance and training efficiency.
• This method facilitates adaptive communication among ASVs, paving the way for more sophisticated multi-agent systems.