Underwater gliders linear trajectory tracking: The experience breeding actor-critic approach.

Area of Science:

• Robotics
• Oceanography
• Control Systems Engineering

Background:

• Underwater gliders are crucial for oceanographic data collection but face challenges in trajectory tracking due to currents and complex dynamics.
• Accurate trajectory tracking is essential for mission success, requiring robust control strategies that can adapt to environmental disturbances.
• Model-free optimization approaches are often necessary due to the inherent uncertainties in glider dynamics and current fields.

Purpose of the Study:

• To develop an effective control strategy for underwater gliders to achieve precise trajectory tracking in challenging current fields.
• To address the limitations of traditional mathematical analysis for model-free optimization problems in underwater glider control.
• To propose a novel neural network control framework capable of enhancing exploration and mitigating current disturbances.

Main Methods:

• The underwater glider trajectory tracking problem is reformulated as a Markov Decision Process (MDP).
• A novel neural network control framework, Experience Breeding Actor-Critic (EBAC), is proposed for model-free optimization.
• The EBAC framework is designed to enhance exploration in high-reward areas and meticulously steer glider heading to counteract currents.

Main Results:

• Simulation results demonstrate the efficacy of the EBAC framework in achieving accurate trajectory tracking for underwater gliders.
• The proposed EBAC method shows superior performance in guiding gliders to precisely follow target tracks amidst current disturbances.
• The EBAC enhances the glider's ability to explore potentially rewarding areas while actively compensating for environmental influences.

Conclusions:

• The Experience Breeding Actor-Critic (EBAC) framework provides a robust and effective solution for underwater glider trajectory tracking in current fields.
• The MDP-based approach combined with the EBAC neural network control offers a promising direction for autonomous underwater vehicle control.
• This research highlights the potential of advanced AI control techniques for enhancing the operational capabilities of underwater gliders.