Deep Reinforcement Learning Controller for 3D Path Following and Collision Avoidance by Autonomous Underwater...

Area of Science:

• Robotics
• Artificial Intelligence
• Control Systems

Background:

• Traditional control theory requires detailed system models for controller design.
• Complex autonomous systems, like underwater vehicles, face challenges in path following and collision avoidance.
• Decision-making in dynamic environments is nontrivial for autonomous systems.

Purpose of the Study:

• To develop autonomous agents for dual objective path following and collision avoidance in underwater vehicles.
• To explore the application of Deep Reinforcement Learning (DRL) for autonomous decision-making.
• To enable autonomous systems to operate without a priori knowledge of the environment or objectives.

Main Methods:

• Utilized state-of-the-art Deep Reinforcement Learning (DRL) techniques.
• Developed autonomous agents capable of learning complex behaviors.
• Trained agents in environments with challenging obstacle configurations.

Main Results:

• Demonstrated the viability of DRL for path following and collision avoidance.
• Achieved autonomous decision-making comparable to human levels.
• Validated the approach in scenarios with extreme obstacle configurations.

Conclusions:

• DRL is a powerful tool for creating autonomous agents in complex environments.
• Autonomous underwater vehicles can achieve sophisticated navigation tasks using DRL.
• This research paves the way for advanced autonomous decision-making in robotics.