Autonomous Underwater Vehicle Docking Under Realistic Assumptions Using Deep Reinforcement Learning

Problem

Autonomously homing and docking an underwater vehicle under realistic, noisy ocean conditions where traditional controllers struggle.

Why it matters

Ocean complexity and variability constrain traditional controllers, limiting autonomous underwater vehicle operations.

What the AI did

Deep Reinforcement Learning policies were trained for homing and docking using a novel observation space combining raw noisy observations with EKF-filtered data.

Results

Stable policies in fewer learning steps, outperforming traditional control methods in simulations

Impact

Simulation evidence suggests DRL with EKF-filtered observations can outperform traditional docking controllers; real-vehicle validation not stated.

Record

Importance rating

More in Robotics & Autonomy