Research on robot path tracking method based on IDDPG-MPC.

Area of Science:

• Marine robotics
• Control systems engineering
• Artificial intelligence in navigation

Background:

• Unmanned Surface Vessels (USVs) face challenges in path-following control due to environmental disturbances and system nonlinearities.
• Traditional control methods like Line-of-Sight/Proportional-Integral-Derivative (LOS/PID) lack robustness and adaptability in complex marine settings.
• Underactuated systems and dynamic nonlinearities further complicate precise navigation for USVs.

Purpose of the Study:

• To develop a hybrid control architecture for enhanced path-following control of USVs in complex marine environments.
• To overcome the limitations of traditional control strategies by combining deep reinforcement learning and model predictive control.
• To achieve high-precision path tracking for autonomous surface vehicle navigation.

Main Methods:

• Proposed a hybrid control architecture combining Improved Deep Deterministic Policy Gradient (IDDPG) and Model Predictive Control (MPC).
• IDDPG as the upper-level module uses deep reinforcement learning to determine optimal heading angle increments based on environmental states.
• MPC as the lower-level module optimizes thrust and rudder angle using a USV dynamics model via rolling optimization.
• Implemented a closed-loop 'perception-decision-execution-learning' paradigm with gradient pruning and a custom reward function for stable training and optimal decisions.

Main Results:

• The hybrid IDDPG-MPC approach demonstrated effective adaptability in complex environments, surpassing traditional methods.
• Achieved a 37% reduction in average lateral deviation and a 21% reduction in heading angle error compared to the ALOS-PID method.
• Validated high-precision path tracking control for unmanned surface vessels through simulation.

Conclusions:

• The proposed hybrid IDDPG-MPC control strategy offers a robust and adaptive solution for USV path-following in challenging marine conditions.
• This approach provides a novel method for autonomous surface vehicle navigation, improving accuracy and reliability.
• The study highlights the potential of integrating deep reinforcement learning and MPC for advanced marine robotic control.