Event-Trigger Reinforcement Learning-Based Coordinate Control of Modular Unmanned System via Nonzero-Sum Game

Area of Science:

• Robotics and Control Systems
• Artificial Intelligence
• Optimization Techniques

Background:

• Coordinate control of modular unmanned systems (MUS) requires minimizing position error and control torque.
• Existing methods face challenges with communication burden between sensors and actuators.
• Event-triggering mechanisms offer a way to reduce communication load.

Purpose of the Study:

• To propose an event-trigger reinforcement learning (ETRL)-based coordinate control strategy for MUS.
• To leverage the nonzero-sum game (NZSG) strategy to transform the control problem into a reinforcement learning (RL) issue.
• To reduce communication burden and improve control performance.

Main Methods:

• Established the dynamic model of MUS using joint torque feedback (JTF) technology.
• Applied the NZSG strategy to formulate the control problem as an RL task.
• Utilized an event-trigger (ET) mechanism to avoid periodic communication.
• Employed an ET-critic neural network (NN) to approximate the performance index function and derive the ETRL policy.
• Verified closed-loop system stability using Lyapunov's theorem.

Main Results:

• The proposed ETRL method effectively controls modular unmanned systems.
• Experimental results demonstrate a 30% reduction in position error compared to existing methods.
• Control torque was reduced by 10%.

Conclusions:

• The ETRL-based coordinate control strategy is valid and effective for MUS.
• The method successfully reduces position error and control torque.
• The event-trigger mechanism significantly alleviates communication burden.