MNNMs Integrated Control for UAV Autonomous Tracking Randomly Moving Target Based on Learning Method.

Area of Science:

• Robotics and Control Systems
• Artificial Intelligence
• Computer Vision

Background:

• Autonomous tracking of moving targets by unmanned aerial vehicles (UAVs) using only airborne sensors presents significant challenges.
• Existing control strategies often lack the adaptability and efficiency required for dynamic tracking scenarios.

Purpose of the Study:

• To develop and evaluate a novel integrated controller framework for UAVs to autonomously track moving targets.
• To demonstrate the effectiveness of a multi-neural-network module (MNNM) based approach for enhanced tracking performance.

Main Methods:

• Proposed a novel integrated controller framework utilizing multi-neural-network modules (MNNMs).
• Designed two distinct neural networks within the framework for target perception and guidance control.
• Employed deep learning and reinforcement learning methods for training the integrated controller.

Main Results:

• The MNNM-based integrated controller demonstrated faster and more efficient training compared to end-to-end deep reinforcement learning controllers.
• Flight tests in simulated and realistic environments confirmed the controller's ability to track randomly moving targets with high velocity.
• The integrated controller outperformed a control mode combining a perception network with a proportional-integral-derivative (PID) controller.

Conclusions:

• The proposed MNNM-based integrated controller offers superior performance for autonomous target tracking in UAVs.
• The trained controller exhibits effective transferability from simulation to realistic environments, enabling robust real-world applications.
• This framework represents a significant advancement in autonomous aerial tracking systems.