High-performance object tracking and fixation with an online neural estimator

Area of Science:

• Robotics and Intelligent Transportation Systems
• Control Systems Engineering
• Computer Vision

Background:

• Vision-based target tracking is crucial for robotics and intelligent transportation systems but often neglects dynamic control issues.
• Existing visual control literature primarily focuses on kinematics, limiting performance in real-world dynamic environments.
• High-performance tracking requires addressing complex dynamic control challenges.

Purpose of the Study:

• To develop a neural network (NN)-based binocular tracking scheme for robust 3D target tracking and fixation.
• To integrate physical system dynamics (Lagrangian dynamics) into the control law for enhanced performance.
• To design a controller that operates with minimal sensory information, specifically without joint velocity data.

Main Methods:

• A neural network (NN)-based controller combined with an observer scheme is proposed.
• The control law explicitly incorporates the vision system's Lagrangian dynamics.
• The neurocontroller-observer design guarantees uniform ultimate bounds on tracking, observer, and NN weight estimation errors.

Main Results:

• The proposed binocular tracking scheme achieves high-performance target tracking and fixation.
• The system effectively accounts for physical dynamics without requiring knowledge of nonlinearities or joint velocities.
• Simulation tests demonstrate the controller's robustness under severe target motion changes.

Conclusions:

• The NN-based binocular tracking scheme offers a viable solution for dynamic target tracking in robotics and intelligent transportation.
• Explicitly considering system dynamics within the control law improves tracking performance.
• The neurocontroller-observer approach provides guaranteed stability and error bounds under general conditions.