Rotor Attitude Estimation for Spherical Motors Using Geometry-Constrained Kalman Transformer Algorithm in Monocular...

Area of Science:

• Robotics and Control Systems
• Computer Vision
• Mechanical Engineering

Background:

• Permanent-magnet spherical motors (PMSpMs) offer three-degree-of-freedom omnidirectional motion.
• Accurate rotor attitude estimation (RAE) is critical for the closed-loop control of PMSpMs.
• Existing RAE methods may lack robustness or real-time capabilities for complex spherical motor dynamics.

Purpose of the Study:

• To propose and validate a novel visual RAE method for PMSpMs.
• To enhance the accuracy, robustness, and real-time performance of RAE.
• To compare the proposed method against existing algorithms and sensor technologies.

Main Methods:

• Development of a visual RAE system using a monocular camera and a visual feature component (VFC) on the rotor.
• Implementation of the Kalman filter and geometric constraint Transformer (GK-TransT) algorithm for enhanced tracking.
• Comparative analysis with TransT, KCF, and CSRT algorithms, and validation against a MEMS sensor using a hydraulic rotary table benchmark.

Main Results:

• The GK-TransT algorithm achieved high tracking precisions of 90.9% (main) and 94.4% (auxiliary) feature points.
• The system demonstrated an average processing speed of 61.23 FPS with a single-frame latency of 16.33 ms.
• The GK-TransT method outperformed MEMS sensors in accuracy and showed superior robustness under occlusion and motion blur.

Conclusions:

• The proposed GK-TransT visual RAE method is highly applicable for PMSpMs due to its precision, speed, and robustness.
• The developed RAE test bench validates the method's effectiveness and practicality.
• This visual RAE approach provides a reliable alternative for controlling omnidirectional spherical motors.