Optimization method for catching deviation of table tennis training robot based on physical motion model and YOLOv3

Area of Science:

• Robotics
• Computer Vision
• Control Systems

Background:

• Table tennis training robots face challenges with receiving deviation due to inaccurate detection, trajectory prediction errors, and system delays.
• Existing systems often struggle with the dynamic and fast-paced nature of table tennis.
• Improving the precision and responsiveness of these robots is crucial for effective training.

Purpose of the Study:

• To enhance the performance of table tennis training robots by addressing key issues like receiving deviation and system delay.
• To implement a "perception-prediction-control" full chain collaborative optimization strategy.
• To provide technical support for the intelligent upgrade of table tennis training robots.

Main Methods:

• Improved You Only Look Once version 3 (YOLOv3) detection network with convolutional attention modules, adaptive spatial feature fusion, and CIoU loss for small object detection.
• Constructed a high-order physical motion model incorporating spin and aerodynamics, fused with an extended Kalman filter for optimal trajectory estimation.
• Designed a composite control strategy combining feedforward trajectory planning and feedback compensation to mitigate system delay.

Main Results:

• The enhanced detection model achieved 98.8% ± 0.5% accuracy and 97.5% ± 0.6% recall.
• The extended Kalman filter with the high-order model reduced average trajectory prediction error to 8.7 mm ± 1.2 mm (75.6% reduction).
• The control strategy decreased median hitting deviation to 12.4 mm, increased catching success rate to 95.2%, and reduced system delay from 42ms to 32ms.

Conclusions:

• Multi-module collaborative optimization significantly reduces catching deviation in table tennis training robots.
• The proposed "perception-prediction-control" framework offers a viable solution for enhancing robot performance in dynamic environments.
• This research provides valuable insights for motion target tracking and interception devices.