An online human-robot collaborative grinding state recognition approach based on contact dynamics and LSTM

Area of Science:

• Robotics
• Human-Robot Interaction
• Control Systems

Background:

• Physical human-robot collaboration (PHRC) requires accurate state recognition for safety and efficiency.
• Distinguishing between human-robot contact and robot-environment contact is crucial in collaborative tasks like grinding.

Purpose of the Study:

• To propose a novel contact dynamics-based method for recognizing human-robot collaborative grinding states.
• To enhance the safety and performance of physical human-robot collaboration systems.

Main Methods:

• Dynamic models were established to differentiate between human-robot contact and robot-environment contact dynamics.
• Feature selection techniques (Spearman's correlation, random forest recursive feature elimination) were employed to reduce data redundancy and computational load.
• A Long Short-Term Memory (LSTM) network was utilized to build a collaborative state classifier.

Main Results:

• The proposed method achieved a recognition accuracy of 97% within a 5 ms timeframe.
• A recognition accuracy of 99% was achieved within a 40 ms timeframe.
• The approach effectively reduces data redundancy and computational burden.

Conclusions:

• The contact dynamics-based method is highly effective for recognizing human-robot collaborative grinding states.
• The proposed approach offers fast and accurate state recognition, crucial for real-time PHRC applications.
• This method contributes to safer and more efficient human-robot collaboration.