Hybrid Recurrent Neural Network Architecture-Based Intention Recognition for Human-Robot Collaboration

Area of Science:

• Robotics
• Artificial Intelligence
• Machine Learning

Background:

• Human-robot collaboration (HRC) necessitates intelligent human intention recognition for effective task execution.
• Current deep learning methods for motion prediction in HRC show limitations, impacting task performance.

Purpose of the Study:

• To propose a hybrid recurrent neural network (RNN) architecture for enhanced intention recognition in cooperative assembly tasks.
• To improve the accuracy of human motion prediction within a collaborative workspace.

Main Methods:

• Developed a hybrid RNN integrating improved Long Short-Term Memory (ILSTM) and improved Bidirectional LSTM (IBi-LSTM) units.
• Utilized state and gate activation functions to enhance network performance.
• Implemented a quartile-based amplitude limiting algorithm for spatiotemporal data filtering.
• Validated the approach using a UR5 collaborative robot and a human motion capture system.

Main Results:

• The proposed hybrid network demonstrated superior performance in predicting human operator motion compared to existing deep learning methods.
• The IBi-LSTM layers effectively captured complex sequential features, while the LSTM layer handled forward dependencies.
• Data filtering significantly improved the accuracy of network training and testing.

Conclusions:

• The developed hybrid RNN architecture offers a more precise method for human intention recognition in collaborative robotics.
• This advancement contributes to more seamless and efficient human-robot collaboration in assembly tasks.