IPTGNet: an adaptive multi-task recognition strategy for human locomotion modes

Area of Science:

• Robotics
• Biomechanics
• Machine Learning

Background:

• Lower limb exoskeletons face challenges in accurately processing complex human motion.
• Effective human locomotion mode recognition is crucial for advanced exoskeleton control and assistance.

Purpose of the Study:

• To propose an innovative multi-task recognition model, IPTGNet, for human locomotion modes in lower limb exoskeleton applications.
• To enhance the accuracy and stability of human motion processing in exoskeletons.

Main Methods:

• Developed IPTGNet by parallelly fusing Temporal Convolutional Network (TCN) and Gated Recurrent Unit (GRU).
• Employed an improved particle swarm optimization (IPSO) algorithm for dynamic hyperparameter tuning.
• Integrated a finite state machine (FSM) to correct transition states during locomotion.

Main Results:

• IPTGNet demonstrated faster and more stable convergence compared to existing methods.
• Achieved a high recognition rate of 99.47% for human locomotion modes.
• Reported a low standard deviation of 0.42%, indicating robust performance.

Conclusions:

• IPTGNet offers an effective solution for multi-task recognition in lower limb exoskeletons.
• The proposed model significantly advances the precision and reliability of human motion interpretation for exoskeletons.
• This work provides a foundation for more sophisticated and intuitive exoskeleton-human interaction.