Decoding Multi-Finger Motions and Grasp Types with Grasp-Specific Models and Lightmyography Based Muscle-Machine Interfaces
Zhe Wang, Bonnie Guan, Shifei Duan, Kean C. Aw, Minas Liarokapis
AI summary
Problem
The decoding effectiveness of lightmyography (LMG) signals for continuous multi-finger motion prediction and grasp classification remains insufficiently explored compared to established methods like sEMG.
Approach
Six participants grasped five objects while LMG and motion capture data were recorded, then decoded using random forest, CNN, and MLP models trained with both generalized and grasp-specific strategies.
Key results
- MLP achieved highest grasp classification accuracy (98.81% average)
- Random forest significantly outperformed CNN and MLP for continuous finger motion regression
- Grasp-specific regression models improved prediction accuracy over generalized approaches
- LMG signals reliably decode five-finger kinematics with low error (<6.5° RMSE)
Why it matters
Provides a robust, low-cost alternative to sEMG for controlling advanced prosthetic and robotic hands by validating LMG's efficacy for multi-finger decoding.
Abstract
Efficiently decoding human movement and/or in- tention is essential for controlling advanced prosthetic and robotic systems. Various muscle-machine interfaces have been researched for this purpose, including electromyography and lightmyography based interfaces. However, the decoding ef- fectiveness of lightmyography signals for multi-finger hand motions remains insufficiently explored. This study investigates the decoding of human multi-finger movements using different machine learning methods. Lightmyography and finger mo- tion data were collected from six participants grasping five common objects. Data were preprocessed using the sliding window method and decoded using three machine learning algorithms: random forest, convolutional neural networks, and multi-layer perceptron. Moreover, models were trained in a grasp-specific manner increasing decoding accuracy. Finally, statistical analysis demonstrated that the random forest model significantly outperformed the other methods, establishing it as the most suitable technique for decoding multi-finger motions from lightmyography signals.