Position control of human-like robotic hand index finger using machine learning
The hand has been a great tool for humans with the advantage of dexterous function and power to do daily activities such as turning a doorknob, wearing clothes, using a screwdriver and picking up objects. These dexterous hand motions are achieved due to the biomechanics of hands and redundant mechan...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2020
|
| Subjects: | |
| Online Access: | http://eprints.utm.my/id/eprint/93046/1/MuhammadHazwanAbdulMSKE2020.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The hand has been a great tool for humans with the advantage of dexterous function and power to do daily activities such as turning a doorknob, wearing clothes, using a screwdriver and picking up objects. These dexterous hand motions are achieved due to the biomechanics of hands and redundant mechanism of hand muscles. These muscles receive brain signals and react accordingly to perform different hand motion gestures when dealing with different objects. Sub-conscious mind in our brain instructs actuation of different extrinsic and intrinsic muscles in order to move our fingers to different positions for desired functions. The index finger plays an important role in performing hand gestures for human daily activities. This work presents position control of index finger of human-like robotic hand (HR Hand) using machine learning. The HR Hand is a replication of the human hands in terms of bones, ligaments, muscles, extensor mechanism, tendon and its pulley system. The muscles of the hand were fabricated using thin multifilament McKibben muscles following design of index finger by A.A.M. Faudzi. Motion of the HR Hand is captured using ViconTM motion capture to build a K-nearest neighbour (KNN) and Artificial Neural Network (ANN) model. This model is used to predict the combination of muscles used for fingertip position control. Hyperparameter optimization is done using grid search method to obtain better accuracy. The ANN model showed to have better accuracy of 71.36 % in predicting actuation of muscle class compared to the 65.38 % by KNN. The ANN model is applied in building a feed-forward controller and verified on the system with 2.7 cm steady state error. Using this model, future control of HR Hand is expected. |
|---|