Machine Learning Algorithms suitable for hands


These are the references required for understanding the kinda software developed for moving the fingers in a hard to operate in a useful fashion:

We use K-nn and K-nn-G to avoid analytical closed form solutions for the hand dynamics.

Word-distance functions allow for similar motor/servo/pully sequences of actions for short periods of time:


If the smart prosthetic use these algorithms, then we can build network system where the adaptive learning from a group of hands could aid in training new groups of hands.

This is possible today thanx to Blue Tooth and smartphones with internet connection.

We should not think of single hand, we should think of a family of hands share their adaptive learning to move their fingers and grasp better.

This happens with human toddlers, they spend almost a decade to train their fingers and hands. There is no way for them to share the training. But with prosthetic the train could be shared for average similar patients.


There is a terrific difference between the robot hand/graph and Prosthetic’s.

Robot hand is a precision equipment with fixed and precise coordinate systems. The objects it manipulates are well planned for industrial applications and not so arbitrary + actions are well defined e.g. painting a part or drilling a hole…

However the prosthetic hand has not fixed coordinate system and the hardness causes some wobbling.

Objects are quite difference e.g. from ball to ball the radius and friction constants vary drastically. From pencil to pencil the grasp is drastically different.

While robot hand grasp could be solved numerically with high accuracy, the prosthetic hand cannot. We need to have NEAREST approximations to solutions which are adaptive i.e. change and adapt over time and usage.


This is a thought, not sure yet:

If the hands are all the same manufacturing and systems inside, then the hands could share training datasets and talk to each other to better the movements.

In a human baby takes a good 10 years and millions of training trials to train the hands, and at that left hand in general is less functional.

If the hands have the same electromechanical controls and micro-controller is the machine learning based, then they should be able to share training data sets.

This IMHO should reduce the training of the hand to days or months than years.