Robots are caught in a catch-22: need the numbers to be useful, need to be useful to justify the numbers. Open source can overcome this.
This is where we're aiming for:
The first mechanical draft has been assembled but it's somewhat unreliable - we're rebuilding, particularly the electronics and hope to be ready to demo after the summer with the intention of integrating the latest in deep learning via an embedded Jetson TX2 - please let us know if you're interested and we'll release the full instructions...
A single motor fitted to an assembled hand shows the kind of grip that can be attained:
As an introduction to the project we recommend building the mechanics of a robot hand. There is a complete instructional video on youtube:
The 3d printed parts can be download from thingiverse: https://www.thingiverse.com/thing:549095
Though it can be awkward to remove all the supports after extrusion printing - it's much easier to buy a complete print at a commercial quality from Shapeways with the added advantage that it comes with all the printed parts pre-assembled:
Simply press in 6x10x3mm bearings (available from Technobots website, part no: 4255-130) between the finger sections and add the torsional ligaments by threading nylon fishing line (100daN approx. diameter 1.1mm) through the finger sections to link the top of one joint to the bottom of the section two along i.e. the top of the knuckle is linked to the underside of the middle section and the top of the first section is linked to the top of the tip section and vice versa. Attach dyneema tendons by spreading the ends of a dyneema kiteline and embedding the fibres in polymorph. A single spring made from 3mm polyurethane tubing is attached to the top of the knuckle to extend the fingers.
An update to the hand build with the finger sections uploaded as separate stl files: https://www.thingiverse.com/thing:727323