Dexter, a really great robot arm project, just won top honors in the 2018 Hackaday Prize, and walked away with $50,000 toward continuing their project. As a hat tip to Hackaday and the community, Haddington Dynamics, the company behind Dexter, agreed to open-source their newest version of Dexter as well. As James Newton said when accepting the trophy during the award ceremony, “because of your faith in us, because of this award, we have been moved to open-source the next generation of Dexter.” Some very clever work went into producing Dexter, and we can’t wait to see what further refinements have been made!

Dexter isn’t the only robotic arm in town, by any means. But in terms of hobbyist-level robotics, it’s by far the most complete robot arm that we’ve seen, and it includes a couple of design features that make both its positional accuracy and overall usability stand out above the rest. This is a robot arm with many of the bells and whistles of a hundred-thousand dollar robot, but on a couple-thousand dollar budget.

Moreover, many of the design innovations can be taken piecemeal and used in your own project. For instance, our judges particularly liked Dexter’s clever encoder design. Optical encoders usually use a disk with holes in it; shine a light through the disk onto a sensor and you can determine the position by counting the number of resulting pulses.

Dexter takes this a step further, using the analog value of the light sensor, allowing it to track a few thousand positions per hole, multiplying the usable resolution. This means that, with a 3D printed encoder wheel and some serious math to transform this into a position, Dexter’s encoder can resolve over a million positions per revolution. That’s mind-blowing precision, and it’s available relatively cheaply due to advances in computing power.

Specifically, an FPGA and some custom “gateware” logic code runs the conversion of encoder data to arm position. That’s right, they’re dedicating an embedded Linux computer with FPGA peripherals doing the heavy lifting to calculate two joints’ position. You might think that this is overkill, but with an arm like this any small angular error in the first joint that sits on the table is multiplied by the reach of the arm. You want positional accuracy of 0.1 mm at the end of a half-meter arm, you’re looking at 0.01 degree angular resolution. Getting that resolution with a 3D-printed encoder wheel and some heavy computation is a great innovation.

Combine such detailed positional feedback with some good software and you can design an arm that’s both compliant, strong, and easily trainable just by moving it around. Dexter uses harmonic drives for some of the axes, and while you don’t normally want to force a harmonic drive, the extreme resolution of the positional sensor lets the robot know when it’s being pushed out of position and either apply more force to hold position, or drive the motors …read more

Source:: Hackaday