Reflections:

Electrical: Utilizing EMG signals as our input was much harder than we had expected. We knew that it would be challenging, but Jon had worked with EMG signals in the past, and easily overcome the obstacles. We purchased reusable silver-chloride surface electrodes. Our initial circuit, which used LM741 op amps to make instrumentation amps, did not give us a reliable signal. At one point one of our members was instigating a DC shift when he flexed his muscle, before we rectified and bin-integrated the signal (which physiologically can’t happen, though it did look cool). We later went to fabricated instrumentation amps, but these didn’t provide a signal with consistent characteristics either. 60 Hz noise proved to be a very large problem. The bottom line is that EMG is really great, but if that’s the only hard part of your project, you might want to think about doing something else. If the mechanical design will be tough as well, be sure to have a first prototype that uses switches, so that your device still works even if the EMG doesn’t.

Mechanical: We designed our terminal device using ProE 2001. This worked really well, because the geometrical transition from tip to lateral prehension was crucial. ProE allowed us to easily make necessary changes, and see the results. Because ProE works well with assemblies, moving parts around hinges, etc… was quickly accomplished. However, our initial design was too heavy and too complicated. We made 2 - 4 prototypes of every mechanical piece, changing things accordingly with each version. From past experience we made one version of every part, even when we knew we’d have to make another version halfway through to fix a problem. This worked really well, because in the process we might discover another fix that needed to be made. When attempting to make a product like this, expect to have to make multiple versions of all of the pieces, save weight wherever possible, and try to make the parts as simple as possible.

Motors: We broke our first set of motors at the very end of the fabrication of the hand. We’d planned to epoxy them to the aluminum, and in clamping them down, we broke the gearbox on each of them without realizing it. This sent us on a mad dash to Tom Thumb, where owing to a sale we were actually able to get twice the torque for the same cost and almost the same size. Luckily we’d started soon enough that when we broke the pieces at the end of our fabrication, we still had enough time to go out and get new ones, and epoxy those in (this time using a metal sheet to distribute the load and applying much less pressure). Lessons learned: be very careful applying pressure to the sides of servo motors and plan to have your mechanical design done soon enough so that when parts break, you can start from square one.

Microcontroller: In order to make a fluid, humanoid transition between the two modes of operation, it was necessary to use three servos. However, the Handyboard can only support two PMW servos. As a result we decided to use an OOPic microcontroller, since Rajib had one sitting around. This worked very well, and we highly recommend that future groups consider the OOPic. It’s relatively inexpensive and designed to be used with servos in robotics. It allowed us to use three servos and the code was much easier to write than that used by the Handyboard. In addition, the processing time was much quicker with the OOPic than with the Handyboard, giving us a much faster response time. We had to solder in ribbon wire, which was a bit of a pain, but using the OOPic was a very smart decision and we’d definitely do it again.

Part Interactions: We’d originally intended to use gears and belts to interface the parts. However, after consulting Rich we decided against this, and just connected the servos directly to the pieces we wanted them to move. This proved to be a very wise thing; it greatly simplified our lives and cut down on tinkering with interfaces. The only interface that we had other than screwing servos to metal was the interface between H1 and H2, in which a rubber band controlled contraction, and a piece of fishing line connected to a pulley (which was screwed to a servo) controlled extension. This was a good concept, and we liked it a lot, as it provided variable force – however, the rubberband stretched slightly overnight, so that our maximum contraction force decreased, and our fishing line stretched slightly. The hand still functioned great, and most people didn’t notice, but Jon had spent a lot of time optimizing the exact placement of the fingertips, which was now off by a millimeter or two. Aligning the correct position and then keeping it there while you tied a couple of square knots was also a pain. So, the lesson for future groups is that pulleys and cords are easy to design, but they change with time – avoid them if possible, or beware the implications.

 

Overall: We knew from the very beginning that our project was going to be very time consuming and difficult. We started right away on the mechanical design, and made full use of CNC. Having two members that could machine was also a huge plus. We employed a variety of materials to cut down on weight and to cut down on milling time, but aluminum ended up being the material of choice for anything that needed a precision cut. Our first H1 part was made out of wood, but had aesthetic problems with splintering. However, our final H2 part was made out of wood, and that worked really well. Experience and trial and error for material selection works well – don’t be afraid to try new materials. Get the mechanical side done as soon as possible; we were working on that until the very last week, even though we started early and put in an enormous amount of time. Use the OOPic if possible – it makes programming a delight, provides a faster response time, and allows the use of more servos. And plan to make two or three attempts on a project of this type – even if the entire thing has been designed in a program such as ProE. In a project where a couple of extra grams prevented servos from moving, yet some parts were too thin and broke, expect to have to make multiple attempts to achieve the right balance. A project of this variety is definitely rewarding, however. The outcome was great, and in the end it all came together, aside from the EMG. Use of switches, however, allowed our team to adequately demonstrate that a dual mode prehension device is fully feasible. Future projects of this type should be even better!

 

 

Best of Luck,

Team JAT

Jon Sensinger, Neal Poeppelmeier, and Rajib Nandi