One of the components of the summer math class I’m taking prior to actually starting the PhD program at the University of Michigan is a “research project” in our field, out of which we are to present some math. Since I’m the only computer science person in the program, they grouped me with the only electrical engineer in the program. The problem is, despite the fact that EECS is one department at Michigan (and MIT and a lot of places), they’re really two separate fields. So what kind of project does a CS student interested in AI and an EE student interested in bioengineering work on? Robotics, of course!
Teddler.jpgA few weeks ago, when we first met with our advisor, he introduced us to the topic of passive dynamic walking and a paper by Tad McGeer. There were some differential equations involved, so we thought we’d tackle the project. Then we (mistakenly) thought, “Hey, building one of these little walking robots seems easy. Let’s do it to make our project that much cooler.” Our first problem was that we tried to simply build a synthetic wheel, ignoring that line in the paper that said “Meanwhile an infintesimal shortening of the other leg will keep it clear of the ground…” But little legs of balsa wood do not spontaneously shrink (or grow). Our first try simply rocked back and forth until it tipped over.
We realized that our robot needed either knees (much more work than we wanted to do) or it needed to wobble slightly side to side. Our advisor found us another paper to look at — one with a very familiar second author, Teresa Zhang. I left the meeting and immediately called Breath to let him know that I was going to reference his girlfriend’s work. He was highly amused, and didn’t even seem to mind that I was calling him while he was at work. Once my partner and I developed a rough idea of how to make our robot wobble, I got Teresa herself on the phone and ran the idea by her. The good news was, while she thought we were being really ambitious, she didn’t think we were insane.
Toddler, the robot Teresa worked on with Russ Tedrake at MIT, has gotten pretty famous. Toddler also “learns” how to walk on different sources — something our robot isn’t even close to doing. We may not have put as much thought into the calculations as Teresa put into Toddler, but our little Teddler (named in honor of Toddler and the University of Michigan’s Department of Mathematics alumni Ted Kaczynski) waddles and walks down inclines fairly consistently. And for a little guy made mostly of balsa wood, Elmer’s glue, duct tape, and some washers, that’s pretty impressive. A 30 second (poorly lit) clip of him toddling down my new coffee table is here.


One response

  1. So (as you know) there’s a lot of both wasted and turbulent motion rolled into the robot’s rock: Would downriggers (that do not “quite” touch the underlying plane) at some convenient junture along the sidearms help to stabilize Teddler and solve one of the apparent problems until another solution comes along? OR would splaying the legs outward at 10 degrees achieve the same advantage? [Just notions of The Unknowing, I know (that much) but I ask because Teddler sure looks like any learning toddler I’d ever seen — whose outstretched arms are as often reasons for falling as for saving a fall: either downriggers or splaying might enable foreshortening those longish armatures without any loss in efficiency — indeed, perhaps with some gain.]
    And THANKS! for making the trip ERhode! SWMBO and I sure enjoyed being with you!