Looking for the perfect last-minute stocking stuffer for that very special geek in your life? Check out Chico Bicalho's mechanical Critter Collection (available online at Kikkerland), a collection of small wind-up toys — or "kinetic sculptures" if you wanna be more artsy about it. The original Critter was the brainchild of a motion display company called Handscraft Motors located in New York City, and built by hand using a vintage 1950s Japanese gearbox someone picked up at a surplus store. The Spousal Unit received one of the more recent designs from a friend of ours for Christmas: Awika, essentially the love child of two earlier designs, Bonga (a determined little crawler/climber: "it will crawl over anything to get to its objective, whatever that is") and Sparklz (known for shooting sparks, and hence carries a warning to keep it away from flammable liquids or vapors).
The accompanying booklet claims, "Awika is a very good climber," and challenges the user to try it out with various small objects serving as barriers, such as an open book. We did this with a Vosges chocolate gift box, with pretty disappointing results: mostly, Awika clawed ineffectually at the barrier, and if we really cranked it up and stored as much potential energy as possible, Awika managed to get part of the way up the barrier before flipping over onto its back. We doubt its parent, Bonga, would perform much better. Still — sparks shoot out of Awika! That makes it better than Bonga. Even the cat, usually evincing a studied feline ennui, was cautiously enthralled enough to pad over from her comfy perch by the window.
We can't fault Awika too much, since the task of walking and/or climbing over barriers is quite a bit more complicated, biomechanically speaking, than one might think. Bipedal (human) forward walking employs the double pendulum strategy: one leg leaves the ground and swings forward from the hip (pendulum #1), then strikes the ground with the heel and rolls through the foot to the toe. As the foot hits the ground the other leg swings into motion (pendulum #2). Movement is coordinated so that one or the other foot is always in contact with the ground.
Furthermore, human beings can learn to adjust our gait to compensate for changes in terrain: we walk differently on ice (i.e., verrry carefully) than we do on dry ground, and our gait while walking uphill differs slightly from our gait while walking downhill. Straightforward walking requires very little mental effort, but clamboring over rocks or uneven terrain causes the brain to kick in at a higher level of processing to analyze the terrain and send signals to adjust the length of one's stride, speed, the angle of the torso, and a host of other modifications.
So getting a wind-up toy to walk or climb is a bit of a daunting challenge; heck, it's daunting even for robots, most of which have some form of a rudimentary brain, like an artificial neural network (ANN). The human central nervous system makes use of a biological neural network: definitions can get fuzzy, but basically we're talking about collections of neurons linked together into a network, capable of exhibiting "global complex behavior, determined by the connections between the processing elements [neurons] and element parameters." In biological networks, these grouped neurons perform specific physiological functions. The artificial versions aren't all that different, except they're far more simplified. ANNs use artificial neurons — computer processing elements — connected together to form a network of nodes. The best thing about ANNs is that adding appropriate computer algorithms ("designed to alter the strength, or weights, of the connections in the network to produce a desired signal flow") makes them capable of adaptation.
As microelectronics has advanced, so, too have the ANNs used in robotics. Early attempts at walking robots had six legs, apparently because of limitations in microprocessor technology, but electronics have become so advanced that tripod and biped robots are much more commonplace today. The robots the aliens built in H.G. Wells' classic War of the Worlds were essentially tripods, since that's a design that has an inherent stability — unless you manage to knock out one of its three "legs." But the tradeoff is efficient movement: most tripod robots actually built in the lab don't do much better than a slow sort of shuffle. Anything more than that, and the legs inevitably become tangled.
Earlier this year, scientists at Virginia Tech introduced STriDER, a three-legged robot that does manage to achieve a reasonably elegant walking gait (h/t: Cognitive Daily). To step forward, the robot shifts its weight onto two legs so that it falls forward away from the third leg. Then its body flips upside down so the third leg can swing up between the other two just in time to catch the ground, so STriDER ends up back in a stable tripod stance. It just needs to switch its choice of swinging leg to change directions. Whether STriDER will actually prove to be a useful design in terms of real-world deployment remains to be seen, but the researchers think it might be useful to deploy small sensors.
Bipedal robots are certainly being built, and they're getting better and better. Scientists at the University of Goettingen's Computational Neuroscience program have built a walking robot that not only mimics the biomechanics of human walking, but also simulates the neuronal principles that form the basis of human adaptivity while walking, according to a paper they published back in July in PLoS Computational Biology. In order for a robot to walk effectively, it has to be able to achieve comparable levels of coordination, meticulously adjusting all the aspects of movement control (angle of knee joints, hip momentum, the balance point of the torso, etc.) in response to whatever it encounters along its path.
Apparently, RunBot currently holds the world record in speed walking for dynamics mechanics, clocking in at 3.5 leg-lengths per second (impressive considering the thing is merely 30 centimeters high); the previous champion was MIT's "Spring Flamingo" robot, which is four times as tall and yet has a stride of just 1.4 leg-lengths per second. and by adding an infrared eye, the researchers have enabled RunBot to detect a slope in its path and adjust its gait in response. (Here are some movies of RunBot in action.) The secret lies in part in how it mimics the human gait: RunBot leans forward slightly when walking and uses shorter steps. It's also worth noting that this is learned behavior: RunBot figures out through trial and error, in much the same way that a human child will do so. For instance, the first time it tries to climb a slope, RunBot will keel over backwards, mostly because it hasn't yet learned to change its gait in response to visual cues from the surrounding environment. But after a few tries, RunBot has learned its lesson, and adjusts its gait in proportion to the steepness of the slope it is attempting to climb.
How does it accomplish this reflexive response? A lot of robot designs are quite complicated: Honda's Asimo robots, for example, use elaborate control systems to maintain balance. RunBot has relatively few sensors and a simple program that mimics the way neurons control reflexes in humans and animals, and relies on inertia as part of its gait. The result is a quicker, more "natural" walk. RunBot really only needs to detect (a) when a foot touches the ground, and (2) when a leg swings forward. When one foot touches the ground, the opposite leg swings forward. The "knee" of the swinging leg bends automatically until a hip sensor tells it to straighten out just before hitting the ground. Once that other foot lands, the cycle repeats. And repeats.
RunBot gets its visual cues from a handy infrared eye, but that information is processed by a fairly simple neural network using a hierarchical organization for the control of its movements, similar (but simpler) to how the human brain functions. At lower rungs of the ladder, so to speak, movement occurs in reflexive response to input from peripheral sensors, and built-in control circuits ensure that the robot's joints don't over-stretch. It also makes sure that the robot takes its next step as soon as the foot touches the ground. The higher centers of organization kick in whenever an adaptation in gait becomes necessary, such as hitting uneven terrain or encountering a steep slope. Right now RunBot mostly walks in circles around a circular room, connected to the center by a boom. The next step is to develop a freestanding version, and the researchers don't anticipate major problems in achieving that, since the presence of the boom plays only a tiny role in RunBot's walking ability.
Once robots conquer walking and climbing, there's always dancing. You heard me. Scientists at the University of Tokyo in Japan are using video motion-capture systems to record the movement of dancers performing traditional Japanese folk dances, many of which are in danger of being lost. This data is then translated into a limb-motion sequence for humanoid robots, which can then perform those dances — kind of mechanical libraries. So far, the robots do okay in mimicking upper body movements, but they have trouble maintaining balance trying to perform some of the more intricate footwork. But they're still working on it, and I'd wager they'll crack this particular problem in due time. At which point, perhaps the Spousal Unit will receive a miniature dancing robot for Christmas capable of performing the Moon Walk, or the Macarena, or even something more ballroom-esque, such as the fox-trot or rhumba. We draw the line at David Levy's robotic vision for 2050, though. Some things should stay relegated to science fiction.