Image: PETER DILWORTH
To the high-pitched hissing sound of electric servos, Troody slowly tilts forward. Fourteen tiny motors in her limbs begin to hum as her shins lift up from her feet. She stretches her legs and rises to an upright position. Still wobbling back and forth a bit, she takes her first timid step forward, then another¿and then she is walking.
That scene played out for the first time this past spring, when Troody¿a 10-pound, two-legged robotic replica of a Troodon dinosaur from the Cretaceous period¿took her maiden stroll through Peter Dilworth's lab. As scientists continue to argue over whether it will ever be possible to clone a dinosaur ¿ la Jurassic Park, Dilworth¿a research scientist at the Massachusetts Institute of Technology's Leg Lab, a part of M.I.T.'s Artificial Intelligence Laboratory¿simply built one himself. He chose to model a Troodon for its small size and big brain: "The ratio of its brain size to its body weight makes it the brainiest of the dinosaurs," Dilworth says, "so I thought that was kind of interesting."
His objective was straightforward: "It was for the robot to stand up from a sitting position, to walk and possibly run, go back to walking and stop and then sit down and be stable, and not fall over in any phase," Dilworth explains. But as he found out, making a robot walk is a lot more challenging than it seems. "It has proven extremely difficult to make bipedal robots," says Gregory Paul, an independent dinosaur paleontologist who collaborated with Dilworth on the project. "People assume that flight is hard to do while walking is easy; actually it's the reverse."
With one look at Troody's smooth movements, however, it is clear that the researchers have succeeded. "A lot of people comment on how it [Troody] looks very biological," Dilworth says. "That's one thing that I strive for, and it's a result of the way we do the control system in the robot, a technique which uses springs in series with the motors, which softens the way the robot feels." Indeed, Troody the Troodon, like all other robots at the Leg Lab, uses so-called electric series-elastic actuators, devices invented by Lab director Gill Pratt and one-time graduate student Matthew Williamson.
Before the development of these electric series-elastic actuators, researchers relied on hydraulic actuators to create walking robots (see sidebar). In comparison, the electric actuators are lighter and require far less power, making it possible to design more self-contained machines. But they, too, have their drawbacks¿including low torque and thus a need to run very fast to deliver high power. That in turn creates a need for gears to allow the robot to move slowly. The gears, however, lower the robot's tolerance for physical shocks¿which isn't a good thing for a machine that's supposed to walk.
Pratt's team solved the problem by adding a spring between each of a robot's joints and respective gears, thus buffering the blow. The springs also became a vital part of the robots' control systems. Traditionally, most robots had control systems that only determined the position of their arms or legs in relation to their environment. "You couldn't really tell if you hit the wall if the only thing you could measure was the position of the arm," Dilworth explains. "If your position is off by a little bit, you might be stopped a little before the wall, or, even worse, you might be stopped a little inside the wall and you'd have these massive torques on your actuator, trying to position the arm inside the wall."
Image: PETER DILWORTH
In the spring-actuator design, sensors measure the force acting on the springs. "You take your arm and you'd command it to apply a force in the direction of the wall," Dilworth continues. "So if the arm is floating in the air, it'll just move towards the wall because it's pushing into that direction. Once it hits the wall, it'll load up until it's applying the desired force you told it to and just rest up against the wall with that amount of force. If you were to grab the arm, pull back and let it go, it would flop back against the wall again, which is analogous to what a person or an animal would do. So that shows how force control is a very good way to do biologically inspired robots."
But force control alone is not enough to keep a robot on its feet. "There is a gyroscope that gives you pitch and roll and linear velocity and acceleration," Dilworth adds. "That sort of tells you which way is up. The robot, of course, is kinematic, so it knows all of its joint angles. It knows all of its masses, so it knows where its center of mass is, and the robot can compute various types of balance strategies." To improve the biped's walking, Dilworth even designed an additional device, which he calls a PQ control system: "It's a control system that looks at what the state of the robot is kinematically and figures out from that where it should apply forces with its feet to remain stable while it's walking."
Although these control systems are very advanced, Dilworth is quick to clarify that Troody has no built-in intelligence. "The robot really is an RC car," he says. "There is a joystick, and you say forward, right and left. The legs are really equivalents of wheels on a car-just a way to travel around and move forward. She's walking blindly. If she stubs her toe on something so she can't move her leg forward, then she might tip over. If the ground is just higher, then the control system is robust enough to deal with that, but it's not a high-level intelligence."
It took four years of experimenting with different designs before Dilworth finally got Troody up and walking. Now he hopes that museums, theme parks and other institutions may want to exhibit such a dinosaur robot, and to that end, he is building a larger version of Troody that will sport feathers and other lifelike details. "Since dinosaurs are extinct and no longer something you can see, I figured this would be a possible opportunity to try and represent one in a way that hadn't been done before," Dilworth says. "If you have a robotic dinosaur walking around, you may get a new sense of what they looked like in a unique way."
Dan Stone, the founder of Dinodiscovery¿a company that brings dinosaur exhibits to schools¿certainly feels that way. "Many of the children do not have an opportunity to go to the museums, so in that manner we can bring the exhibit to them," Stone says. "We have an educational program that enhances the curriculum that the kids are involved with in their schools, and it's just something that helps generate more excitement for the field of science." As for Troody, he says, "Kids would love it. The movement would be very educational and exciting for them."
Legged robots are still rather primitive and thus don't have a lot of real-world uses right now, apart from appearing in exhibits. "I tried to think of a way that I could work on a legged robot and have it actually have some type of commercial potential without having to think decades into the future," Dilworth says. "As for a robotic dinosaur, really all it has to do is just walk slowly from point A to point B on a flat ground and people will be perfectly happy with that." So far, the response to Troody has been very positive. "There is a great interest in both the robot aspect of it and the dinosaur aspect of it," Dilworth notes, "so the combination of the two was kind of a winning combination."