Inspired perhaps by the success of drone aircraft in Iraq and Afghanistan, the U.S. Department of Defense is funding projects to build machines that walk like us—machines that could carry loads, perform search and rescue, or even assist in combat.
One of the greatest challenges for a bipedal robot is navigating common obstacles like curbs and stairs. Keith Buffinton, a mechanical engineer at Bucknell University, recently received a $1.2 million Navy grant to tackle the problem. He video-taped students regaining their balance after a shove and realized that walking is actually a type of controlled falling, with each step an act of recovery. That insight inspired new algorithms for managing hips, knees, ankles, and toes. Collaborating with the Institute for Human and Machine Cognition in Florida, Buffinton’s team has built a partial bipedal robot (torso, legs, and feet) that could soon walk over simple obstacles, allegedly with better balance than a person. In 2010 the group will add arms and a head with stereoscopic vision.
Boston Dynamics, an offshoot of MIT, is taking a similar approach with Petman, which struck out this year on its first exploratory walks. This robot not only stands like a person but also simulates human gestures, body warmth, and—creepiest of all—it can sweat. Those traits are important to the Army, which wants to use Petman for testing chemical-protection gear in battlefield conditions starting in 2011.
Anybots of Mountain View, California, is trying something different with Dexter: It incorporates self-teaching software to help the robot learn how to walk. Dexter recently began its first tentative movements. That’s one small step for a robot, but if this approach succeeds, a giant leap for robotkind. It would certainly beat anything with wheels.