All the news that's fit to assimilate[ Home | Blogs | Events | Robots | Humans | Projects | Podcasts | About | Account ]
|Target Environment||Locomotion Method|
|Sensors / Input Devices||Actuators / Output Devices|
|6 axis gyro/accelerometer
|12 pneumatic cylinders|
|Control Method||Power Source|
|Autonomous||Power and compressed air through tether|
|CPU Type||Operating System|
|Python and C++||45 Kg|
|Time to build||Cost to build|
|URL for more information|
Dexter is quite different from other robots that have walked on two legs. The Honda ASIMO and related robots use a walking algorithm called Zero Moment Point or ZMP, a geometrical constraint that guarantees stability. To use this approach, a robot must have stiff joints (driven by geared servo motors) and fairly large feet. In the simplest version, the robot is given pre-planned movements that guarantee that a perpendicular drawn from the center of whichever foot is on the floor passes through the center of gravity, with some compensation when it accelerates. Such a robot does not need active balance feedback to walk. While the most advanced ZMP-based robots do include active balance control to adapt to sloped floor surfaces or external forces, this is a refinement to a passively stable gait.
Dexter has a different, more human-like body on which ZMP control does not work. Its joints, driven by air cylinders, are springy and flexible like human muscle. There are no stable postures that it can be put in where it can balance without active feedback, so it has to constantly adjust based on its sense of balance, the robot equivalent to your inner ear. It walks and balances the same way humans do, even wearing the same shoes humans wear.
Dexter's harder-to-control body has major advantages in the real world. It can walk just as easily on soft surfaces, like the deep carpet shown in the video, as on hard surfaces. Because its joints are flexible and able to absorb impact, it will be able to run at high speed over uneven ground and jump over obstacles. If it accidently steps on your toe, it won't hurt any more than a person stepping on your toe. But most importantly, because there is no geometrical principle by which we could have programmed a walking motion, it had to learn to walk. Its learning software will soon lead to a much wider range of walking abilities than could ever have been programmed.