News‎ > ‎

Machine learning and recovery procedure

posted Jan 18, 2011, 10:45 PM by Cesar Harada

I found the recovery procedure particularly inspiring : if protei tips over, certain sensors can be activated in order to start a tipping back procedure.

"While BigDog runs field trials (see article “Carrying loads on the four-legged walking robots“), his little brother – LittleDog – improving their skills in the laboratory. Both robots are the development of Boston Dynamics, which operates on a request of the U.S. Advanced Research Projects Agency of the Ministry of Defense (DARPA) and were originally intended for use by the military. However, now DARPA has selected six research teams from various universities, including Massachusetts and Stanford, to develop a more sophisticated control algorithms LittleDog, that later can be implemented and older brother. 

Each leg has three joints LittleDog with servo motors, with the robot controlled by the onboard computer.Built-in gyro helps the robot to navigate in space, and the external system monitors the accuracy of each limb and joint, when LittleDog moves. 

Gradually, scientists have corrected many of the shortcomings of the robot, it now moves much more confidently: moved through the dips in the ground, easily overcomes the rocky surface and not succumb to the stairs. The latest version LittleDog even become a self-learning, its programmed so that on their own experience to become more effective in navigating the difficult terrain, it has the ability to evaluate the difference between a sustainable footing and shaky, and depending on it to distribute the load on the limb."