Fastest running robot in the world learns mountaineering

Researchers simulate neural bases of motion adaptation

Running robot Runbot © Bernstein Centers for Computational Neuroscience
Read out

He is the fastest of his kind, but in the meantime he can storm to summits: RunBot, among all the dynamic machines of the world record holder in fast-moving, has expanded his repertoire. With an infrared eye, the running robot recognizes whether there is a slope in front of it and adjusts its gait uphill with pinpoint accuracy. The neuronal basis for this adaptation has now been simulated by a team of scientists using a "learning" exercise program.

Just like a human, he leans his upper body forward and makes smaller steps. A team of scientists led by Prof. dr. Florentin Wörgötter has simulated the neural basis of this adaptive performance with the help of a "learning" movement program. The results of this research at the Bernstein Center for Computational Neuroscience at the University of Göttingen are presented by PLoS Computational Biology in its online edition of July 13, 2007.

Human gait is a marvel of coordination. The angle of the knee joints, the speed of the hip swing, the center of gravity of the upper body and many other elements of the movement must be precisely coordinated. In doing so, humans adapt to different external circumstances. On ice, it runs differently than on solid ground, uphill other than downhill.

Motion control hierarchically ordered

"The ability of the robot to switch from gait to gait in a flash without stumbling is based on the hierarchical organization of motion control, which is similar to that in humans, " explains Prof. Dr. med. Florentin Wörgötter, head of the research group at the Bernstein Center for Computational Neuroscience at the University of Göttingen. In a study now published in PLoS Computational Biology, researchers report their findings.

It turns out that the sequence of movements on the lower hierarchical levels is progressively pushed forward by peripheral sensors. Control circuits ensure that joints do not stretch, others take the next step as soon as the foot touches down. Only when the gait has to be adjusted does higher levels of organization intervene: In humans, it is the brain with the interaction of its strongly networked neurons. In the walking robot, the signal from the infrared eye triggers this adaptation process via a computer-based neural network, albeit a much simpler structure. display

To learn from mistakes

The hierarchical organization of the motion control makes it possible to achieve the change of gait by shifting a few parameters - the remaining sizes are automatically adjusted by the autonomous control circuits. On the first attempt to climb a mountain, RunBot tilts backwards. He has not yet learned how to react to what his "eye" perceives with an altered exercise program.

Similar to children, RunBot learns from his downfalls; In this way, the neural interconnection between eye and motion control is expanded. Only when this connection is present, step length and body posture can be controlled by the visually triggered signal. With a steep mountain, the movement program of the running robot is strong, with a shallow mountain only slightly changed.

(Bernstein Centers for Computational Neuroscience, 16.07.2007 - NPO)