The human gait is a marvel of coordination. All aspects of movement control – from the angle of the knee joints to the momentum of the hip up to the balance point of the torso – need to be meticulously adjusted. In addition, the gait is adaptable to different environments. Walking on ice is different from walking on solid ground, walking uphill is different from downhill.
In their study, publishing in PLoS Computational Biology July 13, 2007, scientists around Florentin Wörgötter, Bernstein Center for Computational Neuroscience at the University of Göttingen, have simulated the neuronal principles that form the basis of this adaptivity in a walking robot. "RunBot", as it is called, lives up to its name – it holds the world record in speed walking for dynamic machines. Now its inventors have expanded its repertoire. With an infrared eye it can detect a slope on its path and adjust its gait on the spot. Just as a human, it leans forwards slightly and uses shorter steps. It can learn this behavior using only a few trials.
The robots ability to abruptly switch from one gait to the other is due to the hierarchical organization of the movement control. In this respect, it resembles that of a human and can hold as a human model. On the lower hierarchical levels, movement is based on reflexes driven by peripheral sensors. Control circuits ensure that the joints are not overstretched or that the next step is initiated as soon as the foot touches the ground. Only when the gait needs to be adapted, higher centers of organization step in – a process triggered by the human brain or, in case of the robot, by its infrared eye leading on to a simpler neural network. Because of the hierarchical organization adjustment of the gait can be achieved by changing only a few parameters. Other factors will be automatically tuned through the regular circuits.
At its first attempt to climb a slope, RunBot will fall over backwards, as it has not yet learned to react to its visual input with a change in gait. But just like children, RunBot learns from its failures, leading to a strengthening of the contact between the eye and the sites of movement control. Only once these connections are established, step length and body posture are controllable by the visually induced signal. The steeper the slope, the stronger RunBot will adapt its gait.
CITATION: Manoonpong P, Geng T, Kulvicius T, Porr B, Wo¨ rgo¨ tter F (2007) Adaptive, fast walking in a biped robot under neuronal control and learning. PLoS Comput Biol 3(7): e134. doi:10.1371/journal.pcbi.0030134
CONTACT:
Prof. Florentin Woergoetter
Computational Neuroscience
University of Goettingen
Bunsenstr. 10 (at the MPI)
D-37073 Goettingen
Germany
Phone: +49 (0)551 5176528
FAX: +49 (0)551 5176449
email: worgott@bccn-goettingen.de
web: http://www.ifi.informatik.uni-goettingen.de/cng/
PLEASE MENTION THE OPEN ACCESS JOURNAL PLoS COMPUTATIONAL BIOLOGY (www.ploscompbiol.org) AS THE SOURCE FOR THIS ARTICLE AND PROVIDE A LINK TO THE FREELY AVAILABLE TEXT. THANK YOU.
PLoS Computational Biology is an open access, peer-reviewed journal published weekly by the Public Library of Science (PLoS) in association with the International Society for Computational Biology (ISCB).
Journal
PLoS Computational Biology