In this lecture, we will begin by revisiting the physics of legged locomotion, how this physics impacts stability analysis, and what sort of solutions have been proposed for making humanoid robots walk during the last decades. We will observe that the walking motion control that power most of the existing humanoid robots appear to be all variants of a common structure, which is that of Model Predictive Control. We will then spend some time discussing computational aspects, focusing especially on efficient ways to solve the Quadratic Programs that appear at the core of walking motion control, with an insider's view on the new algorithm currently developed in Aldebaran Robotics, which should be distributed in the next versions of NaoQi and will be available in open source.
Universität Bonn, Institute for Computer Science, Autonomous Intelligent Systems