The goal of this paper is to present a real-time controller for an omnidirectional wheeled humanoid robot which can be strongly disturbed and tilt around its wheels. It is based on two linear model predictive controllers, managed by a tilt supervisor, which detects changes of the dynamic model caused by the tilt of the robot. Experimental results are proposed on the robot Pepper made by Aldebaran, showing good performance in term of stability an robustness.

While realizing a task, human-beings are able to use grasping inside their environment in order to keep the most stable balance. Although such a behavior is quite natural for humans, it is very difficult to find the best formulation to adapt human motion to humanoid robots. This paper proposes a conceptually simple framework of human posture control based on optimization which takes into account grasp and friction and achieves robustness against external disturbances. A new stability criteria is also introduced. Contrary to most other approaches, our method deals not only with unilateral contacts with friction but also with bilateral grasps. This allows for arbitrarily pulling, pushing or twisting on a handhold. Additionally, and in contrast to classical methods based on ZMP, our method also accounts for contacts not being all in the same plane.