Abstract
Probabilistic model for balance recovery and biped locomotion
Walking and balance is a major issue in humanoid robotics. Most of the time this problem is treated using Zero Moment Point (ZMP) control. While this method works, it lacks the reactivity and the stability we found in humans. Moreover the speed of the walk is greatly limited by this approach.
In our case we looked at the notion of capture point introduced to cope with push recovery that requires a step (Pratt et al. 2006) . Here we extended this notion to allow walking and balance and tested it in simulations. We then applied this method in a Bayesian frame to deal with uncertainty of the model and of the sensors. It also grants the possibility to easily integrate other constraints such as obstacles or corridors.