Abstract
In the context of legged robotics, stability (or equilibrium) is of the utmost importance. Indeed, as legged robots have a non-actuated floating base they can fall. To avoid falling, we must be able to tell apart stable from non-stable motion. This thesis approaches stability from a reduced model point-of-view: our main interest is the Center of Mass. We show how to compute stability regions for this reduced model, at first based on purely static stability. Although purely geometrical in nature, we show how they depend on the admissible contact forces. Then, we show that taking into account robustness, in the sense of acceleration (or contact forces) affordances transforms the usual two-dimensional stability region into a three dimensional one. To compute this shape, we introduce novel recursive algorithms. We show how we can apply computer graphics techniques for shape morphing in order to continuously deform the aforementioned regions. This allows us to approximate changes in the parameters of those shapes, but also to interpolate between shapes. Finally, we exploit the effective decoupling offered by the explicit computation of the stability polyhedron to formulate a linear, minimal jerk model-predictive control problem. We also propose another linear MPC problem that exploits more of the available dynamics, but at an increased computational cost. We then adopt a hierarchical approach, and use those CoM results as input to our whole-body controller. Results are demonstrated on real hardware and in simulation.