Online adaptation of reference trajectories for the control of walking systems

A simple and widely used way to make a robotic system walk without falling is to make it track a reference trajectory in one way or another, but the stability obtained this way may be limited and even small perturbations may lead to a fall. We propose here a series of heuristics to improve the stability that can be obtained from such a tracking control law, through an online adaptation of the choice of the reference trajectory being tracked. Encouraging simulations are obtained in the end on a simple planar biped model.