A Newton method with always feasible iterates for Nonlinear Model Predictive Control of walking in a multi-contact situation

Diana Serra (1) , Camille Brasseur (2) , Alexander Sherikov (2) , Dimitar Dimitrov (2) , Pierre-Brice Wieber (2)

1 DIETI - Dipartimento di Ingegneria Elettrica e delle Tecnologie dell'Informazione [Napoli]
2 BIPOP - Modelling, Simulation, Control and Optimization of Non-Smooth Dynamical Systems

In this paper, we present a Nonlinear Model Predictive Control scheme, which is able to generate walking motions in multi-contact situations. Walking up and down stairs with an additional hand support is a typical example, which we address in simulation. Computing such a nonlinear control scheme is usually done with a Newton method, a potentially time-consuming procedure involving iterative linearizations. We propose here a Newton method which is specifically designed to provide at each iteration a feasible solution, always satisfying the (nonlinear) dynamic balance constraints. This results in a significant reduction in computation time, by minimizing the number of necessary iterations to reach a feasible solution.

Domaines: Robotique [cs.RO]; Optimisation et contrôle [math.OC]

Diana Serra, Camille Brasseur, Alexander Sherikov, Dimitar Dimitrov, Pierre-Brice Wieber. A Newton method with always feasible iterates for Nonlinear Model Predictive Control of walking in a multi-contact situation. IEEE-RAS 2016 - International Conference on Humanoid Robots (Humanoids), Nov 2016, Cancun, Mexico. pp.932-937, ⟨10.1109/HUMANOIDS.2016.7803384⟩. ⟨hal-01418402⟩