Section: New Results

Robust Model Predictive Control for biped walking motion generation

Participants : Pierre-Brice Wieber, Diana Serra, Alexander Sherikov, Dimitar Dimitrov.

One of the main sources of nonlinearity in the Newton and Euler equations of motion of biped walking robots lie in the vertical motion of the Center of Mass. We proposed last year an approach that considers this nonlinearity as an uncertainty, in what would else be a linear system. We proposed then to use a robust linear MPC approach accordingly. The use of a linear approach allows fast computations to generate walking motions online. This year, we further developed this approach, by adapting the bounds on the uncertainty at each iteration of a Newton scheme, when solving the original nonlinear problem [35]. By using a robust approach within a Newton scheme, every iteration can be ensured to satisfy all dynamic constraints, so that we can limit the number iterations depending on the available computing power and always obtain a feasible solution. We also developed this year an application of this MPC approach to cases of collaborative carrying of heavy objects with a human partner [32].