Synthesis of a Predictive Push-Recovery Controller: Simulation Results on the iCub Humanoid Robot

When balancing, a humanoid robot can be easily subjected to unexpected disturbances due to external pushes or modeling errors. In these circumstances, reactive movements become a necessary requirement in order to avoid potentially harmful falling states. In our previous work, we designed and implemented on the real platform, a strategy based on simple models. In this paper we conceive a Model Predictive Controller which enables the prediction of future evolutions of the robot, taking into account constraints switching when performing a step. The control inputs computed by this strategy, namely the desired contact wrenches, are directly obtained on the robot through the momentum-based whole-body torque controller implemented on iCub. The proposed approach is validated through simulations, revealing higher robustness and reliability when executing the recovery strategy.