Real-Time Motion Generation for Mobile Manipulators via NMPC with Balance Constraints

Abstract

We present a novel real-time motion generation approach for mobile manipulators which maintains balance even when the robot is called to execute aggressive motions. The proposed approach is based on Nonlinear Model Predictive Control (NMPC) and uses the robot full dynamics as prediction model. Robot balance is maintained by enforcing a constraint that restricts the feasible set of robot motions to those generating non-negative moments around the edges of the support polygon. This balance constraint, inherently nonlinear, is linearized using the NMPC solution of the previous iteration. In this way we facilitate the solution of the NMPC and we achieve real-time performance without compromising robot safety. We validate our approach in scenarios of increasing difficulty and compare its performance with two other methods from the literature. The simulation results show that our method can generate motions that maintain balance in challenging situations where the other techniques fail.