Systems with legs and arms are becoming increasingly useful and applicable in real world scenarios. So far, in particular for locomotion, most control approaches have focused on using simplified models for online motion and foothold generation. This approach has its limits when dealing with complex robots that are capable of locomotion and manipulation. In this presentation I will show how we apply MPC for locomotion and manipulation with different variants of our quadrupedal robot ANYmal. In contrast to traditional methods, we jointly optimize for footholds and body motion while accounting for the full system kinematics and centroidal dynamics. Moreover, I will present some insights how we ensure constraints and synthesize complex manoeuvres with online and offline methods.
