Learning Competing Constraints and Task Priorities from Demonstrations of Bimanual Skills

As bimanual robots become increasingly popular, learning and control algorithms must take into account new constraints and challenges imposed by this morphology. Most research on learning bimanual skills has focused on learning coordination between end-effectors, exploiting operational space formulations. However, motion patterns in bimanual scenarios are not exclusive to operational space, also occurring at the joint level. Moreover, bimanual operation offers the possibility to carry out more than one manipulation task at the same time, which in turn introduces the problem of task prioritization in bimanual settings. Here we address the aforementioned problems from a robot learning perspective. We go beyond operational space and present a principled approach to simultaneously learn operational and configuration space constraints, as well as the evolution of task priorities from demonstrations. Our method extends the Task-Parameterized Gaussian Mixture Model (TPGMM) to the use of projection operators which allow for tackling such problems. The approach is validated in two different bimanual tasks with the COMAN and WALK-MAN humanoids that either require the consideration of constraints in both operational and configuration spaces, or the prioritization of tasks.