Learning context-adaptive task constraints for robotic manipulation

Constraint-based control approaches offer a flexible way to specify robotic manipulation tasks and execute them on robots with many degrees of freedom. However, the specification task constraints their associated priorities usually requires human-expert often leads tailor-made solutions for specific situations. This paper presents our recent efforts automatically derive constraint-based robot controller from data adapt respect previously unseen situations (contexts). We use programming-by-demonstration approach generate training in multiple variations (context changes) given task. From this we learn probabilistic model that maps context variables respective soft priorities. evaluate 3 different dual-arm an industrial show it performs better than comparable reproduction accuracy contexts.