Geometric Task Networks: Learning Efficient and Explainable Skill Coordination for Object Manipulation

Complex manipulation tasks can contain various execution branches of primitive skills in sequence or parallel under different scenarios. Manual specifications such branching conditions and associated skill parameters are not only error-prone due to corner cases, but also quickly untraceable given a large number objects skills. On the other hand, learning from demonstration has increasingly shown be an intuitive effective way program for industrial robots. Parameterized representations allow generalization over new scenarios, which however makes planning process much slower thus unsuitable online applications. In this article, we propose hierarchical compositional framework that learns geometric task network (GTN) exhaustive planners, without any manual inputs. A GTN is goal-dependent graph encapsulates both transition relations among constraints underlying these transitions. This improve dramatically offline efficiency, performance, transparency decision process, by leveraging task-parameterized models. We demonstrate approach on 7-DoF robot arm simulation hardware solving tasks.