A proposal for a global task planning architecture using the RoboEarth cloud based framework

As robotic systems become more and more capable of assisting in human domains, methods are sought to compose robot executable plans from abstract human instructions. To cope with the semantically rich and highly expressive nature of human instructions, Hierarchical Task Network planning is often being employed along with domain knowledge to solve planning problems in a pragmatic way. Commonly, the domain knowledge is specific to the planning problem at hand, impeding re-use. Therefore this paper conceptualizes a global planning architecture, based on the worldwide accessible RoboEarth cloud framework. This architecture allows environmental state inference and plan monitoring on a global level. To enable plan re-use for future requests, the RoboEarth action language has been adapted to allow semantic matching of robot capabilities with previously composed plans