Complex Task Planning for General-Purpose Service Robots

Service robots, in contrast to industrial robots, are devices aimed to operate in the service sector in replacement of or collaboration with human performers—in particular, domestic service robots carry out daily household chores. Recently, their popularity has increased and the range and difficulty of the activities they can perform are reaching amazing performances. There have been proposed several systems that can generalize single tasks such as object manipulation or scene understanding but they still fail on complex task planning, i.e., when and where they can perform such tasks. In this work, we propose the use of expert systems as a core module to understand user commands, infer the task context, request missing information, and plan action where each step of the plan may consist of basic state machines to generalized deep-learning models.