Planning perception and action for cognitive mobile manipulators

Abstract

We present a general approach to perception and manipulation planning for cognitive mobile manipulators. Rather than hard-coding single purpose robot applications, a robot should be able to reason about its basic skills in order to solve complex problems autonomously. Humans intuitively solve tasks in real-world scenarios by breaking down abstract problems into smaller sub-tasks and use heuristics based on their previous experience. We apply a similar idea for planning perception and manipulation to cognitive mobile robots. Our approach is based on contingent planning and run-time sensing, integrated in our "knowledge of volumes" planning framework, called KVP. Using the general-purpose PKS planner, we model information-gathering actions at plan time that have multiple possible outcomes at run time. As a result, perception and sensing arise as necessary preconditions for manipulation, rather than being hard-coded as tasks themselves. We demonstrate the effectiveness of our approach on two scenarios covering visual and force sensing on a real mobile manipulator.

Global navigation approach for assistant robot

“…However, in our case there is no need to build a map, because it is possible to use an external camera. On the other hand, the case of manipulator robot must take into account self-collisions between its links while its end effector is going to the goal (Gaschler, Nogina, Petrick & Knoll, 2014), so 3D path planning algorithms are proposed too.…”