Generalized Plan Design For Autonomous Mobile Manipulation in Open Environments

Autonomous robotic agents acting in open environments have to master situations and action effects they did not anticipate. To deal with these issues we propose an information processing concept for plan interpretation that is based on three concepts: • statically defined, and dynamically inferred knowledge, • context-definition on a task level to implicitly reparameterize all sub-actions during a task, • and control structures for plans that can monitor plan execution for unexpected events and respond appropriately. Implications of plans realized according to these design patterns are explained on the example of a reasoning intense autonomous manipulation task.