Cognitive Reasoning for Compliant Robot Manipulation

Abstract

Physically compliant contact is a major element for many tasks in everyday environments. A universal service robot that is utilized to collect leaves in a park, polish a workpiece, or clean solar panels requires the cognition and manipulation capabilities to facilitate such compliant interaction. Evolution equipped humans with advanced mental abilities to envision physical contact situations and their resulting outcome, dexterous motor skills to perform the actions accordingly, as well as a sense of quality to rate the outcome of the task. In order to achieve human-like performance, a robot must provide the necessary methods to represent, plan, execute, and interpret compliant manipulation tasks. This dissertation covers those four steps of reasoning in the concept of intelligent physical compliance. The contributions advance the capabilities of service robots by combining artificial intelligence reasoning methods and control strategies for compliant manipulation. A classification of manipulation tasks is conducted to identify the central research questions of the addressed topic. Novel representations are derived to describe the properties of physical interaction. Special attention is given to wiping tasks which are predominant in everyday environments. It is investigated how symbolic task descriptions can be translated into meaningful robot commands. A particle distribution model is used to plan goal-oriented wiping actions and predict the quality according to the anticipated result. The planned tool motions are converted into the joint space of the humanoid robot Rollin' Justin to perform the tasks in the real world. In order to execute the motions in a physically compliant fashion, a hierarchical whole-body impedance controller is integrated into the framework. The controller is automatically parameterized with respect to the requirements of the particular task. Haptic feedback is utilized to infer contact and interpret the performance semantically. Finally, the robot is able to compensate for possible disturbances as it plans additional recovery motions while effectively closing the cognitive control loop. Among others, the developed concept is applied in an actual space robotics mission, in which an astronaut aboard the International Space Station (ISS) commands Rollin' Justin to maintain a Martian solar panel farm in a mock-up environment. This application demonstrates the far-reaching impact of the proposed approach and the associated opportunities that emerge with the availability of cognition-enabled service robots.