Seite - 134 - in Proceedings - OAGM & ARW Joint Workshop 2016 on "Computer Vision and Robotics“

4.4. Human-Robot InteractionandInformationExchange Current research on HRI investigates different ways how robots and humans interact, the main ones beingvoicecontrol [13], interactionprimitive [7],motion recognition [10], forceadaptation [17]and sharedpresence[30]. Howevernoneof thesemethods iswellestablished incommercialapplications. In this project, we focus on new paradigms of HRI in the context of collaborative industrial robotics and emphasize the distinction between collaboration and other forms of human-robot interaction, which usually view the problem as robot control or human-robot communication via tele-operation [11]. We include natural interaction mechanisms (acoustic and gestural interface), human factors [32] and a visualization component supported by augmented reality functionalities in an intelligent, context-sensitivecontrol system. Whilehumancollaboratorsprimarily interactusingspeechandges- tural input, they receive situation-dependent information about current and future tasks, the robot’s movement path and possible dangers. Here we explore the use of a head-worn AR system to ensure unobtrusive, hands-free collaboration and explore the optimal information flow to avoid cognitive load and distraction from the task. Moreover, the physical state of the human affects both the robot behavior and the feedback channel. This all strengthens the collaborative aspect of the interaction by increasing communication quality, trust, security awareness, and work efficiency. Recently, a first implementation of the interaction system was set up including the speech interface, a basic dialogue manager and a tablet application, serving as basic sensor interface. The human collaborator is thus able to interactwith therobotusingspeechinput, includingbasicrobot taskqueuemanipulationcom- mands, while monitoring the sensory data. As a next step, the acoustic interface will be extended to includenatural languageunderstanding together withamoreadvanceddialoguemanager. Moreover, a tablet application and wearable sensors will be used to analyse behaviors, emotions and actions of human collaborators. We will use the results of this analysis to implement the context-sensitive visualizationandcontrol system. 4.5. RedundantSensitiveRoboticManipulation Whenever a physical human-robot interaction is supposed to take place to fulfill a shared task, hu- man ergonomic operation and safety are important aspects which must be observed. Robot safety is provided by the electromechanical system in different ways and often in a redundant fashion. In practical terms, the main options to reduce the risk of human injuries are safety-related monitored stops, speed and separation monitoring, and power and force limitations, as manifested in [ISO/TS 15066:2016]. The implementationof theseoptions isdone(i)bymeasuring thedirectenergy transfer betweentherobotandanobjector (ii)bymonitoringtheenvironmentusingelectromagneticorsound basedsensors. Morespecifically, aphysicalcontactcanberecognizedbymeasuring theforce, torque orcurrentat theendeffector, the robot’sbase,orat each joint. Inaddition, a sensitiveskinappliedon themanipulatorcanmeasureacontact forceaswell. Ifadirectcontact isnotdesired, theenvironment can be perceived by different sensors operating at distance (see Section 4.1.). All sensor data can be fused to expand the knowledge of the environment, thus being used to control the robot’s movement in a human-safe manner. That means that the dynamic movement of the robot must be planned and executed inan adaptiveand reactiveway. If a kinematically redundant system has to perform a given task, the additional freedom can be used to enhance safety (by increasing the distance of the manipulator’s parts to a human or reducing the velocityof the robot’sarmsegments) andergonomics forahumanoperator (byconfiguring the robot joints in a way that the robot does not disturb ergonomic human motion). Such systems allow a 134