Page - 47 - in Proceedings of the OAGM&ARW Joint Workshop - Vision, Automation and Robotics

IV. USER EVALUATION The goal was to explore if there is a difference in the UX between a robot with remote-HRI (robot A) and a technical revised version of this robot with physical-HRI (robot B). Both robots offered two control modes: remote control via touch-panel and direct-manual control via physical guidance. The touch-panel for remote control featured a graphical user interface consists of buttons to steer the robot and to save the taught movement trajectories. The physical-HRI mode enabled the operators to control the robotic arm directly, manually and without an additional intermediate layer. Robot A was optimized for remote control, whereas the improvement of robot B consisted of an extended physical HRI. Five participant were recruited to participate in the two studies. This small participant number can be sufficient to identify the most severe usability problems and was already discussed by [9]. The current study was conducted one year after the previous one. Within this time, robot A was upgraded to robot B, so robot B could only be examined after robot A. However, both studies had the same structure: (1) Introduction of the robot: Each participant was introduced to the robot and its control mechanisms. The participants were assigned the task to parameterize the process points in a predefined robot program. That means they had to bring the robot’s tool to a precise position above the screw and that they had to adopt the position parameters to a program in the UR-teach pendant (for robot A) or to the XROB-user interface. Fig. 1 shows the screw positions as process points. For process quality precision of the parameterization is crucial. As Fig. 5 points out especially lateral or orientation deviances are critical for process effectivity while vertical positioning could be effectively observed visually during teach in process. Fig. 5 - screwing process - error sources & real view In order to relief stress and increase compliance, the participants were assured that the focus of investigation was only the robot’s performance and there were no negative implications for them. (2) Conducting the user study: Each participant was audio- and videotaped with two cameras in order to generate a holistic perspective. This included a head mounted camera (first-person view - Fig. 6) and a hand camera (context oriented view). (3) Post-study questionnaires, including NASATLX, SUS, and self- developed items. The aim of the analysis was to compare the temporal demand, and the UX (including usability and performance expectancy) of the first and second version of the robot prototype. The findings are used for a the third (and last last) technical revision (design of the user interfaces of robot C, D and E) before the robot is deployed in the normal factory environment. The analysis of the video data (comments, reactions and feedbacks) consisted of (1) a rough clustering of all relevant issues, (2) a detailed description of their key features, and (3) overlapping topics were merged to categories or differentiated from each other. Fig. 6 – Head Mounted Device for gaze tracking - gaze tracking results V. RESULTS A total of five male assembly workers were recruited to participate in both studies (a representative sample for the factory with which we collaborated). The sample might be rather small but even for companies with several 1000+ employees it was difficult to find workers who work at a special part of the assembly line, predictively for the whole project duration (2 years+) who fulfill requirements (left- / right-handedness, age, robot training,…). Each participant was interviewed for 30 minutes and filled in demographic questionnaires afterwards. The mean age of the study participants was 45.4 (SD=5.7) and they had no prior experience with robotic systems. Four out of five participants had experience with computers and automated systems previous to the studies. The teaching using robot A yielded requirements regarding robot hand guidance. Gear friction yields stacking and imprecise movement. Locking of certain degrees of freedom (e.g. rotation or translation,…) is asked for by the users as well as semiautomatic tool alignment and expected to improve both programming time and process quality. A state of the art force torque sensor was integrated (in robot B) as well as buttons to call perpendicular realignment or locking of rotational or translational degrees of freedom. That should make the robots more effective. Additionally a RGB-D sensor as well as a 2D sensor for position deviation correction were added (see Fig. 7). Robot B was evaluated with exactly the same assignment of parameterization of the process points. The teaching duration using remote (robot A) and physical (robot B) control mode was extracted from the video recordings. Table I shows a decrease in average duration by 23.11%, and a strong shift from software- to 47