An adjustable autonomy paradigm for adapting to expert-novice differences

Know thy user: Designing human-robot interaction paradigms for multi-robot manipulation

“…The robot's autonomy can be adjusted intelligently by encoding the user's interruption preferences as rewards, allowing the robot to make optimal use of the human's time [25]. Here, we evaluate an adaptive user interface that modifies the robot's autonomy based on a user model learned during the calibration period [17]. Muszynski and Behnke [20] presented an interface for personal service robots that allowed the users to select the appropriate level of autonomy based on their perceived workload.…”

“…Chen and Barnes's study on multi-robot supervisory control indicates that individual differences in spatial ability, attentional control and video gaming can affect the quality of humanrobot interaction [3]. In our previous work [17], we created a user modeling system specifically for this scenario, in which short teleoperation sequences were used to classify user expertise levels on three aspects of the task: navigation, manipulation, and coordinating the robots. Here, we use the same modeling paradigm to predict which users are likely to prefer an adaptive version of the interface that modifies robot autonomy vs. a configurable version in which the user creates autonomous behavior macros that can be activated on command.…”

“…In this paper, we evaluate an adaptive version of the interface that adjusts the robot's autonomy based on the user's expertise [17] vs. a configurable version that allows the user to create macros for autonomous behaviors using a simple learning from demonstration paradigm [16]. The operator interacts with the robots through a combination of arm gestures, tracked using a Microsoft Kinect sensor, plus gamepad controls executed with a Nintendo Wii remote.…”

“…In our delivery task, the human driver commands a pair of robots to move stacks of objects from starting piles to the goal location in a lightly cluttered environment [15]- [17]. Each robot is composed of a non-holonomic wheeled base and two 3-DOF arms for lifting objects.…”

“…with good user modeling mechanisms [17]. Learning from demonstration (LfD) has been successfully used in many robotics applications to create policies based on a set of demonstrations provided by the user [1].…”