Cognitive Telepresence in Human-Robot Interactions

Abstract

Remotely operated semi-autonomous robots have great potential in many applications important to the economy and the environment in the Middle East. Applications include construction and maintenance of nuclear power plants and underwater oil wells, remote sensing for weather forecasting and pollution detection, and healthcare. Such tasks often require an operator without substantial technical expertise to remotely control a complex robot that is situated in an unknown and rapidly changing environment. In such case, the limited autonomy of the robot can cause failures that endanger the successful and efficient completion of the mission. In this research, we argue that cognitive telepresence (CT), that is, the ability of the user to comprehend and control the robot's cognition, is central to the reliability and robustness of robot's autonomy, which in turn, significantly increases the chance of success. Correcting the robot's behavior without switching to lower autonomy levels (manual teleoperation) will allow better exploitation of the currently available automation algorithms. To validate the applicability of our approach, we conducted experiments with human subjects. The subjects directed a simulated semi autonomous robot in a scenario that models a real-world application. The task was to disarm a minefield without entering into any of several threat zones in the environment. The simulated robot had all the major capabilities required for accomplishing the task. However, the robot's autonomy was imperfect, as it sometimes misclassified mines and failed to detect threat zones in the environment. Thus, the users had to be able to correct the robot's cognition to achieve mission success. This scenario, although hypothetical, illustrates the challenges that modern robotic applications must address in the real world. The tests were conducted on five systems that differed from each other only by the levels of CT achieved by the users. The results of the experiment indicate that CT as a design feature of human-robot interaction systems can improve the overall performance of human-robot teams. It can also reduce the workload of the user and, without any changes to the robots' artificial intelligence, allow the completion of tasks that were unattainable with systems that provide lower levels of CT.

Date of Award	2012
Original language	American English
Supervisor	Jacob Crandall (Supervisor)