Study on the effects of robot behaviors and their interactions on human trust

Date

2022-12-21

Authors

Horn, Matthew William

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

This dissertation considers the problem of integrating three different robot behaviors in different combinations and settings to ascertain their effect on human trust. A system is developed to optimally combine the different behaviors such that any negative combinations are avoided. Three non-verbal robot behaviors are tested: a legibility pathing behavior, a usability avoidance behavior, and a usability eye contact behavior. Three user studies are performed to ascertain behavior interactions in transportation tasks in differing environments with multiple possible goal locations. These studies are accomplished online through videos of tasks performed in a simulated environment. The study into these Human-Robot Interactions is important because as technology progresses, robots are being introduced into more spaces that are sometimes shared with human workers. These areas may necessitate the robot to make cooperative actions, show intent, or keep nearby people safe from hidden hazards which are prevalent in nuclear or chemical facilities such as radiation, radioactive particles, odorless fumes, strong chemicals stored in beakers, etc. When robots are introduced to such environments, it is necessary that people feel trust in the robot and feel safe around them. Three behaviors are studied in a simulated environment on the TIAGo++ platform. These behaviors are combined and evaluated to ascertain complementary behaviors and adversarial behaviors and their variables that trigger these interactions. The TIAGo++ platform is simulated in Gazebo, a ROS-enabled simulation platform to create three common scenarios a person might face when robots are moving near them to transport items. Results are varied between combinations of the behaviors, and depend on the variables used to instruct the behaviors' motions. A set of trust-based questions, ranking questions, and pathing testing is utilized to tune the behaviors on a final system to avoid negative behavior interactions. The final tuned simulated system has been found to negate the negative associated behavior interactions in a new environment versus other partially tuned systems. The robot system that always takes the safest path away from workers was also found to not trigger negative behavior combinations, however, did not score as well in user studies as the final tuned smart system. The complete quantitative results are presented and analyzed in this report.

Description

LCSH Subject Headings

Citation