Autonomous and Intrinsically Motivated Robots for Sustained Human-Robot Interaction

Abstract

A challenge in using fully autonomous robots in human-robot interaction (HRI) is to design behavior that is engaging enough to encourage voluntary, long-term interaction, yet robust to the perturbations induced by human interaction. It has been repeatedly argued that intrinsic motivations (IMs) are crucial for human development, so it seems reasonable that this mechanism could produce an adaptive and developing robot, which is interesting to interact with. This thesis evaluates whether an intrinsically motivated robot can lead to sustained HRI. Recent research showed that robots which ‘appeared’ intrinsically motivated raised interest in the human interaction partner. The displayed IMs resulted from ‘unpredictably’ asking a question or from a self-disclosing statement. They were designed with the help of pre-defined scripts or teleoperation. An issue here is that this practice renders the behavior less robust toward unexpected input or requires a trained human in the loop. Instead, this thesis proposes a computational model of IM to realize fully autonomous and adaptive behavior generation in a robot. Previous work showed that predictive information maximization leads to playful, exploratory behavior in simulated robots that is robust to changes in the robot’s morphology and environment. This thesis demonstrates how to deploy the formalism on a physical robot that interacts with humans. The thesis conducted three within-subjects studies, where participants interacted with a fully autonomous Sphero BB8 robot with two behavioral regimes: one realizing an adaptive, intrinsically motivated behavior and the other being reactive, but not adaptive. The first study contributes to the idea of the overall proposed study design: the interaction needs to be designed in such a way, that participants are not given any idea of the robot’s task. The second study implements this idea, letting participants focus on answering the question of whether the robots are any different. It further contributes ideas for a more ‘challenging’ baseline behavior motivating the third and final study. Here, a systematic baseline is generated and shows that participants perceive it as almost indistinguishable and similarly animated compared to the intrinsically motivated robot. Despite the emphasis on the design of similarly perceived baseline behaviors, quantitative analyses of post-interaction questionnaires after each study showed a significantly higher perception of the dimension ‘Warmth’ for the intrinsically motivated robot compared to the baseline behavior. Warmth is considered a primary dimension for social attitude formation in social cognition. A human perceived as warm (i.e. friendly and trustworthy) experiences more positive social interactions. The Robotic Social Attribute Scale (RoSAS) implements the scale dimension Warmth for the HRI domain, which has been validated with a series of still images. Going beyond static images, this thesis provides support for the use and applicability of this scale dimension for the purpose of comparing behaviors. It shows that participants prefer to continue interacting with the robot they perceive highest in Warmth. This research opens new research avenues, in particular with respect to different physical robots and longitudinal studies, which are ought to be performed to corroborate the results presented here. However, this thesis shows the general methods presented here, which do not require a human operator in the loop, can be used to imbue robots with behavior leading to positive perception by their human interaction partners, which can yield sustained HRI.