Accepted Papers

Abstract: In order to achieve efficient collaboration during task completion in groups, temporal alignment is essential, i.e., synchronisation. We believe that efficient entrainment in mixed human-robot teams can positively affect human-robot collaboration. However, few studies have investigated how groups of humans entrain with each other to acquire new knowledge transferable to human-robot collaboration. This paper proposes a study design to get new insights into how dyads and triads of human workers entrain in assembly tasks simulating the industrial context. We argue that the investigation of both dyadic and non-dyadic (i.e., triadic) configurations is essential, as this will give us insights into how, and if, the complexity of reaching temporal synchronisation through entrainment increases with additional actors. Lastly, we propose a follow-up study investigating how the mechanisms utilised in human-human entrainment can be replicated in an industrial robot, ultimately improving human-robot collaboration in mixed teams.

Abstract: How does physical interaction with a real or virtual playmate affect the prediction behavior? In this study, we use the mirror game paradigm to observe the motion of human and virtual players in leader-follower modality with a leader and two followers. The main aim of this study is to observe the use of haptic information between players in the mirror game and propose a cognitive model for the virtual follower with haptic interaction in the game. It is shown with mathematical metrics that the haptic interaction between the followers enhance the tracking performance. The proposed virtual follower utilizes the haptic information to enhance the prediction of leader’s motion and this cognitive model shows a human-like behavior in the mirror game.

Abstract: The present paper outlines an approach based on joint action to overcome the Out Of The Loop (OOTL) phenomenon in HRI. Our proposed solution involves endowing artificial agents, specifically robots, with the same temporal adaptation and anticipation mechanisms that underlie mutual influence in human-human interactions. Specifically, we argue that independently from the motoric repertoire of the robot, its behaviour should be characterized by reactive period and phase correction together with temporal prediction mechanisms that facilitate precise coordination with the user. This will optimize the balance of self-other integration and segregation, allowing the human to stay in the loop during joint action with a robot.

Abstract: Despite the daily presence of joint actions involving collaboration with other people, very little is known about how we successfully collaborate with partners who possess different characteristics and skills. It is unclear how people attribute error and success to themselves and their partner in situations where both actors are noisy, and all information regarding each actor’s performance must be extracted from the common outcome of the joint action itself. The aim of the current study was to test whether participants were able to estimate the variance associated with the actions of a virtual partner, and attribute the relative weight of contribution to themselves and their partner in order to complete a joint task as precisely as possible. We developed a joint reaching task in which participants collaborated with three different simulated players: one with better performance than the participant’s, one with the same, and one with worse. In each collaboration, participants could assign a weight to their own contribution before performing the reaching task. They would then see the outcome of the joint action. We hypothesized that, if participants were able to behave optimally in these three conditions, they would assign the weight to their own relative contribution according to the perceived variance of their partner (e.g. they would assume more control with a worse partner). But the results provided a more complex situation, as two distinct behavioral strategies emerged: one “near-optimal strategy,” in which participants were able to adjust to the second player’s performance, and one “non-optimal strategy,” in which participants discarded the contribution of the second player and took the majority of control regardless of their partner’s ability. Such results deepen our understanding of behavioral strategies underlying joint action, while also contributing insights for the development of effective human-robot interaction applications.

Abstract: Joint action problems remain a difficult endeavor when social robots are involved. The level of patience that a person has when interacting with a robot has been shown to vary based on their expectations of the robot’s capabilities and the difficulty of the task. Social robots, especially humanoid ones, require an additional level of consideration when designing behaviors and interactions to meet such expectations. We aim to explore an approach rooted in the perceived embodiment of a humanoid robot to improve the interaction experience. This manuscript outlines previous work and inspirations for our development thus far. By utilizing a definition of embodiment, we have identified four mutual perturbatory channels that humanoid robots share with people. Through an experimental joint action scenario, we aim to quantify the bandwidth of these perturbatory channels, which we can use to further maximize the embodiment of the robot and improve the user-experience of interaction with the robot.

Abstract: When studying human-robot collaboration, people focus on improving robot policies to create fluent coordination with human teammates. However, the effect the environment has on human-robot joint action is often overlooked. To thoroughly explore environments that result in diverse behaviors, we propose a framework for procedural generation of environments that are (1) stylistically similar to human-authored environments, (2) guaranteed to be solvable by the human-robot team, and (3) diverse with respect to coordination measures. We analyze the procedurally generated environments in the Overcooked benchmark domain via simulation and an online user study. Results show that the environments result in qualitatively different emerging behaviors and statistically significant differences in collaborative fluency metrics, even when the robot runs the same planning algorithm.