"You Might Like It": How People Respond to Small Talk During Human-Robot Collaboration

TL;DR

The paper investigates how people respond to small talk from a non-anthropomorphic robot manipulator during collaborative assembly, revealing substantial engagement and a subset of user-initiated conversations. Using a Wizard-of-Oz setup with a Panda robot and a negative-feedback manipulation in a lab study (N=20), the authors show that most participants reciprocate small talk and continue conversations, though initial negative feedback reduces the likelihood of extending dialogue. Key contributions include a detailed behavioral taxonomy for initial reactions, engagement, and feedback responses, plus design implications for integrating small talk into industrial robots without compromising task performance. The findings highlight both the potential to enhance trust and cohesion through social cues and the risk of distraction, underscoring the need for adaptive, user-aware conversational strategies in physical HRI settings.

Abstract

Social communication between people and social robots has been studied extensively and found to have various notable benefits, including the enhancement of human-robot team cohesion and the development of rapport and trust. However, the potential of social communication between people and non-social robots, such as non-anthropomorphic robot manipulators commonly used in work settings (\eg warehouse and factory), is less explored and not well established. In this work, we investigate people's engagement and attitudes towards a non-anthropomorphic robot manipulator that initiates small talk during a collaborative assembly task and explore how the presence of negative team feedback may affect team dynamics and blame attribution. Through an in-person study with 20 participants, we observed a response rate of 77.60% in response to the robot's small talk attempts. Nine participants continued engaging with the robot by initiating their own questions, indicating sustained interest in the conversation. However, we also found that the first negative feedback decreased the participants' willingness to extend the conversation. We additionally present participants' initial perceptions of small talk for physical robot manipulators and discuss design implications for integrating small talk into non-social robots, along with various aspects of small talk that may influence physical human-robot interactions.

Figures (5)

• Figure 1: The experimental setup. The robot can grab a pipe from the pipe dispenser and place it on a pipe ramp, which delivers it to the participant. The participant can grab a joint from the joint containers on the left and place a faulty joint on the joint ramp for it to be retrievable by the robot.
• Figure 2: An overview of the collaborative tasks the participants completed with the robot. The team negative feedback was introduced halfway through the first and second task structures. To analyze changes in user behavior across instances of negative feedback, we delimit observation windows with respect to the negative feedback announcements. Window A is considered our baseline as this corresponds to the only portion of the interaction that has not yet been influenced by any negative feedback statements. In the lower half of the figure, we include an example of one user's conversational interaction with the robot after the first negative feedback announcement (P10).
• Figure 3: Different aspects of participants' response to the first robot small talk initiation: (1) conversational reciprocation, (2) task disruption, and (3) affective reaction. For each aspect, the states are mutually exclusive. The numbers denote participants who exhibited the corresponding responses.
• Figure 4: The conversation extension rate is a ratio of the number of participant responses and participant-initiated questions made in direct response to a robot small talk statement, over the total number of robot questions and small talk statements. (a) The average conversation extension rate across all participants over three separate time windows: Window A) before first negative feedback, Window B) between first and second negative feedback, Window C) after second negative feedback. A significant difference was found between the mean conversation extension rates of Windows A and B. (b) Scatter plot depicting the conversation extension rate for each individual participant across their entire interaction.
• Figure 5: Examples of positive and negative user interaction with the robot small talk behaviors. The top row shows a positive example where the participant expressed genuine interest in learning more about the robot (P7). The bottom row is an example of a user who perceived the interaction negatively and exhibited a potential mental model mismatch of expectations (P12).