Abstract
Nonverbal communication is important for attaining fluent interaction between people and robots. In the present work, we explore the potential benefits of adding nonverbal cues in the form of dynamic LED light to a hexapod robot. The effects of the added light signals on evaluations of the robot’s sociality and people’s trust in the robot were investigated through an experimental study where the robot functioned as a guide for a human. The light signals used consisted of (1) a periodically dimmed signal mimicking respiration and (2) a signal indicating the robot’s current movement. Study participants found the robot to more often provide appropriate information with the added light signals. Interestingly, we also found the added light to increase discomfort (Robotic Social Attributes Scale) in the interaction with the robot. This previously unreported phenomenon highlights an important issue for robot designers to bear in mind when adding dynamic light.
Introduction
Through nonverbal social cues, people communicate and negotiate intentions, their comprehension of one another, as well as information about shared objectives. Eye gaze, body posture, and facial muscle activity, have been studied extensively within psychology as implicit indicators of affect. In social interactions, nonverbal cues are thus often attributed more weight than words. It is generally agreed that nonverbal communication (NC) is also central to attaining well-functioning human–robot interaction (HRI) and that the use of NC is a prerequisite for smooth and natural interaction. NC has, for instance, been shown to improve efficiency and robustness in collaborative tasks. 1 While robots may make use of similar NC as humans do, they also possess novel distinct NC modalities including light. To further investigate how light for NC can improve HRI, we explore here the effects of adding dynamic light signals to a hexapod robot in a task where the robot serves as a guide for a human.
Background
Light has been used extensively for signaling within aviation and marine technology as well as in appliances, computer technology, and mobile phones. Within robotics, lights are used both to convey state information and as an expressive feature. 2 Dynamic shifts in color and intensity have previously been used to increase the complexity level of the information that can be conveyed by means of one or more light sources (e.g., in light motions or light animations). Humanoid robots, drones, and autonomous vehicles all frequently use LED lights to indicate the state of the robot to the user, but light signals have also successfully been used for more complex aspects of social interaction including conveying intent and affect.3,4 General principles for how to design light signals for robots are still lacking and designs often draw inspiration from common place light signals in everyday life and the signals of fictional robots. 2 Hence, more research is needed in order to develop repertoires of reliable and efficient light signals that can be used for interaction whose precise effects and interpretations have been validated empirically.
Materials and Methods
Through an experimental study, we investigated how adding dynamic light signals to a walking robot affects people’s impressions of its sociality and trustworthiness.
Robot platform
We used the MORF (Modular Robot Framework) hexapod robot with the BioMORF configuration that includes an added pneumatically actuated soft robotic skin and a neural control system (see Fig. 1). The neural controller consists of two central pattern generators (CPGs) controlling the robot’s gait and the cyclical inflation of the compartments of the soft skin to emulate respiratory motion (for details see 5 ) Individually addressable RGB LED lights (WS2812B 5050 RGB) were added below each of the four air compartments of the silicone skin, with five LEDs under each anterior compartment, and seven LEDs under each posterior compartment. The lights were positioned across the robot’s body to allow full visibility from all directions and to increase visibility of the actuated soft sleeve. The brightness range of each series of LEDs were calibrated individually empirically to compensate for the slightly uneven thickness of the silicone skin, so that an equivalent light intensity was obtained for each compartment.
Top: The two robot configurations used in the experiment. Middle: An overview of the types of light signals used by the robot (see also link to video in Sect. 3.3). The light signal for turning right (not shown) is the inverse of the signals for turning left. Bottom: The physical premises where the experiment took place and the path taken by the robot.
Robot programming and behavior
The ROS (Robot Operating System) software was used with four ROS nodes running to facilitate control of the robot via a joystick. The added light signals consisted of two overlaid patterns: (1) a periodically dimmed signal mimicking respiration and (2) light sequences indicating the robot’s current movement (see Fig. 1 middle). The intended functions of the added light patterns were respectively:
To emulate breathing to make the robot appear more lifelike, through the light dimming in sync with the inflation of the robot’s soft skin. We hypothesized that a more lifelike appearance might improve people’s impressions of the social interaction with the robot, by making the robot appear warmer and more competent. To display the robot’s current activity to the human user to make it transparent and understandable. We hypothesized that this transparency in actions might lead to increased trust in the robot.
For the cyclical dimming in light pattern (1), we used a central patern generator (CPG) control scheme from prior work. 5 A base frequency (10 beats per minute (BPM)) is gradually increased up to a max value (36 BPM) over 9 seconds when the robot walks. The frequency gradually decreases again towards the base value, when the robot is at a standstill. A green color was selected for the light based on user preferences expressed in preliminary tests. Light pattern (2), was superimposed on top of the cyclically dimming light pattern. For the moving forward/backward signal we used a blue light motion on the LEDs moving in the direction of movement (e.g., from back to front to indicate going forward), inspired by the directional signals used on the side mirrors of some car models. We used the car convention of blinking the left or right side lights at 1 Hz, to signal a turn taking place, switching between a green and a blue color. The stopping signal consisted of all LEDs pulsing in red color for 5s at the current CPG frequency to emulate brake lights.
Human–Robot interaction experiment
A between-subjects study was carried out using the Wizard-of-Oz method with two conditions corresponding to a “lights OFF” and “lights ON” configuration of the robot. Video showing the robot’s behavior in each condition is available at: https://youtu.be/Rrwt2OHdoP0. The interaction experiment took place in the entry hall of the Mærsk Mc-Kinney Møller Institute at the University of Southern Denmark (SDU). The layout of the area can be seen in Figure 1.
Measures
Participants performed self-reporting post-interaction. We used the Robotics Social Attributes Scale (RoSAS) 6 to measure impressions of the robot’s sociality with a 7-point Likert scale used to rate the 18 subitems of the 3 main constructs (Competence, Warmth, Discomfort). The question posed was: “Using the scale provided, how closely are the words below associated with the robot you have just experienced?’, with “1 = not at all”, “4 = a moderate amount”, and “7 = very much so.” To assess trust in the robot, we used the 14-item subscale of the Trust Perception Scale-HRI (TPS-HRI). 7 The questions posed for each subitem were “What % of the time will this robot be …” or “What % of the time will this robot …,” rated on a scale from 0% to 100% in steps of 10%.
Procedure
Participants interacted with the robot individually according to the following experiment protocol. The participant was randomly assigned to one of the two experiment conditions upon arrival. They were guided to a classroom and received information about the experiment and signed informed consent. Subsequently, the experiment task to follow the robot and pick up a box and drop-it off again was explained to them. They were shown an image of the robot and given opportunity to ask questions. After this, they were brought back to the entry hall and asked to wait for the robot to enter and stop in front of them (see Fig. 1—bottom). They followed the robot to where the box was hidden and picked it up and carried it to the drop-off point, following the robot. The interaction lasted approx. 7 minutes. Following interaction, the participant was brought back to a classroom to fill out the questionnaire on an iPad and receive experiment debriefing. The wizard controlling the robot was positioned on the first floor balcony overlooking the area. The wizard was only partially visible to participants with their hands and the game controller used for controlling the robot concealed behind fencing.
Participants
Convenience sampling was used with 35 international students recruited comprising 15 different nationalities and 19 different study programs (see Table 1 for demographic data). Data from one participant were excluded from analysis due to the robot malfunctioning. Independent t-tests confirmed that there were no significant differences between the groups with respect to age (p = 0.12) and familiarity with robots (p = 0.17).
Results and Demographic Data from the Interactions Experiment with Means (M) and Standard Deviations (SD)
Indicates reverse-coded items.
Indicates that a t-test was used for the comparison.
Results
Values for the RoSAS main items were calculated as the mean values of their six subitems. Likewise, the trust score was calculated as the mean of all subitems of TPS-HRI with reverse coding used for relevant items (see Table 1). Internal reliability tests were conducted to confirm the internal consistency of the main data. The internal consistency of the RoSAS main items was tested using the mean inter-item correlation, an appropriate measure for scales with less than 10 items. Values within the ideal 0.15-0.5 interval were obtained for all main constructs (Competence: 0.28; Warmth: 0.47; Discomfort: 0.25). As TPS-HRI consists of 14 items, Cronbach’s α was instead used as reliability test. An α value of 0.667 was obtained, indicating acceptable internal consistency. Normality of the data was checked with Shapiro–Wilk tests and independent t-tests and Mann–Whitney U tests were used for comparisons as appropriate.
We found a statistically significant difference [t(29) = 2.09, p = 0.048] between the two configurations for RoSAS Discomfort, with a medium effect size (Cohen’s d = 0.72) indicating that participants found the “lights ON” condition more discomforting. A statistically significant difference (U = 81, p = 0.024) was also found for the “Provide appropriate information” item of TPS-HRI, with a large effect size (Cohen’s d = 0.80), indicating that participants found the robot more prone to provide appropriate information in the “lights ON” condition.
Discussion
We found no significant differences in the overall trust rating between the two configurations, indicating that the added lights did not affect people’s trust in the robot. However, a significant difference was found for the single item” Provide appropriate information” of TPS-HRI. We hypothesize that this difference is attributable to the move forward/backward, turn, and stop signals. It is likely that, based on prior knowledge of the light signal conventions used in cars, participants were able to reason and decipher what actions each signal indicated. Hence, they found them appropriate and informative, as they revealed to the user the current activity of the robot. No significant differences were found for Warmth and Competence on the RoSAS. However, surprisingly, the rating of Discomfort was significantly higher for the “lights ON” configuration. We hypothesize that the increase in Discomfort could result from the added periodically dimming light pattern that mimics breathing, which, based on the robot’s activity level, increases and decreases in frequency to emulate physical exhaustion and restitution. This pattern may make the robot appear more lifelike and animated, as prior work shows point-light animations of biological motion can give rise to strong and spontaneously attributions of animacy. 8 This in turn could result in the robot appearing more uncanny, which could increase ratings for several Discomfort subitems (e.g., awkward, scary, strange, awful, and dangerous).9,10 An alternative hypothesis, to explain the increase in Discomfort, could be that the colorful lights amplify the insect-like look of the robot and trigger entomophobic responses in some people. The increased discomfort could thus result from a perceptual interaction between the robot’s insect-like appearance and the colored lights and might not generalize to other morphologies.
Conclusion
In the present work, we explored potential benefits of adding dynamic LED light to a hexapod robot that functions as a guide to a human. Specifically, a light signaling consisting of periodically dimmed light, mimicking respiration, overlaid with blinking and moving lights, indicating the robot’s current action, was used. We implemented these two overlapping signals, aiming to attain a coherent and convincing light design that contributes to both improving practical functionality and the robot’s aesthetics. Through an empirical study, the effects of the added light on people’s impression of the robot’s sociality and their trust in the robot were evaluated. We found no effect on the robot's perceived Warmth and Competence nor the overall trust rating. However, study participants found the robot to more often provide appropriate information with the added light signals. Our study also revealed a previously unreported phenomenon: the addition of lights can, in some cases, lead to increased user discomfort. This finding represents a valuable contribution to the field of robot design, as it highlights an important issue to bear in mind when incorporating dynamic light. Determining which specific light signals and configurations that trigger discomfort warrants further investigation.
Footnotes
Acknowledgments
The authors thank Cao Danh Do for controlling the robot during the interaction experiment and Mads Bering Christiansen and Naris Asawalertsak for support on using and augmenting the BioMORF platform.
Ethics Declaration
The authors declare that they have no conflict of interests. The procedure used in the study adheres to the tenets of the Declaration of Helsinki. Informed consent to participate and to collect and process personal data were obtained from all participants included in the study. The authors affirm that all human research participants signed informed consent regarding publishing their data.
Authors’ Contributions
H.J.C.V.: Data curation (lead); formal analysis (lead); investigation (lead); software (lead); validation (lead); visualization (lead); writing—original draft (equal); writing—review and editing (supporting). P.M.: Conceptualization (equal); methodology (equal); project administration (equal); resources (equal); supervision (equal); funding acquisition (equal); writing—review and editing (supporting). J.J.: Conceptualization (equal); methodology (equal); project administration (equal); resources (equal); supervision (equal); funding acquisition (equal); writing—original draft (equal); writing—review and editing (lead).
Author Disclosure Statement
The authors declare no competing interests.
Funding Information
The research was internally funded by SDU Biorobotics, University of Southern Denmark and in part by the Marie Skłodowska-Curie Actions-Doctoral Networks (Grant Number 101119614, MAESTRI) [PM].