A Pilot Investigation of Human Preference for Robot Arm Visual Form

First impressions matter, and a robot's appearance has a significant impact on a user's perception of it, as well as their decision whether to use it. At the same time, there is a knowledge gap on non-anthropomorphic robot design form's effects on human perceptions and therefore interactions with non-humanoid robots. In this paper, we begin to evaluate the effects of two specific non-anthropomorphic form conditions: robot arm link concavity and roundness. After systematically varying robot arm stimulus models' morphological characteristics, we conducted an online survey-based within-subjects pilot study (N = 10) to gather participant ratings of attributes for each arm. We found that the perceived safety was significantly higher for the most rounded link compared to the intermediately rounded link. Participant free-response feedback can also support future hypothesis generation, such as a possibility that convex forms may seem more humanlike. This work contributes a starting point for future exploration in the realm of non-anthropomorphic robot design.


INTRODUCTION
The impression of a device happens immediately upon frst sight, and infuences if it is deemed usable [14].This visual impression is determined by the morphology of the technology, making shape a critical consideration in the design of any new device (including robots) [9].This important truth, combined with the limited amount of existing research on robot visual form (especially for non-anthropomorphic robots), inspired the presented work.
Specifcally in assistive robotics, a market of technology intended for interaction, this impression is an important factor in determining the equipment's acceptability, and therefore its purchase and subsequent use.Thus, exploring the efect of robotic form design on perception of the robot is important to ensure that this technology gains actual use; these assistive technologies must be both useful and socially accepted.For example, many technologies intended for older adults, a common user base for assistive robots, have the visual appearance of medical devices and quickly fall into disuse [6].Further, populations in need of assistive technology are oftentimes subject to othering within society [1,13], a situation which could also be exacerbated by the visual form of assistive technology such as robot arms for aid in self-feeding.Overall, a clearer understanding of how assistive robot arm form afects perception of the robot itself can inform engineers on how to design their robot arms most appropriately, with potential for extension to other types of robots.
Other studies have evaluated perceptions of overall robotic designs characteristics such as anthropomorphic shapes and overall size.However, there is a gap in exploration of non-anthropomorphic shapes beyond basic ideas such as the negative perception of spiky or sharp elements.Robot arms are a ftting vehicle for expanding the knowledge base on non-humanoid robot form; they are commonly used in a broad range of contexts and already vary in terms of shape, kinematics, color palette, and material.Do nuanced diferences in a beginning subset of these features matter signifcantly to human onlookers?This pilot study sought to explore the efects that two systematically varied robot arm design elements -concavity and roundness -have on perception by human interlocutors.We presented these robots arms with a purported assistive feeding use case, to help viewers imagine these arms in intimate and important interactions with people.

RELATED WORK
A key uniqueness of this work is the exploration of form in a non-anthropomorphic model.The robotics community knows that the humanoid versus non-humanoid appearance of a robot signifcantly afects how people will interact with it [7].There are also many prior studies investigating and characterizing the efect morphology has on humanoid robots with varying anthropomorphic robot design conditions [3,4,10,12,15].Perhaps due to the varied nature of robot structures, associated with required capabilities, domains, and manufacturing requirements, work exploring nonanthropomorphic features has been much more sparse, with studies of these robots' visual features only recently drawing attention [11].Of the most related studies, much of the previous work on nonanthropomorphic features have focused on the form of existing robots [9,16,17].
One broadly recognized visual attribute of importance, even within the non-anthropomorphic robot literature, is robot size.Past studies of form have largely explored this factor, fnding it to infuence human perception of the robot's character.Smaller robots are viewed as more emotionally capable than larger robots, and therefore are more likely to elicit empathy from people when mistreated [12].Of the existing work exploring robot arms specifcally, it has been found that size played a large role in perception of stimuli [16].Accordingly, the present pilot study aims to expand the current body of knowledge by investigating less understood visual form attributes.
Past work on anthropomorphic robots helped us to curate the currently studied visual form attributes.For example, the overall shape of a humanoid robot impacts user emotion and perceptions of robot's personality [8].This past efort uncovered these efects by systematically adjusting robot morphology, varying the shape of diferent parts of the robot in visual stimuli.We follow a similar systematic adjustment of two design elements (i.e., concavity and roundness) in order to evaluate each variable's efects on perception of non-anthropomorphic robot arm.

METHODS
This exploratory pilot study investigated the impacts of changes in robot arm form design -specifcally concavity and roundness levels -on human perception of the robot.Within the study, we presented the mock context that all presented robot arms were intended for assistive feeding applications, a case in which we subjects online survey to explore these efects.Responses were anticipate visual form to matter greatly.We employed a withincollected with the approval of our university ethics board under protocol #IRB-2019-0172.

Stimulus Design
The presented robot arm models consist of two varying morphological characteristics: concavity and roundness of links.All other features (color, fnish, number of links, background, size, and pose) were consistent for all arms.The colors and semigloss fnish for the models were selected to closely mimic the commercially available Kinova Jaco robot arm's color and material.All stimuli were modeled in Autodesk Fusion 360, rendered with a neutral toned background to minimize distraction, and posed at a three-quarter  view to efectively display each arm's attributes.The dimensions of our variations are presented as proportions of end of link width, .Concavity: The three diferent link concavities in our robot arm stimuli were convex, straight, and concave.A profle view of each variation is presented in Fig. 1.All variants had a link length that was fve times the link end width.The convex shape curved outward along the length, and the 3 center width was straight 2 × .The shape had a rectangular profle, so the width in the middle was the same as the width on the ends.The concave shape had an inward-arching curvature along 5 the length, with a center width that was 8 × .Roundness: The three levels of roundness in our robot arm stimuli were squared, rounded, and round.The end-of-link properties of each variation appears in Fig. 2. All variants maintained the same end-of-link cross-sectional width.The squared edges were characterized by a square-shaped cross-section with corners rounded at a radius of 1 × .The 4 rounded edges were characterized by a squareshaped cross-section with corners rounded 3 at a radius of 8 ×.The round edges were characterized by a circular-shaped cross-section in which the diameter was equivalent to the end width.
Resulting Stimuli: The combination of these two independent variables results in nine arm stimuli, as presented in Fig. 3.

Measures
We gathered participant perceptions of each robot arm stimulus using self-report questions.We used the Robot Social Attributes Scale (RoSAS) to measure social perceptions of warmth, competence, and discomfort on 9-point Likert scales [5].We also administered the Godspeed safety questionnaire on a 5-point semantic diferential scale [2].Additionally, we collected data on price sensitivity for each arm stimulus, as well as a baseline image of a Kinova Jaco arm.Specifcally, participants estimated how much the cost of each arm.Free-response data after each stimulus and at the end of the study captured which features stood out or strongly infuenced responses.Finally, we collected basic demographic information on participant age, gender, ethnicity, nationality, and hometown.

Participants
The pilot study was distributed online through an online surveying platform, Prolifc, and yielded 10 complete surveys.The fve men and fve women in the study were all from the United States.60% of participants self-identifed as White, 20% as Black, 10% as Asian, and 10% as Pacifc Islander and White.Participant age ranged from 20 to 63, with a median of 27.5.

Procedure
The pilot study was a self-contained Qualtrics survey lasting roughly 30 minutes.First, participants confrmed the ability to view images in their device's browser and provided informed consent.We gave framing information with the context that the presented robot arms are intended for use in assistive feeding.Participants were asked to describe the purpose of the robot arms presented in this study, to confrm their grasp of the context.Next, the baseline Kinova Jaco arm image was displayed, and the participant entered its estimated cost.
In the main portion of the survey, each stimulus image was presented in a randomized order.For each stimulus, the participant completed the RoSAS, Godspeed safety, and price questions.Participants could also provide free-response feedback to the following question after each arm: "What part(s) of the robot arm stood out to you most or most strongly infuenced your responses?" After the fnal stimulus, the participant completed an open-ended question about their overall experiences with and opinions of to the stimuli.As a check, the participants were asked to again summarize the intended use context of the arm (assistive feeding).Lastly, the participant answered the demographic questions.

Analysis
The scale-wise results were analyzed using descriptive statistics and repeated-measures analysis of variance (rANOVA) tests, with factors of concavity and roundness and a = 0.05 signifcance level.For the price sensitivity analysis, we began by normalizing the prices indicated for each stimuli, using a given participant's indicated price for the baseline Kinova arm, as presented in Eqn. 1.We then performed a rANOVA test, with factors of concavity and roundness, for the normalized prices.P-values were adjusted with a Bonferroni correction to account for multiple analyses with the RoSAS subscales, Godspeed subscale, and price sensitivity, yielding an adjusted signifcance level of = 0.01.

− =
(1) A simplifed thematic analysis was conducted on the free-response sections.

RESULTS
Concavity: There were no signifcant diferences found across concavity level for any of the measures ( (2, 18) = 0.194 − 1.959, = 0.170 − 0.825).The descriptive statistics for each concavity level appear in Table 1.Notably, the concave stimuli tended to yield the lowest warmth and competence ratings, as well as the highest discomfort.The convex group was the only group estimated to cost more than the Kinova arm.

Convex
Free-Response Themes: Some participants noted a diference in thickness between the upper and lower links as having an infuence on their responses ( = 3).Related to concavity, a few participants expressed the theme of the arms' "human-likeness" ( = 3).These participants discussed the extreme convex curved arms as "humanlike" or as one participant described as "similar to a bicep".Additionally, participants had varying opinions on the positive or negative connotation of limb thickness.= 5 participants identifed thicker links as positive or good, using terms such as "sturdy" or "cool, " vs. "awkward" or "strange" for thinner models.The rest of the group was divided, with = 1 of the opposite opinion, = 3 having unclear stances, and = 1 participant identifying preferring neutral or non-extreme models.Some participants mentioned material as a factor that stood out ( = 4), but also stated that this factor did not infuence their perception on the arms once they realized that the material did not change between models.

DISCUSSION
The goal of this pilot study was to evaluate the general populations' perceptions of various visual form characteristics, concavity and roundness, on a robotic assistive feeding arm using a set of robot arm stimuli model images.We measured participant impressions of each arm through Likert-style questions, cost estimates, and free-response.
Looking at the link concavity results, the concave group consistently had the lowest average score for both warmth and competence factors, and the highest average score for the discomfort factor.The link roundness results showed signifcant diferences in perceived safety between the round and rounded models, with the round model rated higher.There was no signifcant diference found between the more extreme round and squared models, and the rounded model tended to gain higher competence ratings than either the round or squared models.The trend in low ratings for the concave group, in combination with the half of participant's textual responses indicating a distaste for "thin" models, indicates that there may be more to explore in the perceptions and preferences of these extreme "thick" or "thin" model conditions.
This study contributes to the knowledge of non-anthropomorphic robot design by means of being a frst step in developing a more thorough and systematic investigation into variation of visual conditions and design characteristics.At the same time, our sample was quite small, and the free response comments may indicate a need adjust our stimuli models to appear more realistically mobile and possessing equally thick upper and lower limbs to garner more real-world-relevant responses to the stimuli.From here future work could include more systematic variations and combinations of conditions, such as adjusting the number of links in the arm or the material fnish of the arm.A full study with a fully powered participant sample is underway for further exploration into these fndings.The fndings of this work can help inform designers and roboticists who wish to facilitate better frst impressions of robots.

Figure 1 :
Figure 1: Lengthwise link profles for the concavity variations.From left to right: convex, straight, and concave.

Figure 2 :
Figure 2: End of link cross-sections of the three conditions with corner curvature radius, , proportional to end width, .From left to right: squared, rounded, and round.

Figure 3 :
Figure 3: The nine stimuli used in the study.Concavity varies top to bottom: convex, straight, and concave.Roundness varies left to right: squared, rounded, and round.

Table 2 :
Means and standard deviations of measures for each roundness level, denoted as ().