The use of robotic arms in the rehabilitation context is increasing, leveraging the robotic capabilities related to object manipulation. Similarly, social robots are used in rehabilitation to provide features of social interaction with people. Even though these two robotic aspects are gaining visibility, the combination of both manipulation and social capabilities is often neglected. This work aims at filling this gap, introducing both manipulation and social capabilities in a scenario of human-robot interaction for rehabilitation. This work will define the best social configuration for a robotic arm, in terms of social cues. In particular, sound and expressive eyes cues will be linked to social movements designed using Laban Movement Analysis. Various combinations of movements and social cues will be combined and embedded in a robotic arm. Then 15 participants are recruited and asked to perform a human-robot interaction task (handover) with the robotic arm. A questionnaire is used to evaluate the user impressions in terms of Valence, Arousal, the Godspeed components of Animacy and Safety, and two custom questions related to movement fluidity and likeability. The results show that the best combination for this exercise is the one where both sound and eyes are present, even though the data show high scores of Arousal and Animacy for the configuration composed of only sound and movement.
The robotic world is advancing towards technological improvements and the application of
robots is spreading to various fields. In fact, it is possible to assist in the advancement of
Social Robots [
This study is based on the definition of an HRI task for the evaluation of the proposed novelty. The interaction task design is simple, yet relevant in the realm of HRI and is a starting point for future implementation of this system. A handover task will be presented, which is composed of multiple steps that can be used singularly or grouped in the execution of a rehabilitation exercise (e.g. tower of London [17]). In this section, the procedure to put the experiment in place is described and further details are provided regarding the setup.
In the first place, this work focused on how to design the social cues that should be used during HRI, to achieve a particular task objective. As described in Section 1 this work uses 3 social cues. I) Robotic Movement: stemming from a previous work of the same authors [15], the movement design is based on Laban Movement Analysis (LMA). LMA was defined by Rudolph Laban in 1948 [18] as a system for describing, discussing, and documenting human body movements according to diferent components, namely Laban Efort Graph (LEG) referring to the body’s inner use of energy, and Laban Space (LS), referring on how the body uses the surrounding space. The LEG component is composed of four parameters that can have one between 2 values and the LS component can vary according to the 3 parameters (corresponding to 3 planes of movement) that change between two opposite values. A schema of such a configuration and all the possible values are summarized in Table 1.
The diferent combinations of LEGs parameters and the LS parameters originate diferent movements, that are designed to convey emotional information. The set of movements to be used in this work comes from the previous work [15] where various Laban Movements were shown to a pool of 120 participants and their preferences were collected. The most preferred movements were the one modeled after the happiness emotion and the one modeled after the anger emotion; in this work, the happy movement will be called Indirect and the angry one will be called Direct. Their Laban configuration is presented in Table 2.
I) Eyes: the face was selected from another previous work [
The overall system is run on a Ubuntu OS and takes advantage of the backbone infrastructure provided by ROS [19]. Each of the social cues is written in a node and registered to a common topic, which is published by a /controller that takes charge of the synchronization of the system. It is worth highlighting that even though the overall system is ROS based, the robotic arm movements were achieved thanks to Frankx [20] which is a movement library developed for the Franka Emika Panda robot and that allows control over velocity, acceleration, and jerk. The ifnal part of the ROS system is a node for saving data after the interactions, which connects to the disk and saves all the interaction session data in a CSV format, for later analysis. The data saved are user feedback regarding the interaction session and general information on the robot, such as time duration of the movement, and information on which action it is performing. At last, a node that records bag data from a RealSense camera (Intel, USA) is synchronized with the system for autonomous recording and saving of the diferent sessions. The overall schema for all the ROS nodes is depicted in Fig. 1. Another element of the system, unrelated to ROS, is a Docker container that hosts a Flask app; such an app allows connection from external elements to the app that is used to administer the questionnaires to the participants in the experiment and allows to save the data on the disk passing by a ROS node, contacted using RosBridge [21]. All system is triggered by a ROS service which is manually called by the experimenter. This choice was done to allow full control over the session duration and manage any unexpected event that could have happened during the experimentation.
A total of 15 participants were enrolled for this test. The people enrolled were researchers, medical doctors, psychologists, and nurses that volunteered in the experiment. No eligibility criteria were selected and all volunteers were accepted into the study. The participants are composed of 60% male and 40% female, and their education level is composed of 3 groups (33.3% Bachelor’s Degree, 53.3% Master’s Degree, and 13.3% Ph.D.). The mean age of participants is 30.47 years, with a standard deviation of 4.64. Before starting, all the participants were informed of the test and it was given to them a formal description of the experiment on paper; all of them were asked to sign informed consent.
The task to be performed during this experiment is the handover of a colored ball between the robot and the participant, with the participant that places the ball in a dedicated box to end the interaction. The social cues will be proposed in 4 possible combinations (only movement, movement and sound, movement and face, movement, face, and sound), coupled with the 2 movements (Direct or Indirect). Hence, there will be a total of 8 trials for each participant (4 social cues combinations by 2 movements). The experiment can be divided into 3 main phases, namely the robotic arm movement, the human-robot handover, and the rating of the interaction by the user. The robotic arm movement is in charge of picking up the colored ball and passing it to the user. The main design idea of the movement is introduced in the previous paragraph, and can be further described as a combination of 5 main blocks: i) The robotic arm turns toward the colored balls, ii) The robotic arm performs a movement to get closer to the balls and picks one of them, iii) The robotic arm turns toward the user, iv) The robotic arm performs a movement and gets closer to the user to handover the ball, v) The robot goes back to the idle position. The choice of which ball to pick, the order of movements, and the social cues combination were random and automatically decided by the system. After the robotic arm movement, the human-robot handover phase started, with the participant that grasps the ball ofered by the robot, and puts it in a dedicated box with the same color as the ball displayed on top. Eventually, the participant is asked to complete a questionnaire to rate the experience of that specific session. The questionnaire used in this work was structured with a first part where it asked the user to rate the interaction according to the Self Assessment Manikin (SAM) [22], in terms of Valence, Arousal and Dominance. Then the user is asked to complete two elements of the Godspeed questionnaire [23], specifically the Animacy and the Safety component. At last two custom questions were asked, regarding the fluidity and the likeability of the movement. At the end of the 8 interactions, the experimenter asked 5 open questions to the participants, to acquire their qualitative feedback. Such questions were: 1) What do you think about the movements? 2) What do you think about the sound? 3) What do you think about the face? 4)
Did the movements convey to you any emotion or sensation? 5) Imagine bringing the robot home to use for your own purpose, which configuration (sound, face, movement) would you like to have? In this work only answers to question 5 will be analyzed. A schema of the phases for this experiment is available in Fig. 2.
The analysis focused on the data collected from the questionnaires. For the SAM scores, the values were extracted on a scale of 1 to 9; the Godspeed questions were summed up together to obtain an aggregated index with all the scores for the Godspeed elements where the maximum value is 25 for animacy and 15 for safety and the minimum values are respectively 5 and 3. Finally, for the likeability and fluidity scores, the value is taken and evaluated on a scale from 1 to 5. The means for each answer’s group were evaluated for each interaction. The sessions reported in Table 3 are divided into columns by social cues (no social cues, eyes, sound, both sound and eyes) and by Direct (D) or Indirect (I) movement. The means extracted from the questionnaire were then compared across all the diferent iterations and the comparison between the diferent movements and social cues combination are analyzed and discussed. Due to the low number of subjects (15), only qualitative analyses were conducted and no statistical significance was looked for. Finally, the open answers to question 5 were analyzed using content analysis [24] and 2 main components were extracted, namely which configuration among the 4 presented was the one preferred and which movement was the most liked by participants.
NO SOCIAL
CUES D In the first place, the mean values for the SAM, Godspeed, and Custom values were computed. The complete results are depicted in Table 3. The Valence has the highest values for the session with all the social cues and the Indirect movement, asserting that this interaction is the most positive. On the contrary, Arousal has the maximum value in correspondence to the session with only sound and Indirect movement. So the movement that maximizes the Arousal value is the Indirect one; indeed the social cues play a role in the perception, and the session which has the highest Arousal score is the one with sound only.
As for the Godspeed, it is possible to identify the highest Animacy scoring in the sessions with only sound and with sound and eyes, for Indirect and Direct movement respectively. The sound seems an element that for the Indirect movement is capable of increasing the perception of Animacy (17.67 against 16.53 for Direct movement). Regarding Safety, it is possible to notice how the highest value is in the last session for the Direct movement. At a first glance, it is possible to say that the Godspeed parameters used in this study are maximized for Direct movement. Finally, the highest values for both likeability and fluidity are in the session with all social cues and Indirect movement. When looking at the open answers the results partially reflect the values highlighted in the tables. The preferred movement (Figure 3a) is the Indirect one (53.3%), but the 33.3% of people did not express a preference; only 13.3% rated the Direct as the preferred movement. For the configuration (Figure 3b), the preferred configuration of social cues is the one where both sound and eyes are present (60%), while 26.7% of participants liked only eyes and the 13.3% liked only the sound.
The results reported in Table 3 show that the configuration of the social cues with the highest
scores is the one where eyes, sound, and movement are used at the same time. Including only
the eyes or the sound in the session does not seem to afect significantly the perception, even
though it is worth mentioning that the use of only sound has been rated at the highest for the
(a) Preferred movement for the participants
(b) Preferred configuration for the participants
SAM component of Arousal and for the Godspeed part of Animacy. This evidence suggests that
sound was more appreciated than the eyes in communicating social information. The reasons
for this are mainly 2: the focus of the user was on the gripper part of the robotic arm, so the
eyes, placed in the bottom part of the robot, had a less relevant impact on the perception. Then,
being the gaze of the user focused on the robotic gripper, it was easier to feel a social cue that is
perceived through another one of the 5 human senses, the hearing. When evaluating the open
answer left by the participants, the results are not fully aligned with the quantitative data. In
fact, even though the quantitative data for the Godspeed analysis showed an overall preference
in terms of Safety and Animacy of Direct movement, the open answers show a preference for
Indirect movement. Similarly, when asking about the preferred configuration, even though the
quantitative data for the session involving only sound is higher, the preferences are lower for
only sound than for only eyes. This point suggests that there is not a perfect alignment between
the quantitative and qualitative data and this information should be furtherly investigated in
future studies. The answers regarding the most preferred session are similar, being the session
of both eyes and sound the best (60%) for this sample of participants. The results collected
confirm the work from [
The aim of this work was to evaluate the most preferred combination of robotic arm movement and social cues. Secondarily, there was the purpose to understand what kind of movement is preferred by people. The preferred configuration has been verified to be the one with sound and eyes, and the preferred movement, regardless of social cues, is the Indirect one. Further evolution of this study will aim at enlarging the sample size to perform also statistical analysis of the results. Moreover, it will be possible to verify the correlation between perception and users’ personalities. Nonetheless, further experiments should validate the findings of this work and such findings will be used to perform a real rehabilitation exercise with healthy people, to evaluate their perception of the robot and the validity of the social interaction. movement generation for robots and other near-living creatures, in: Springer Tracts in Advanced Robotics, volume 111, 2016, pp. 25–48. doi:10.1007/978-3-319-25739-6_2. [12] H. Knight, R. Simmons, Expressive motion with x, y and theta: Laban Efort Features for mobile robots, in: IEEE RO-MAN 2014 - 23rd IEEE International Symposium on Robot and Human Interactive Communication: Human-Robot Co-Existence: Adaptive Interfaces and Systems for Daily Life, Therapy, Assistance and Socially Engaging Interactions, Institute of Electrical and Electronics Engineers Inc., 2014, pp. 267–273. doi:10.1109/ROMAN.2014. 6926264. [13] H. Knight, R. Simmons, Laban head-motions convey robot state: A call for robot body language, in: Proceedings - IEEE International Conference on Robotics and Automation, volume 2016-June, Institute of Electrical and Electronics Engineers Inc., 2016, pp. 2881–2888. doi:10.1109/ICRA.2016.7487451. [14] C. Clark, L. Sliker, J. Sandstrum, B. Burne, V. Haggett, C. Bodine, Development and preliminary investigation of a semiautonomous socially assistive robot (sar) designed to elicit communication, motor skills, emotion, and visual regard (engagement) from young children with complex cerebral palsy: A pilot comparative trial, Advances in Human-Computer Interaction 2019 (2019). [15] C. La Viola, L. Fiorini, G. Mancioppi, J. Kim, F. Cavallo, Humans and Robotic Arm : Laban
Movement Theory to create Emotional Connection *, in: Ro-MAN, 2022. [16] D. Eizicovits, Y. Edan, I. Tabak, S. Levy-Tzedek, Robotic gaming prototype for upper limb exercise: Efects of age and embodiment on user preferences and movement, Restorative Neurology and Neuroscience 36 (2018) 261–274. doi:10.3233/RNN-170802. [17] P. Anderson, V. Anderson, G. Lajoie, The tower of london test: Validation and standardization for pediatric populatons, The Clinical Neuropsychologist 10 (1996) 54–65. [18] R. von Laban, Modern educational dance., 1948. URL: http://books.google.com/books?id= nsAKAQAAIAAJ. [19] M. Quigley, K. Conley, B. Gerkey, J. Faust, T. Foote, J. Leibs, R. Wheeler, A. Y. Ng, et al., Ros: an open-source robot operating system, in: ICRA workshop on open source software, volume 3, Kobe, Japan, 2009, p. 5. [20] L. Berscheid, T. Kröger, Jerk-limited Real-time Trajectory Generation with Arbitrary
Target States (2021). URL: https://github.com/pantor/ruckig. arXiv:2105.04830v2. [21] R. Toris, J. Kammerl, D. V. Lu, J. Lee, O. C. Jenkins, S. Osentoski, M. Wills, S. Chernova, Robot web tools: Eficient messaging for cloud robotics, in: 2015 IEEE/RSJ international conference on intelligent robots and systems (IROS), IEEE, 2015, pp. 4530–4537. [22] M. M. Bradley, P. J. Lang, Measuring emotion: The self-assessment manikin and the semantic diferential, Journal of Behavior Therapy and Experimental Psychiatry 25 (1994) 49–59. URL: https://www.sciencedirect.com/science/article/pii/0005791694900639. doi:https://doi.org/10.1016/0005-7916(94)90063-9. [23] C. Bartneck, D. Kulić, E. Croft, S. Zoghbi, Measurement instruments for the anthropomorphism, animacy, likeability, perceived intelligence, and perceived safety of robots, International journal of social robotics 1 (2009) 71–81. [24] K. Krippendorf, Content analysis: An introduction to its methodology, Sage publications, 2018.