"Noisy" Matching of Motion Velocity of an Assistive Robot to the Users' Walking Velocity

This study investigates the impact of dynamic matching of robot motion velocity to users' walking velocity in a human-robot approach scenario on three categories: perceived comfort, interactivity, and naturalness. Considering age diversity, participants were divided into two age groups. Young participants tended to rate higher for all three categories when the robot's approaching velocity was dynamically matched. In contrast, elderly participants preferred a steady and slow robot approach for comfort and predictability. These findings contribute to the ongoing effort to design assistive robots that effectively cater to diverse user groups, ultimately enhancing user satisfaction and acceptance. Taken together, this study highlights the importance of dynamically tailoring robot behaviors based on user demographics for positive Human-Robot Interactions.


INTRODUCTION
Human-Robot Interaction (HRI) is a multidimensional feld that collectively contributes to overall user experiences, including the robot's capacity to adapt to its users.This is particularly important regarding assistive robots [3].A crucial factor in this context is the motion velocity of assistive robots.Previous research has pointed out the importance of the robot's ability to adapt its "walking" (whether bipedal or wheeled) velocity to human users for perceived comfort and safety [11,13].This is especially relevant in the context of rehabilitation, particularly in aiding gait training [5].Additionally, natural and intuitive robot behaviors could encourage the users to better understand and interact with the robots.For example, users tend to feel more comfortable with robots exhibiting human-like movements, contributing to increased engagement [1,6].Thus, the capability of assistive robots to align their motion with that of the user emerges as a critical factor, especially in applications requiring close and dynamic interactions.
However, limited attention has been given to the nuances of dynamically matching the robot's motion speed to users, especially when its velocity varies around the user's walking velocity.This variability may impact users' perceptions of the robots and HRI experiences.
This pilot study serves as a precursor to a more extensive experiment and project, ofering valuable insights into the feasibility and potential directions for future research.Our objective was to examine "how matching, varied ("noisy") matching, or mismatching the robot's motion velocity to users' mean walking velocity during a human-robot approach scenario infuences users' perceptions of  comfort, interactivity, and naturalness" in controlled environments depicted in Figure 1.Specifcally, we sought to understand the potential efects of this dynamic velocity-matching paradigm on individuals belonging to diferent age groups, namely young participants (YP) and elderly participants (EP).

METHODOLOGY 2.1 Participants
Ten YP at the University of Oslo and ten EP at a care facility in Oslo (20 participants in total) were recruited for this study.The criteria for the recruitment study were that the participants were healthy, without any obvious cognitive and physical difculties that would make it difcult to carry out the task in the experiment.The basic demographics of the participants are shown in Table 1.

Robotic system
A TIAGo robot by PAL Robotics [8] was used for the study (Figure 2).The robot was controlled through the Robot Operating System (ROS) [9] on a laptop through a network cable between the laptop and the robot.

Experimental setup
The experiment was carried out in two locations: a robotic lab at the University of Oslo for YP and a communal area in an elderly home in southeast Norway for EP (Figure 1).When a participant and the robot "met" by the pre-defned destination marked by the colored tape on the foor, the distance between the robot and the participant was kept within reach of the robot's arm, which falls within the so-called far phase (76 to 122 cm) of the personal zone (46 to 122 cm) [10].

Experimental conditions
There were three diferent experimental conditions, which were semi-randomized for each participant to avoid bias, such as order or practice efects.The three conditions included the following: (1) Default: Fixed approaching velocity of the robot (2) Matched: Matched approaching velocity of the robot to the user's mean walking velocity (3) Varied: Varied approaching velocity of the robot, randomly selected from ± 25 %, around the user's mean walking velocity The motion velocity of the robot in the Default condition was fxed at 0.2 m/s, which is the default velocity defned by the producer of the robot (PAL Robotics).In the Varied condition, the robot's velocity did not change once selected during a sequence and stayed constant in that particular sequence.

Experimental procedure
Participants received a detailed explanation of the experiment's procedure before signing an informed consent form.They were then instructed to walk toward the robot at a normal pace, knowing the robot would approach as they reached a designated spot marked by colored tape on the foor.Crucially, information about the varying robot velocity was intentionally withheld to maintain participant "blindness" to speed changes, preventing behavior adjustments based on expectations.After each robot approach, participants provided ratings and reasons for their experience through short questions.This sequence was repeated three times.Subsequently, participants answered longer questions to elaborate on their experiences.
After completion, participants underwent debriefng to clarify the purpose of the diferent experimental conditions, address concerns, and uphold ethical standards.The entire experiment lasted approximately 20 minutes per participant.

Velocity estimation
Participants' walking velocity was estimated using an RGB camera to capture video images from the side as the participant walked past the camera (Figure 3).Prior to entering into the Field of View (FoV), the participants had already started walking for about 1.5 m and continued for about 3.5 m after entering the FoV.We employed a computer vision technique utilizing the OpenPose library1 to detect the movement, specifcally focusing on the joint number 17, representing the right ear.
The general formula for velocity () is expressed as the derivative of the pixel position of the detected joint () with respect to time (): (1) Δ However, due to frame rate fuctuations exceeding 2 Frames Per Second (FPS) while using the OpenPose library, making the element of time () inconsistent, we opted to consider ∼ Δ and assumed a fxed frame rate of 12 FPS.Consequently, an array of the changes in pixel positions (Δ 1 , Δ 2 , ..., Δ −1 ) was collected.
Finally, the mean Δ was calculated by: (2) − 1 where n represents the number of instances in which OpenPose determines the pixel position of joint number 17 (i.e., the right ear).
The calculated values were converted to meters per second (m/s) and scaled between 0.1 and 1 m/s for the robot.The scaling ensured a proportionate adjustment for the robot's velocity to align with that of the participant.

Measurement
This study adopted a mixed-methods approach, combining quantitative and qualitative data collection.
In the quantitative phase, participants' velocity aligned with the robot's motion was measured, along with numerical ratings using a seven-point Likert scale derived from the Robotic Social Attribute Scale (RoSAS) [4].Ratings were consolidated into three categories: comfort, interactivity, and naturalness.Using a seven-point Likert scale, each item ranged between very uncomfortable (1) and very comfortable (7), non-interactive (1) and very interactive (7), and very unnatural (1) and very natural (7), respectively.
The qualitative aspect involved a thematic analysis of extended questionnaire responses, providing a comprehensive understanding of participants' perspectives.For EP group, the questionnaire was communicated verbally in an interview-style format.

Data analysis
The data, managed and analyzed in Python, underwent a two-way ANOVA to assess age group and experimental condition efects on participants' ratings for comfort, interactiveness, and naturalness.Post-hoc analysis involved a Student's paired two-tailed t-test and Cohen's d for efect sizes.

Ethical considerations
Participants received information about the experiment beforehand, signed a consent form, and were aware of their right to withdraw at any point without the need for explanation or fear of negative consequences.Upon completion of the experiment, participants received a small treat as a token of appreciation.

Velocities of the robot's motion and participants' walking
We observed signifcant diferences (p < 0.001) between the mean walking velocities of YP and EP (Figure 4).On average, YP walked at 0.88 m/s (SD = 0.18 m/s), approximately 39 % faster than EP, who walked at 0.59 m/s (SD = 0.19 m/s).

Participant ratings and experiences
After each sequence, participants provided ratings for comfort, interactiveness with the robot, and overall naturalness (Figure 5).Signifcant age group efects were observed on comfort (F(1,54) = 8.48, p < 0.005), interact (F(1,54) = 10.85,p < 0.05), and natural (F(1,54) = 23.93,p < 0.001).The three experimental conditions did not yield a signifcant result.The interaction between age group and experimental conditions was not signifcant, implying a consistent impact of experimental conditions across the two age groups.
Reviewing the results for EP, our data indicates homogeneous ratings across all conditions, suggesting a lack of notable diferences.Interestingly, the EP assigned higher ratings than YP in every experimental condition.

Qualitative insights and user perspectives
Participants' feedback on robot motion velocity revealed key themes such as perceptions of naturalness, comfort, speed concerns, and the impact of velocity variations on overall HRI.Naturalness perception.Participants (7 out of 10) noted that the robot's motion in the Default condition was perceived as very slow.Several responses stated that "The robot moved too slow and did not feel natural."The robot was indeed slower in the Default condition compared to the other conditions (Figure 4).Although sometimes participants found the Varied condition "startling, " many young adult participants (5 out of 10) found the Varied condition the most natural.For instance, YP1 responded, "It just felt the most natural, he moved at the right pace."In addition, YP3 responded "Nice pace."

Young adult participants (YP).
Comfort, interaction, and speed.Comfort in interacting with the robot seems to be closely tied to its motion velocity.In the Default condition, YP6 noted, "Robot appeared to move slower, which made the interaction weird," and YP10 stated, "If [the robot is] too slow, the interaction becomes awkward." The Default condition was often criticized for its slowness, while the Varied and sometimes Matched conditions were perceived as more aligned with participants' speed expectations.Notably, YP8 mentioned, "The interaction is more pleasant if it happens as fast as you would expect it to."However, a subset of participants (2 out of 10) felt the diferences were subtle or nonexistent, with YP8 expressing, "I barely felt the diferences" referring to the Matched and Varied conditions.

Elderly participants (EP).
Comfort in predictability.EP consistently expressed a preference for the Default condition, emphasizing their comfort with the robot's constant and slow motion.Although this preference might not be immediately evident in their rating responses (Figure 5), qualitative comments, such as EP4's observation that the robot was "a bit brisk, perhaps" in the Matched and Varied conditions, and EP6's remark that "the robot must not be too fast, " underscored the importance of a stable pace for their comfort.
Perception of the robot's motion velocity.The diferences in the robot's motion velocity were less noticeable for EP.All of them reported and found all conditions generally pleasant.This explains their general tendency to give higher ratings than YP.It was evident in their responses in the interview that the attention was drawn to the robot's movement as a whole rather than speed nuances.EP2 responded, "I did not notice anything regarding the speed.All speeds were pleasant." Similarly, EP6 responded, "I did not notice any diferences regarding speed.I was fascinated by the robot that I was not thinking of that." There was a general tendency to favor slow speed (the Default condition) for a pleasant interaction.A directly related comment was made by EP9 "I preferred the lowest speed, not the fastest."

DISCUSSION
Our main focus was on evaluating how noise in matching robot motion velocity to users' walking velocity in human-robot approach scenario infuences the perceived comfort, interactivity, and naturalness of the interaction, particularly considering the age diversity among the participants.
The results indicate that noisy matching could enhance the perceived naturalness of the interaction for YP.This fnding resonates with ecological validity, users' preference when robot behavior is aligned with human-like motions and human expectations [6,12].
Contrary to the undistinguished ratings, likely infuenced by their limited prior experience with robots, EP's preference for steady and slow robot motion underscores the signifcance of predictability and safety.This preference indicates a conservative stance, potentially improving acceptance, which is crucial considering the historical resistance of the older population towards assistive robots [2].
The observed preferences of two age groups underscore the need for adaptable robot behaviors considering individual diferences and user expectations.Recognizing the potential novelty efect, especially for those unfamiliar with real-life robot interactions, is also important.Measures to mitigate such efect, depending on study goals, may include providing opportunities for participants to familiarize themselves with the robot before the experiment through training or long-term interactions where novelty diminishes [7].
In summary, our fndings prompt considerations for the design and customization of assistive robots based on user demographics.Tailoring robot behaviors to align with the preferences and comfort levels of diferent user groups is essential for promoting positive interactions.

CONCLUSION
In summary, our study sheds light on the nuanced user preferences in response to noisy matching of robot motion velocity, emphasizing the delicate balance between adaptability, naturalness, and predictability in designing assistive robots.Future research should delve deeper into the complex interplay between robot behaviors and user expectations, exploring additional demographic factors like cultural background and prior robot experience.

Figure 1 :
Figure 1: The experimental setups for young (left) and elderly (right) participants.The orange and blue areas indicate camera positions for video footage used in observational analysis, which is not reported in this paper.P: Participant.PI: Principle Investigator.R1-3: Researchers.

Figure 2 :
Figure 2: The TIAGo robot used in the experiment.

Figure 3 :
Figure 3: Illustrations of the OpenPose library fgure overlay on walking persons during the experiment.

Figure 4 :
Figure 4: Comparison of the velocity between the robot's motion and participants' walking.

Figure 5 :
Figure 5: Mean ratings of the participants in short questions for three categories (Comfort, Interactiveness, and Naturalness) across the three experimental conditions (Default, Matched, and Varied robot motion velocities).Error bars represent the standard deviation.

Table 1 :
Basic demographics of the participants.