Beyond the Default: The Effects of Adaptable Robot Speed in Industrial Human-Robot Interaction

Collaborative robots have gained popularity in industrial manufacturing due to their flexibility. However, one substantial challenge is to utilize this flexibility human-centered. While there is considerable research emphasis on system-controlled approaches, the exploration of user-controlled adaptation has been comparatively overlooked. Therefore, this study investigated user-controlled speed adaptation in an experimental within-subject design. We assessed mental workload of 36 participants using subjective, physiological, and secondary task performance measures. Moreover, we investigated acceptance, perceived control, and the influence of desirability of control. The primary task simulated assembling circuit boards in a sequential scenario. The results indicated that workload did not decrease with speed adaptation. Conversely, secondary task reaction time was even slower. However, participants generally preferred the adaptation and reported a greater sense of control. In general, the findings suggest that adapting speed could have positive subjective and possible negative performance-related aspects.


INTRODUCTION
In the past decades, a new category of robots has entered the industrial domain known as collaborative robots, or cobots.These robots are specifcally designed to interact with humans in a shared environment physically, eliminating the need for conventional barriers or protective cages [26].The often forecasted synergy between humans and cobots is poised to be a crucial factor in the industry to achieve higher performance and fexibility in production lines [23].Yet, when viewed from a psychological perspective, the collaboration between robots and humans presents countless challenges [26,31].To achieve a smooth collaboration, robots should have the ability to adapt or be adaptable to certain situations [16].The feld of human-robot interaction (HRI) is currently dominated by research on adaptive (i.e., system-driven adaptation) approaches [2,28].Despite the assumption that adaptivity and related increased autonomy of the cobot is expected to enhance performance, it may introduce ambiguous efects regarding key human factors concepts such as workload, acceptance, and situational awareness [28,31].A relevant concern is that adaptability is inherently linked to a reduced sense of control on the part of the human and more perceived responsibility of the robotic interaction partner [15,28].A promising strategy to ensure both fexibility and perceived control on the part of humans could therefore be an adaptable approach (i.e., user-driven adaptation).
The relevant cobot features for adaptation include joint stifness, trajectories, proximity to the robot, or the speed of movements [4,12,17,27,32].The speed of industrial robots was one of the frst variables for which a considerable infuence on human experience and behavior was illustrated [4,12,27].Interestingly, there mainly has been a search for an optimal speed rather than the consideration of adaptable speed.Importantly, previous research has illustrated that both low [27] and high [4,16] robot speed can lead to increased mental workload.Due to these previous fndings, a medium default speed is often recommended to achieve a lower mental workload.Moreover, a recent study [29] suggests that adaptable speed might reduce mental workload, yet due to varied scenario manipulations (such as speed, adaptability, autonomy, trajectories, and command types), it remains uncertain if the workload reduction was primarily linked to speed adaptability.To address this still open research speed condition compared to the default speed condition.In line with earlier research, we used a mixed methods strategy to capture mental workload [34].Besides incorporating a frequently used questionnaire [1,19,20] as a subjective indicator, heart rate was assessed as a psychophysiological indicator [6,18] and secondary task performance as a behavioral indicator [25].
In addition to mental workload, acceptance stands out as an important factor, given the assumption that this attitude shapes both intention and the actual use of the robot [5].The study of [29] also examined the infuence of the adaptability of various workstation factors on acceptance.The researchers concluded that the scenario involving speed adaptation was more accepted by participants than the previous scenario without adaptation.However, as mentioned above additional variables, such as extra information were changed between the two scenarios making a clear attribution of the efect to the adaptability of speed impossible.The present study seeks to address this shortfall.In this context, we propose the hypothesis that participants show higher acceptance of the robot in the adapted speed condition compared to the default speed condition.
While the potential positive efects of adaptability on acceptance and workload are yet to be explored, it is evident that the adaptability of speed enhances individuals' control of the robot.Control is a crucial psychological factor for well-being which matters for HRI [3,35].In earlier HRI studies control was realized via the adaptation of the robot's personality [24], appearance [22], and autonomy [35] and related to multiple positive outcomes.Even though the adaptability of motion characteristics in industrial HRI has not been investigated so far, based on these results, we assumed that participants perceive higher control in the adapted speed condition compared to the default speed condition.Moreover, even if the actual control increases to the same extent for all individuals, the perception can vary between them due to interindividual diferences.Currently, particularly personality traits like extraversion are emphasized as relevant interindividual diferences in HRI [14].However, frst studies exploring the correlation between the desire for control and preferred robot autonomy suggest a potential impact of this trait, too [9,10].Hence we hypothesized that participants, who have higher desirability of control, prefer using the adapted speed.
Taken together, the current study aimed to investigate the impact of the adaptability of a robot's speed in an industrial setting.Therefore, we designed a sequential collaborative scenario [13,26], in which the participants worked with a cobot to assemble LEGO circuit boards.Either a default speed was employed, or an individual speed was determined by using a methods of limits approach [11] in a prior interaction phase.

METHODS
A checklist of the local ethics committee for self-assessment of studies' ethical harmlessness was conducted.The signed checklist, preregistration, and collected data are accessible via the Open Science Framework (https://osf.io/y6vax/).

Participants
Based on an a priori power analysis our target sample size was 34 participants.A total of 43 participants were recruited, both from the local university and through personal connections.Data sets from fve participants were evaluated as pretests due to changes in the study design.Two participants were excluded as a result of an erroneous experimental procedure.The resulting 36 participants were predominantly students or workers in the technical feld (77.8%) with ages ranging from 18 to 33 years ( = 24.11years, = 3.56 years).Half of the participants identifed themselves as female, the other half as male, and no participant as non-binary.The majority had no prior experience working with an industrial robot (77.8%).Participants signed consent forms at the beginning of the experiment.After the experiment, they either received course credit or a little snack as compensation.

Apparatus and task
The laboratory was arranged as a collaborative assembly workspace (Figure 1).The human workspace was arranged partially on the table where the cobot was operating, and partially on an adjacent table.The robot used in this experiment was a cobot from Franka Robotics GmbH equipped with one arm with 7 degrees of freedom, a range of 855 mm, and a maximum cartesian velocity of 2 m/s.Programming was carried out via the internal GUI, in which the speed settings are indicated as a percentage.The main task for the robot during the collaboration was to hand over LEGO building blocks from an adjacent table to the participant.
The trajectories the robot traversed remained constant throughout the experiment and were programmed using the GUI.While transferring the building blocks sequentially, the robot proceeded to the table, seized the board, moved to the handover point, released the board, and then moved to a waiting position outside the handover area.Depending on the speed, the robot would then wait for 0 to 11 seconds to create consistent cycle times regardless of the speed.
The participant's job was to assemble components onto the building block.This was supposed to simulate the manual assembly of a circuit board in the electronics industry.Additionally, a secondary task on a tablet was included, which was previously used in HRI research to simulate multitasking requirements [25].A blue rectangle would unpredictably appear on the screen every 10 to 30 seconds, and the participants were required to tap it within a 3-second timeframe.Failure to do so resulted in it disappearing, which was counted as a miss.This task was chosen, because based on Wickens' theory of multiple resources [33] similar resources to those in the primary task were demanded.Hence, the performance of the secondary task served as a workload indicator, refecting the remaining capacity of resources.

Design & Dependent variables
The study consisted of a within-subjects design with the factor of robot speed adaptability and two levels of default vs. adaptation.
The main dependent variable was mental workload, which was assessed by three measures.Firstly, it was subjectively measured using the simple version of the NASA Task Load Index (NASA-TLX) [1,19,20].The questionnaire consists of six items regarding mental demand, physical demand, temporal demand, performance, efort, and frustration, rated on a scale from 0 to 20 and accumulated for the overall score.Secondly, as a performance measure, the average reaction time and the proportion of misses to presented stimuli in the secondary task were used.Additionally, the primary task performance was measured by tracking the mistakes in assembly.Lastly, the average heart rate was deployed as a physiological measure.Acceptance was measured using three items of the TAM reloaded [5], while the item for intention to use was rephrased to refer to the speed adaptation.Perceived control was assessed using two items from Hinds [21].The desirability of control was measured using the questionnaire by Burger [7] consisting of 20 items.

Procedure
After attaching the heart rate monitor chest strap Polar H10, the participants flled out consent forms and read the instructions for the assembly of the circuit boards.To accustom themselves to the task, they then completed three boards with handovers performed by the experimenter.
Subsequently, both conditions with a prior speed-display block were presented counterbalanced with a short break between the conditions.The speed-display blocks were used for adapting the speed in the adaptable condition and prior to the default condition to guarantee the same initial experience with the robot.Independently of the condition, participants were informed that the robot was going to present its part of the task with increasing/decreasing speed in the following trials.The robot consecutively handed over fve boards, with the speed increasing (in 10% steps) with each handover from the lowest (30%) to the highest (70%) speed setting.Conversely, the robot then handed over fve more boards as the speed decreased from the maximum to the minimum setting.This approach was inspired by the method of limits approach [11].Participants in the adapted condition for this block were instructed to give feedback, once a speed was reached, that they found comfortable during the increasing and decreasing interaction.They were informed, that choosing a higher speed would not result in a faster cycle time because of the additional waiting time.We used the two resulting speed values (i.e., from increasing and decreasing) to calculate the mean speed.This was set as the speed in the adapted speed condition.In the speed-display block of the default condition, participants could not adjust the robot's speed.Accordingly, no feedback was required during increasing/ decreasing speed interaction and the robot was set at medium speed (50% of maximum speed).
In the following interaction blocks with ten handovers each, the secondary task was integrated.After each interaction block mental workload [1], perceived control [21], and acceptance (i.e., perceived usefulness and perceived ease of use) [5] were measured.The preference for default vs. adaptable speed was measured after the last interaction block only.Lastly, the participant flled out the sociodemographic questions and the questionnaire measuring the desirability of control [7].The whole experimental procedure lasted about 50-60 Minutes, while the interaction with the robot took about two-thirds of the time.

RESULTS
All t-tests were conducted two-tailed at a signifcance level of .05.

Default vs. Adapted Speed
In the adapted condition an average speed of 54.72% was chosen, (=7.17%).A one-sample t-test indicated that the speed adapted by participants was signifcantly higher than the default speed of 50%, (35)=3.95;<.001.

Individual Diferences
A linear regression analysis was used to test if the individual desirability of control predicts participants' intention to use an adapted speed.The results of the regression indicated the predictor explained <1% of the variance ( 2 <.01; (1.72, 34)=-0.03;=.883).Even though desirability for control did not predict the intention to use adaptability (=-0.01), the high intercept on a scale with a maximum value of 7 is at least noteworthy (=6.13), as it indicates a possible ceiling efect.

Exploratory Analysis of Sequence Efects
The speed participants chose in the adapted condition was not signifcantly diferent between the frst and second block ( (35)=0.94;=.352), indicating that the following results are a consequence of sequence efects rather than diferences in experienced speed.We compared the frst vs. second interaction blocks with each other for all dependent variables that were assessed in both blocks.To account for the multiple testing issues we used Bonferroni adjustment.

DISCUSSION
The purpose of this study was to gain a better understanding of user-adapted speed in industrial HRI.While none of our hypotheses could be fully confrmed, the results of the current research yield three key fndings.
First, the adapted speed was signifcantly higher than the medium default speed.We selected the default speed based on prior research that explored the optimal speed in industrial HRI [4,16].Nevertheless, the adapted speed was approximately 5% points higher than the default value.Although the medium speed appears to serve as a suitable anchor for an appropriate speed, various factors such as proximity and the size of the robot can impact the preferred speed [4,16,27,29].Thus, the suggested method of indicating a preferred speed via ascending and descending speeds could ofer an easy-to-implement approach to indicating a preferred speed in industrial HRI scenarios.
Second, adapted speed could be a mixed blessing, with apparent positive efects on a subjective level and potential negative efects on a behavioral level.Participants preferred the adapted speed and expressed a higher sense of control in this setting compared to the default speed.Nevertheless, the increased reaction time in the secondary task performance could suggest that the increased (actual and perceived) control might have inadvertently introduced another task: controlling the implementation of the adapted speed.The concern that user-induced adaptability could result in an increased workload due to the introduction of an additional task (i.e., the management of adaptation) is well-documented in humanautomation interaction research [8].While we initially did not anticipate this to be an issue in our paradigm, given that the adaptation was performed before the actual interaction, it appears that even with a static speed, monitoring the robot's speed might have diverted visual resources from the secondary task performance.However, further research is required to clarify this efect.
Third, workload and perceived usefulness were afected by a sequence efect, indicating that the novelty of interaction resulted in a raised workload and reduced perceived usefulness in the frst interaction block compared to the second one.A typical limitation in industrial HRI studies is that participants often interact with such a cobot for the frst time, leading to a substantial novelty efect [4,30].However, considering the novelty efect within the study design is less frequently addressed.This result further underscores the necessity of familiarization phases and careful counterbalancing of conditions in industrial HRI studies to account for a possible within-study novelty efect.
Although the present study represents a frst attempt to address the potential of speed adaptation in industrial HRI, it is appropriate to recognize several potential limitations concerning measurements (e.g., suitability of heart rate as workload measure) or the design (e.g., static adaptation once before actual interaction).We suppose that further research examining the diferences between prior and on-task speed adaptation compared to default settings may shed light on possible negative efects of speed adaptation.Moreover, although the generality of the current results must be established by future research, the present study has provided clear support for a preference for adapted speed and a higher perceived control on the part of the human.