1 Introduction

Understanding product assemblies is important throughout the development and life cycle of a product. Virtual assembly employs cutting edge hardware in the creation, design, and evaluation of assemblies. As part of this technology, bimanual haptics renders realistic force feedback to create an immersive experience for manipulation. Bimanual haptic applications have been used for a wide variety of purposes from explosive ordnance disposal (Kron et al. 2004), to surgical training (Hinckley et al. 1998) and surface and curve manipulation (Owen et al. 2005; Shaw and Green 1994). They have also been used for 3-D object manipulation and interaction in virtual environments (Balakrishnan and Kurtenbach 1999; Bowman and Hodges 1997; Poupyrev and Ichikawa 1999). Fiorentino et al. (2010) found that the incorporation of bimanual haptics aided CAD designers in model creation. Talvas et al. (2013) describe a wide range of bimanual haptic interaction techniques

Bimanual virtual assembly has been found to outperform unimanual virtual assembly in a variety of situations, including perception of weight (Giachritsis et al. 2009; Owen et al. 2005), virtual navigation for the visually impaired (Crossan and Brewster 2006), and cooperative tasks (Gunn 2006; Hinckley et al. 1997). The increased performance of bimanual assembly has been explained using a framework that models the two hands as two motors connected in series. This model, known as the kinematic chain model (Guiard 1987), emphasizes an asymmetric division of labor between the two hands where the dominant hand moves in reference to the non-dominant hand. An asymmetric task is a task in which each hand does a uniquely different task while a symmetric task is where each hand does the same task. For example, simultaneously opening two oppositely hinged cupboard doors is a symmetric task while dealing playing cards is an asymmetric task.

Studies of asymmetric and symmetric bimanual tasks have resulted in a series of insights into the way humans use both hands when completing various tasks. According to Hinckley et al. (1997), bimanual interaction is optimal when each hand assumes its most effective role. In a 2-D bimanual symmetric tracking task, divided attention, task difficulty, and a lack of visual integration can decrease performance (Balakrishnan and Hinckley 2000). However, the most common bimanual interaction is one in which the non-dominant hand is responsible for gross motor movements while the dominant hand performs more fine motor positioning (Hinckley et al. 1998; Marteniuk et al. 1984). Supporting research indicates that the non-dominant hand is generally used for lower-frequency and higher-amplitude movements and the dominant hand is used for higher-frequency and lower-amplitude movements (Peters 1985). This has led to research into the use of devices for bimanual assembly in virtual reality.

Vyawahare and Vance (2009) proposed a bimanual device configuration that incorporates haptic feedback by using a position-tracked glove on the one hand and a haptic device on the other hand. Referencing Hinckley’s model of bimanual interaction, the glove (non-haptic) would be placed on the non-dominant hand and be used to select and position virtual objects and the haptic device would be controlled by the dominant hand where the user would perform fine positioning movements. The main benefit in implementing this configuration is that the user is able to feel haptic feedback during crucial assembly operations and able to reach parts beyond the haptic workspace. In this way, task performance in the configuration supports haptic, kinesthetic, and visual representation of the bimanual assembly process.

A more recent paper outlines a similar asymmetric interface using haptics and a Razer Hydra™ (Vyawahare and Stone 2012). The Hydra provides two controllers with button and joystick input on each controller that are tracked using a magnetic tracking base. Originally created and marketed for video games, the Hydra has been adopted and extended by virtual reality researchers as a tracking and input device. Vyawahare and Stone used the large workspace of the Hydra to augment the interaction method and provide asymmetric interaction. An evaluation of this interface and the interaction method was performed and the utility of these methods was shown to be useful for a variety of tasks (Vyawahare and Stone 2013).

The desire to use one haptic device rather than two for desktop assembly is motivated by three factors. The first of these is cost reduction. Adding haptic force feedback to a simulation increases the cost because of the need for additional equipment. If the use of a haptic device coupled with a non-haptic device yields similar or better performance than two haptic-enabled devices, costs would be reduced by choosing the less expensive configuration. Vélaz et al. (2014) explored three different hardware configurations for bimanual virtual assembly. Their results indicated that a hybrid configuration of a haptic device on the one hand and a non-haptic device with markerless motion capture on the other hand produced the shortest time of completion for the assembly task. They concluded that the hybrid configuration was sufficient to perform virtual assembly of parts where precision fit is not critical. The research presented here compliments their work by exploring a reduced-cost configuration for bimanual assembly of precision fit parts.

The second factor is the desire to improve realism. Replacing a haptic probe with a glove provides the user with the visual representation of finger motion during grasping. However, most current glove interfaces do not have haptic feedback, but support more natural hand motion and interaction than existing haptic devices. Instead of learning how to use an entirely new device, a user can grab, select, and release objects when wearing the glove in a manner similar to the way they would handle objects in the real world.

Finally, desktop haptic devices have limited workspaces. Talvas et al. (2013) point out that this limitation can be addressed by developing unique software algorithms or by using unique hardware configurations. One software algorithm that expands the usable haptic workspace is the bubble technique proposed by Dominjon et al. (2005) and modified to accommodate bimanual interaction by Talvas et al. (2013). This technique provides the user with an intuitive means of essentially moving the haptic device within the entire virtual environment, allowing for haptic interaction at any place within the environment without the need for haptic clutching. Pavlik and Vance (2015) expanded on the bubble technique to allow grasping parts with the haptic device while simultaneously moving within the environment. Other techniques such as scaling the virtual environment (Fischer and Vance 2003) and haptic clutching (Isshiki et al. 2008) have also been investigated.

The study described in this paper is an evaluation of the Vyawahare and Vance (2009) proposed bimanual device configuration for precise fit haptic assembly. Two different assembly tasks are performed using various combinations of desktop haptic devices and a position-tracked glove.

2 Methodology

2.1 Software

The application was developed using SPARTA (Scriptable Platform for Advanced Research in Teaching and Assembly). SPARTA combines VR Juggler (VRJuggler 2013) for stereoscopic rendering and position tracking management, OpenSceneGraph (OpenSceneGraph 2013) for graphics, Voxmap PointShell (VPS) (McNeely et al. 1999) for physics calculations, and VR JuggLua (Pavlik and Vance 2011b) for easy scripting and content creation. SPARTA supports multiple input and output devices including position trackers, stereo glasses, stereo projection systems, gloves and haptic devices (Pavlik and Vance 2011a). It was developed by researchers at the Virtual Reality Applications Center at Iowa State University.

2.2 Hardware

The two hardware devices that were used in this study are the Phantom Omni® haptic device and the 5DT Data Glove 5 Ultra. The Phantom Omni® is priced around $2000 USD and the 5DT Data Glove 5 Ultra is priced around $995 USD. The Phantom Omni® provides the ability to enable or disable haptic force feedback while the glove does not have haptic force feedback . For completeness, we included the haptic disabled Omni device configuration in our treatments.

The user wore position-tracked stereo glasses to view the rear projected stereo image on the desktop screen. The glasses were tracked using an InterSense IS-900 hybrid inertial and ultrasonic tracking system. The gloved hand was tracked using a Polhemus Patriot magnetic tracker. Both commercial tracking systems have low latency with the IS-900 at around 4 ms and the Patriot at around 17 ms.

Interaction using the Omni allows the user to have a full 6 degrees of freedom (DOF) in tracking and movement, along with three DOF haptic force feedback (see Fig. 1). The user’s interactions in this study consisted of moving a virtual cursor around the scene. Once the cursor intersected a part, the part turned slightly transparent indicating that it could be selected. A button on the Omni could then be pressed to select the part and subsequent movement would move the virtual part.

Fig. 1
figure 1

Dual Phantom Omni® configuration

Full size image

Interaction using the 5DT glove was similar to the Omni, also providing six DOF in tracking and movement; however, no forces are rendered to the user’s hand with the Data Glove (see Fig. 2). In this interaction, a virtual hand representation acted as a cursor that could be moved around the scene. Intersection of the virtual hand with a part turned the part slightly transparent indicating that it could be selected. The user then made a fist gesture to grab the part and move the virtual part in the environment.

The six DOF movement and tracking of both systems provides a natural interaction that mimics real physical assembly. No scaling of movement was applied to either device to ensure that it matched real-life interaction.

Fig. 2
figure 2

Desktop setup showing the 5DT Data Glove and the Phantom Omni® configuration

Full size image

2.3 User study

2.3.1 Study design

The two independent variables were the device configurations (five treatment conditions) and the task difficulty (easy and hard). The dependent variable was performance as measured by task completion time. Examining the effect of device configuration on performance time was a within-subject variable. Task difficulty was a between-subject variable. The five device configurations are listed in Table 1.

Table 1 Treatment conditions
Full size table

Each participant was randomly assigned to one of the two task difficulties and asked to perform the assembly using each of the five treatment conditions. The order of the five treatment conditions was randomized to account for fatigue and learning. The easy task consisted of insertion of one virtual object into another (Fig. 3). This required orientation and insertion of the object. The difficult task required object orientation, insertion, rotation, and finally insertion (Fig. 4). This task is similar to the insertion of a key into a lock, followed by rotation of the key in the lock.

Fig. 3
figure 3

Unassembled individual objects for the easy task

Full size image
Fig. 4
figure 4

Unassembled individual objects for the difficult task

Full size image

The assembly of these two simple parts was designed to represent common motions of assembly. Sliding a pin into a sleeve is a common motion that occurs in many assembly operations. In a virtual environment, this is more difficult than in real life, especially if the virtual environment does not impose artificial axial constraints between the two parts. The motion of rotation can also be difficult to perform in virtual reality due to the existence of a virtual spring between the haptic device and the rigid body combined with the need to rotate and re-grasp.

Upon arriving, participants filled out a consent form. Participants then completed a short pre-study questionnaire that gathered basic demographic information along with a self-assessment of previous computer and VR experience. Next, participants watched a short video demonstrating the equipment that they would be using and the interface for interacting with the virtual parts. Participants were given two minutes to become familiar with using the equipment and the application. Different models than the test models were used during this practice period and all participants used the Omni in their dominant hand and the 5DT Data Glove in their non-dominant hand. This let participants experience both hardware devices. Next, the timed data were gathered as participants used each of the five randomized device configurations to perform the assigned task. The piece that resembles a key was always placed on the left in the scene and the other piece was placed on the right. For each configuration, they were instructed to complete the assembly twice. In total, each participant completed ten assembly operations. A picture of the disassembled and assembled objects was placed on the table for reference but no other instructions were given during the treatments. Upon completion of the trials, participants completed a short exit questionnaire to gather preferential data.

2.3.2 Participants

Fifty-two participants (39 males and 13 females) completed the study. While participants were not compensated, some did receive class credit. The participants were recruited mostly from undergraduate engineering and psychology classes. The ages ranged from 17 to 36 years with a mean (M) of 22.17 and a standard deviation (SD) of 4.48. Forty-eight participants were right-handed and four participants were left-handed. No participants indicated that they were ambidextrous. The participants were divided into two groups (easy task and difficult task) resulting in 26 participants in each group. Self-reported computer experience was reported on a Likert scale between 0 (no experience) and 10 (high computer experience) (M = 5.73, SD = 2.57). Most of the participants had little experience with virtual reality as was self-reported on a Likert scale between 1 (none or little experience) and 10 (significant experience) (M = 2.94, SD = 2.38). Prior experience was low as self-reported on a Likert scale between 1 (none or little experience) and 10 (significant experience) regarding both virtual assembly operations (M = 2.01, SD = 1.79) and haptic force feedback devices (M = 1.67, SD = 1.2).

Within the participant pool, there was a significant difference in self-reported computer experience between males and females. The results of an independent two-sample t test, t(50) = −2.70, p = 0.009 indicated a significant difference between groups with males reporting their computer experience as being higher than females.

3 Results

3.1 Quantitative

For each treatment, the participants completed the assembly task twice. For purposes of data analysis, the overall completion time was calculated as the average of the two task times. A comparison of task completion time for each treatment is shown in Fig. 5.

Fig. 5
figure 5

Average task completion times for each device configuration

Full size image

In order to run a balanced ANOVA, five participants who did not finish all five device configurations because of a technical glitch were removed. All five of these participants were right-handed. This left 47 participants who completed all the device configurations. Levene’s test was not significant F(4, 230) = 0.39, p = 0.81 signifying that the assumption around equality of variances was not violated. The results of an Omnibus ANOVA were not significant, F(4, 184) = 0.10, p = 0.97, indicating that there was little difference in performance between the different treatments not including the variable of task difficulty. When including the interaction of task difficulty in comparing the five treatments, this was also not significant, F(4, 1739) = 0.41, p = 0.80. Since the Omnibus ANOVA was not significant, more specific planned linear contrasts that would examine the details of individual device configurations could not be performed. The effect of devices used during the practice session at the beginning was not considered in the analysis since all participants used the same configuration.

In examining the total time taken to finish the tasks between the easy and difficult task assignments, there was a significant difference when running an independent two-sample t test, t (233) = −4.36, p < 0.001 with the more difficult task taking longer.

Irrespective of treatment, in general, the participants exhibited a slight learning effect as they progressed through the study (Fig. 6). In particular, the increase in their performance from the first task that they completed to the second task is evident.

Fig. 6
figure 6

Times for each of the five trials regardless of device configuration and task assignment

Full size image

There was no significant difference in the total time taken to finish the task between groups of left-handed and right-handed participants, t(233) = −1.41, p = 0.16 irrespective of task difficulty.

In comparing the time taken to finish the tasks between males and females, there was a significant difference when running an independent two-sample t test, t(66.22) = 5.34, p < 0.001 with males participants completing the task faster. The results are shown in Fig. 7.

Fig. 7
figure 7

Differences in time taken based on gender

Full size image

In the exit questionnaire, participants were asked to pick their preferred device setup from any possible combination of bimanual or unimanual device combinations and dominant or non-dominant hand. There was little difference based on handedness preference; therefore, these groupings were combined and a summary of the responses is presented in Table 2. Given the multitude of possible device combinations and the low number of participants, a Chi-squared goodness of fit test was not performed.

Table 2 Preferred device combinations
Full size table

While participants were not able to try every possible bimanual or unimanual device configuration, they were able to experience both the Glove and Haptic/NoHaptic Omni. Interestingly, the majority of participants preferred at least one haptic device (42 out of 52 responses) as given in Table 2. Twenty-six of the 52 participants chose the Glove and the Omni. Additionally, participants rated the use of haptic force feedback beneficial in assembling the objects on a Likert scale between 1 (useless) and 10 (useful) (M = 7.55, SD = 2.06). They also rated each of the devices on a Likert scale for helping in the assembly process: Omni (M = 7.26, SD = 2.03), NoHaptic Omni (M = 5.36, SD = 1.84), and Glove (M = 6.55, SD = 2.65).

In general, participants had a favorable view of the hardware being useful in day-to-day use when asked to rate it on a Likert scale between 1 (useless) and 10 (useful) (M = 6.82, SD = 1.98). Almost three-fourths of the participants (72 %) said that if the technology was available they would use it daily. Most (71 %) felt the haptic force feedback increased their ability to assemble the objects with some (12 %) saying it decreased their ability and others (17 %) saying it had no effect. When asked about the use of the haptic device in their dominant hand and their ability to assemble the objects, it was gauged as being quite helpful (M = 7.94, SD = 1.92). Multiple devices were also considered to be quite helpful in the overall simulation experience (M = 7.32, SD = 1.99). When asked how natural they felt their interactions with the environment seemed, participants had a favorable view (M = 6.84, SD = 2.14) and in addition they felt that the sense of moving around was compelling (M = 7.48, SD = 1.70). In general, the virtual reality experience was deemed moderately realistic when compared to the real-world experience of assembling the objects (M = 6.07, SD = 1.93).

3.2 Qualitative

The benefit of this study design was that it gave participants experience with many device configurations and they were able to try both haptic and non-haptic devices. This helped temper their qualitative comments and responses. The last section in the final exit questionnaire was open-ended and asked participants if they had any comments about their overall experience. In general, participants had a favorable opinion and seemed to prefer the haptic Omni as given in the comments in Table 3. In addition, comments from participants seemed to echo the theory of the kinematic chain model as given in Table 4.

Table 3 Participant comments regarding device preference
Full size table
Table 4 Participant comments confirming kinematic chain
Full size table

In general, participants were supportive of the experience saying that they, “enjoyed it.” One participant said, “I thought it was awesome! I thought once I got the hang of how to do it that it was a lot of fun to work with! I am amazed that this is possible!”. Participants even thought of additional use cases for the devices besides virtual assembly such as gaming. One participant said, “The device is great and I hope this device can be obtained at an affordable price because I know people in my country with drawing and designing objects would have their work time lightened with this piece of technology”.

4 Discussion

Although not shown to be statistically significant, the study results show that the use of the Glove in the non-dominant hand and the Omni in the dominant hand for bimanual assembly resulted in similar performance when compared to the other configurations that were tested for this specific task. In general, participants performed equally well through all five of the treatment conditions. However, in answer to the open-ended question, several participants indicated that having one glove and one haptic device was their favored configuration.

While not quantitatively measured, anecdotal observation of the participants indicated that the participants who performed the task in the shortest time were those that used both hands at the same time. The task given to the participants did not require two-handed manipulation. Perhaps a redesigned task which includes a gravity force and therefore requires the use of both hands may have produced different results across treatments. González-Badillo et al. (2012) found that including the weight of the virtual objects affects the task completion time in bimanual virtual assembly tasks. Another interesting aspect to explore would be to examine the number of movements needed to assemble the part.

There is a marked challenge in comparing haptic and non-haptic devices that differ in a variety of different factors (workspace, accuracy, force feedback , etc.). These differences result in potential confounding variables that can be difficult to account for. The additional testing of other configurations of devices would be beneficial in understanding when it is appropriate to use certain configurations of devices. Additionally, it would be helpful to identify the factors that contribute to task difficulty. The lack of torsion feedback for the Phantom Omni could potentially have been a factor in task performance as well as participant device preference. Future testing could compare the 3DOF versus torsional force utilized during different types of assemblies by users.

The significant differences in task time between males and females is an interesting result. It could be due to the reported difference in computer experience between the genders or perhaps a difference in spatial ability. Gender differences in spatial ability has been identified in other studies. Men have been shown to score higher on spatial tests relative to women (Linn and Petersen 1985). The importance of spatial ability in virtual assembly needs further investigation.

There appears to be an incongruity between the time it took participants to put together the objects in the different treatments and their self-reported preference for hardware configurations. One would think that performance would be related to preference; however, participants seemed to predominantly prefer haptics and felt that it was quite beneficial in helping them assemble the objects. Why is it then that participants had a clear preference irrespective of their performance? One possible explanation is the misinformation effect.

The misinformation effect says that presenting information, whether correct or incorrect, between the encoding of an event and recall can influence the memory of the event and impair the ability to accurately recall details about it (Loftus and Hoffman 1989; Loftus et al. 1978). In this case, there was no purposeful misinformation provided to confuse participants but the constant changing of device configurations through switching hands and hardware may have been confusing to participants. Participants were told at the start of each task which device configuration they would be using; however, this misinformation effect may have influenced their memory of the treatments. In addition, because most participants did not have prior experience with virtual reality or haptic devices, the novelty of the devices could have played a role in determining what they remembered.

5 Conclusion and future work

In an effort to evaluate the usefulness of five bimanual virtual assembly device configurations, a user study was performed. The results indicated that there were no significant differences in task completion times for the five bimanual configurations tested and that male participants had faster completion times than female participants. This may be attributed to a difference in computer experience between genders. Participants also indicated that haptic feedback was beneficial in their ability to assemble virtual products and the majority of the participants preferred using at least one haptic device.

There are additional variables that were not included in this study that we would like to examine in future work. In the present study, performance was measured by the time taken to perform the task. However, the motivation for the study was to evaluate a bimanual interface that includes haptics yet expands the ability to manipulate objects to an area larger than that of the haptic device. In addition, we wanted to give participants the flexibility to perform the task in their own desired way. To better evaluate the performance, it would be beneficial to save the location and orientation of the objects to determine how users are orienting and positioning the objects prior to insertion to see whether there are differences in the way they assemble parts when using different hardware configurations as well as how much of the workspace of the glove participants used. This additional information could be used in comparing the potential benefits of the expanded workspace enabled through glove interaction as compared to a haptic device with a smaller workspace. Another possible improvement would be to force the assembly process as a bimanual task that would require use of both hands at the same time. This might be a good configuration when addressing performance characteristics of the devices. The downside of forcing the user to use both hands like that is that the experience becomes less natural and may not necessarily mimic real-world assembly.