Improving Mobility by Assisting Cognition: A Case Study | Accessibility VR Meetup Recap

ROMAIN: So hi, everyone. And thank you for joining me for this talk. I am Romain Delgrange, and I work in the lab of applied psychology and ergonomics at the University of Paris. I will present to you a prototype of a navigational aid to assist way finding.

We do so by evaluating tasks in the video game Grand Theft Auto V, and I'll show you the results regarding the prototype we were testing and highlight the methodological specificities of this work, which could represent an interesting starting point for studies in cognitive science, exploring the cognitive and behavioral processes that occur in realistic virtual environments. In this first part of the talk, I will present to you the theoretical framework of this study.

And our objectives. I'd like to introduce our framework by asking the question: What is spatial navigation? Spatial navigation is actually something we all do every day. Whether you try to reach the destination outdoors, to meet people, or go to places, or indoors, when you move within a building, or even in the virtual environment, you navigate in space.

It is the first step of most daily activities. It is sustained by two key components: Locomotion and way finding, which will be our focus of attention for this talk. Way finding is the cognitive counterpart of locomotion. It is the construction and use of internal spatial representations in order to find and follow a path linking two places.

There are two types of internal representations. Landmarks, routes, and survey. Landmark representations are seen from your own point of view. They are the first representations you construct whenever you get around a new environment.

For example, you know there is a blue building at the corner of a street. Routes are visual and sensorimotor representation. They represent the path that link multiple landmarks, as seen and felt from your point of view. Finally, survey representations are the most complex type of spatial representations.

They are independent of your own point of view. And it's like a view from above. These representations require you to have already explored the environment, so you can imagine it, as if you were looking at a map. Considering the central role of cognition in way finding, way finding can be challenging to anyone.

Depending on how difficult an environment is to read, or depending on your current state of mind, fatigue, or stress. It can be especially challenging for people facing permanent cognitive disabilities. Little is known on how people with cognitive disabilities deal with way finding in their everyday life. However, some studies have related that these people avoid going outside, for fear of getting lost.

They also express several needs. They would like to go out shopping or visiting relatives, going to museums or libraries, if they could. In a previous study, my team and I explored the travel chain of people with cognitive disabilities. Our results showed that these people often get lost and have to ask for help, more than people without cognitive disabilities.

For example, they ask a by-stander to help them get to their destination. But asking for help is not a strong problem solving strategy. People with cognitive disabilities often face difficulties to explain what the problem is. People around them, as well as transportation workers, therefore, face difficulties to help them.

Since way finding is a costly cognitive task, many aids exist and aim at reducing the cost of this activity. One could ask why people with cognitive disabilities do not use these aids and prefer to ask a by-stander, or a relative, for help. These aids are available on smartphones and provide audio instructions with a voice, just like a real person would. But if these principles seem similar, there is actually a major difference.

These aids mostly use bird's eye view survey information of the environment, like "head North". Thusly, a real person uses landmark information, when they describe a route. This might be one possible reason why people with cognitive disabilities ask someone instead of an app, despite being able to use a smartphone for all other purposes. Some studies have suggested at least two principles that could be adapted to navigational aids, to make them more human friendly, especially for people with cognitive disabilities.

First, they should use landmark-based information. Second, they should use audio instructions. However, very few studies have tested prototypes that could meet the needs of people with cognitive disabilities. The scarcity of studies on this matter can be explained at least in part because of two challenges: The difficulty to provide realistic virtual environments to experiment in a standardized manner, and the difficulty to recruit and categorize people with cognitive disabilities.

The first challenge deals with experimental virtual environments. Many studies in the field of spatial navigation use virtual reality. This tool helps to overcome the limitations of the real world. It allows standardization of experimental conditions for many different participants.

However, there is no turnkey solution. It means that most of the studies which have used virtual environments had to develop them from scratch. And lifelike virtual environments are hard to develop. But video games do provide realistic virtual environments.

However, they are rarely, if ever, used in such studies. A second challenge deals with the cognitive disabilities themselves. Some studies do simulate disabilities. Blindfolding is one of the tools that is frequently encountered.

But how can we simulate a cognitive disability? Unlike other disabilities, they cannot be standardized by a device. Therefore, the study of way finding in people with cognitive disabilities remains an emerging field. But we know from other studies that a certain type of task can limit cognitive resources, and subsequent performance on tasks.

The interfering tasks. These tasks are used to explore the role of short-term memory or working memory in certain conditions. It could be a possibility to use these kinds of tasks not to explore working memory, but to observe the performance under cognitive constraint. Therefore, our main objective for the study is to assess the effects on way finding and usability of a navigational aid whose principles could meet the needs of people with cognitive disabilities.

But we have two experimental goals in this case. First, we want to assess the use of a video game as a virtual reality tool to study way finding. Secondly, we wanted to explore the use of an interfering task as a means to simulate cognitive disability. Assessing this methodology could make this study a starting point to facilitate further research on this matter.

If this first study shows interesting results, its methods could be used to design more specific studies on accessibility for cognitive disabilities. In this section, I will highlight the main parts of the study. I will detail the experimental design, the setting of the material, and the data I was interested in collecting. This section of the presentation is based in part on the article "Exploring Human Behavior with Grand Theft Auto V", which you can find in open access of the website of the International Journal of Virtual Reality.

Grand Theft Auto is commercial software, and all the uses that I will discuss in this talk have been licensed by Take Two Interactive Software Corporation. Our study distinguished two major factors: The prototype for navigational assistance that was assigned to the participant, between a landmark assistant and a survey assistant, based after the spatial representations that I mentioned earlier. And the existence of a simulated disability that was based on the implementation of an interfering task.

I will discuss these two factors in the next slide. 90 participants took part in the study and were divided into four groups, depending on these two factors. For all 90 participants, the procedure was the same, and I will highlight three main steps in this talk. First, a way finding task in Grand Theft Auto V. Secondly, there was a landmark recognition test.

And finally, there was a usability evaluation made by the participant of the assistant they had to use. So our first experimental factor was the type of navigational assistant that was assigned to the participant. The two prototypes of navigational assistants we used in this study were operated thanks to a wizard protocol. Meaning that the participants were under the impression that they were using a fully functional device, but it was a speaker wired to my laptop, which transmitted instructions I had previously recorded.

So the instructions were divided into three parts. The first and third part being the same, but the second being different, depending on the assistant. The first always stated the landmark that was visually close to the participant. For example, it said: You are near the Astro movie theater.

The landmark assistant used the previously stated landmark to ask participants to perform an action linked to the landmark, according to their own point of view. For example, go along the theater on the right. Whereas the survey assistant did not make use of the landmark that was stated in the initial statement. And gave a direction independent of the participant's point of view.

Like: Head Southwest. Which is extremely close to what we can hear when we use Google Maps or other applications like this. In the end, both assistants gave the same closing statement. For example, then stop at the intersection.

The second experimental factor in this study was the condition in which the way finding task was performed. Participants in the simulated disability condition had to perform a spatial tapping task while they moved in Grand Theft Auto. This task required the participant to continuously type on a small keyboard, at a pace of one key per second, and counterclockwise. This task was chosen because of its specific effect.

It induces a limitation on spatial working memory. It limits the retention of spatial information. It is therefore adapted to spatial tasks. Apart from the prototype assistants and the task, Grand Theft Auto represented an important part of our study material.

I suggest you turn around to look at the video on the other side of the room, and I will show you how we designed the environment.

THOMAS: So let's follow Romain to the back of the room. We're always trying to innovate here at our presentation. Joly, can you make sure you let us know that you can see the video on the YouTube stream as well? It's gonna be behind you.

We will click play once you're ready to stream.

ROMAIN: Okay.

THOMAS: So Romain, can you restart?

ROMAIN: Sure, yes. So apart from the prototypical assistant, Grand Theft Auto represented an important part of the study. Like I said. And the experiment was carried out using the director mode, which is a feature included in the game.

That enables the environmental and visual parameters of Grand Theft Auto V to be controlled. Grand Theft Auto V was configured specifically for the experiment so as to be identical across conditions and participants. For example, the time was set at 8:00 in the morning. The time clock was frozen.

And the weather was clear and sunny, like you can see on this video. So that it enabled the participants to orient themselves thanks to the location of the sun in the Eastern direction of shadows. Pedestrians and traffic densities were set to low, in order to make the environment vivid, with limited risks of collisions between participants and non-player characters. Importantly, the point of view was set on first person, as Grand Theft Auto V is by default a third person video game, but allows first person view.

So that it could mimic human vision as closely as possible. Also, some of you who have played the game might wonder about the mini-map, which is by default a part of the gaming interface. The game also provides useful settings to modify the head-up display. We have therefore hidden entirely this interface, so you can see on the video, so that the screen was free of any information, including the mini-map.

So I can answer a few more questions about the environment at the end of the talk, if you would like. I'm going back to the slides, so please follow me.

THOMAS: Thank you. That was great. And Joly, are you back at the slides? Joly is back.

ROMAIN: Okay. Great. So with that in mind, regarding the navigational assistants, the interfering task, and Grand Theft Auto -- the way finding task was set as follows: The participant was sat in a chair in front of the screen which displayed Grand Theft Auto V. A loudspeaker transmitted the instructions of the navigational assistant.

The participant could move in Grand Theft Auto V, thanks to a gaming device that is a navigational controller. It allows the participant to move in-game by moving their feet with rocking and pressing movements. The participant in the dual task condition had to perform the spatial typing on the keyboard. The first task of the experiment was the way finding task.

It was divided into two major steps. First, a guided trip and then a short trip. During the guided trip, the participant had to follow the directions given by the navigational assistant to get from their starting point to two different shops successively. The route was the same for all participants.

This enables the comparison of the efficacy of both navigational aids. As soon as the participant reached the second and last shop, they had to come back freely to their starting point. Meaning that they did not receive directions from the assistant anymore. Importantly, they had to come back by using the shortest possible way, according to them.

This allows us to see if the participant had constructed internal spatial representations of the district while performing the guided trip. During the way finding task, the participant's screen was recorded on video. Performance was measured by cutting the video recordings with open software which allows accurate analysis of the videos by being able to measure everything that happens on the screen. For the guided task, we wanted to compare the way finding performance of the groups of participants.

I collected a number of mistakes, visual explorations, and stops. These data were objective markers of the efficacy of the assistant. For the shortcut task, the assistant was not used anymore. We wanted to observe, as a participant, while following the instructions, their constructed spatial representations of the environment, at least, being able to come back to their starting point by a shorter route.

So I collected route segments in the wrong directions, as well as route segments that were previously used during the guided trip. I also timed the duration of the shortcut and calculated an indicator of the other quality of the shortcut, by calculating the ratio of new route segments in the right direction in the overall shortcut trip. After the way finding task, the participants completed the landmark recognition task. During this task, the screen displayed several landmarks, some of which had appeared on screen during the way finding task.

The participant had to recognize which landmark had been seen, and which landmark did not appear during the trip. Scores and response times were recorded. The last measure was the usability evaluation of the prototypical aid. While objectively measured, effectiveness and efficiency on a given task, like the evaluation we made on the videos, indeed play a part on the experience a user will have of any system, subjective evaluation of these parameters measures the actual usability of a system.

There is a possibility that a system enables its users to effectively complete a given task, but at the expense of much time and effort that lead to this system being unused. The usability of both navigational assistance was assessed in order to deepen the analysis of the effect of overall user experience. So at the end of the task, the participants had to complete the system usability scale, which is a questionnaire that gives an overall usability score that represents a perceived efficacy of the system, as well as the satisfaction by the user during use.

This questionnaire is composed of 10 items, and a score higher than 71 out of 100 indicates a good system. Now that I have summed up the procedure of the study and the main data we collected, I will present you the results we obtained and what we can say from them. I will also emphasize the interests of the tools we used in this methodology. So I'll first take a look at the impact of the navigational assistants on way finding performance.

When we look at these graphs, we see that with the landmark assistant, the graph on the left, the landmark assistant, which was our new prototype, the participants made fewer mistakes. Also, they made fewer visual explorations of the environment, and they made fewer stops also. So it enables us to say that the landmark assistant induces a better efficacy of way finding, as a participant can go from their starting point to their destination without making a lot of mistakes, and without making a lot of hesitations, which contrasts with the survey assistant.

The shortcut trip enabled to measure the internal spatial representation of the district, while being aided by the assistant. Two results can be highlighted from the analysis of the video recordings. After having used the landmark assistant, the participants, while making fewer mistakes and so on, perform a lower quality shortcut. In particular, they retrace their own steps from the guided trip, rather than finding a new shorter route.

So while the way finding performance suggests that a landmark assistant is superior than a survey assistant, these results provide a more balanced picture. When they are not guided anymore by the navigational assistant, people who are guided by the landmark information are less able to come back to their starting point by using a shortcut. This suggests that they have not constructed sufficient spatial representations of the environment. These objective measures are interesting, but we also want to see how they translate into the subjective evaluation of usability that is made by the user at the end of the experiment.

So let's take a look at the usability scale. When we look at these results, we see that the landmark assistant...

THOMAS: Sorry, Romain. Can we pause for one moment? We want to make sure that the stream video is clear. So I think we want to just have you move slightly to the...

Your avatar move slightly in the opposite direction. A little bit more.

ROMAIN: Here? Okay. Sorry. So...

I think this slide was over... Okay. So we were talking about usability. So we see that the landmark assistant is evaluated as good, which means it is considered an acceptable system.

Whereas the survey assistant is just evaluated as okay, which means it is a marginally acceptable system. It means that the users judged the landmark assistant as more effective, and that they are more satisfied when they use this prototype. Note that these results, as well as the previous results on way finding performance and construction of spatial representations are similar between participants, whether they performed the interfering task or not. So judging by these results only, one could note that the interfering task is not an effective way of simulating a disability.

But other results allow us to sketch out a more nuanced picture. The interfering task presents several effects that I will highlight here. Interestingly, these effects only deal with the construction of spatial representations, which was to be spatial tapping with spatial memory. First, when performing the spatial tapping, participants are less able to come back to their starting point during the shortcut trip.

Secondly, the performance is lower on the landmark recognition task. Their score is lower and their response time is longer. It is interesting to note that this lower performance comes after the participants have performed the interfering task. The participants do not perform the interfering task during the recognition task.

That means that the interfering task indeed had an effect on the retention of spatial information during the trip, and that we can observe this effect afterwards. There is also one unexpected result. Even if a few participants performing the interfering task manage to come back to their starting point, those who manage to do it do it faster than the participants who did not perform the task. So there seems to be a boost effect, meaning that some participants, a few of them, have focused their attention more as a response to this demanding task.

So we have seen that landmark based information induces better way finding efficacy. Moreover, a prototype using this type of information is judged more usable. However, these contain some benefits. They are considered less usable than landmarks.

An interesting perspective could be to combine both types of information into one navigational aid. However, further research is needed to determine the way this information could be combined, and if they can actually work together on an app. Some perspectives opened by these results suggest to test a prototype using both types of information, but also to develop a prototype closer than what we did to the real thing. We used a simplified aid, which only gave audio instructions, like the audio features of Google Maps, but without the user interface.

Further research could use these results to try to develop a prototype closer to actual apps, giving both audio instructions, as well as visual information. On a methodological point of view, this study presented the use of an interfering task to simulate a cognitive disability. This tool, while well known in the field of cognitive science, has not been used before, in order to simulate a disability. Its goal was, of course, not to completely resemble a real disability, as the target of the disability was the spatial working memory only.

However, some insights can be obtained from our results. First, the task is effective to some degree. Some performance regarding the construction of spatial representation are indeed lowered by the completion of the task. However, a few participants seem to benefit from the task, which was not expected.

However, despite these divergent results, the ease of implementation of an interfering task, which requires little to no equipment, opens perspectives for future research. Other interfering tasks can be assessed for this purpose. For example, counting backwards is another frequently used task that could well be used to simulate another type of cognitive disability. In order to confirm the results obtained with this simulated disability, however, the performance on specific tasks of participants without disabilities performing the task and participants with an actual cognitive disability should be compared.

Finally, we used a video game, a widely available video game, as a research tool to conduct the study. There are several qualities and interests of the video game-based methodology. First, the effects of landmarks on way finding that had been found in different studies and that are expected to be found in real life are found with this video game-based methodology. It means that there is an experimental validity to the real world.

Also, the environment and settings can be precisely described and they can be reproduced for any research using Grand Theft Auto V and in any lab using the game. There is a real transparency of this method. It is easy to replicate the study. Moreover, all useful settings, like I mentioned before, could be operated in-game.

I will conclude by saying that this game methodology represents a low cost and turnkey solution for all researchers who want to observe the cognitive processes and behaviors that can occur in a lifelike virtual environment. The price for the lab was just $20 US, and there was no need for any other lab customizations. Of course, there are a few downsides to the use of a video game as a research tool. You can't change everything in a video game like you could do in a lab made virtual environment.

However, this is a very good compromise between the realism of the environment and the controllability of the environment. For researchers who try to develop a protocol without much time, budget, or people, a video game is an interesting means to achieve a high level of fidelity, a lifelike environment, and substantial methodological qualities, just with the features that are already implemented in the game. I want to conclude this talk by thanking you all for your attention.

I hope you enjoyed it and I hope it gave you a few ideas for your next project. I am available to discuss with you this presentation. And to answer some of your questions. So I am also available on social media, if you have any question after the talk.

Thank you very much.

THOMAS: Thank you! All right. I hear people doing some clapping. You can actually do claps by typing the digit 5, I think.

That's an accessibility feature. Hi, Tamaris and Suzanne. Yeah. I wanted to start with a first question, Romain, about the simulation of the cognitive disability, just to reemphasize.

Could you talk about using the numpad controller? And how you used that to actually simulate cognitive disability?

ROMAIN: Yes. So this task was for participants to use a small keyboard, which was a numpad, which displayed several keys. And the participant had to type on those keys at a pace of one key per second, and counterclockwise. The goal of this task is to limit the retention of spatial information in working memory.

Actually, this task has been used in multiple studies on way finding. But only to explore the role of working memory. So it was a first to use this task only to provoke a difficulty in spatial processes. It was -- the keyboard was wired to my laptop, and I recorded the keys that were pressed by the participant.

So to ensure that the task was indeed performed during the way finding task. So it was actually quite easy to implement. And it was -- like I said, it was effective to some degree. Because we see that the performance is lowered on the tasks that require spatial representations.

However, if I may highlight one limit of this task, I would say that this interfering task is known to provoke a small interference. It is not impossible for the participant to perform the task. That's why the difference between the groups of participants with and without the interfering task are not that different. But they are different.

But the difference is not extremely high. So it was a first try. The results are interesting, and promising. So it would be interesting to deepen this type of protocol in future research.

THOMAS: Thank you! If anyone has a question, if you can unmute yourself and ask it, or type it into the chat... And Stacey, if there are any questions from YouTube, please let us know. Thank you.

May I ask a quick question? Yeah. So how big was the sample size of your test subjects? How many people did you test with?

ROMAIN: We had 90 participants. Which were in four different groups. So each group was between 21 and 24 participants. So it was quite a large sample.

And it was one of the interests of this protocol to simulate a disability. Because since it's often very difficult to recruit people with cognitive disabilities, it was a way to have enough data to perform analysis by recruiting people without disabilities, and having 90 people to perform the same tasks.

And follow-up question. How many percent of your participants actually have played that game prior to the experience? Or the experiment?

ROMAIN: That's an interesting question. Actually, that's something that I did measure. And I did not have the time to present these results in this presentation. So thank you for your question.

Actually, a third of the participants had previously played Grand Theft Auto V. So it represented something like 30 participants. So I had designed a questionnaire to evaluate their previous use of the game for personal purposes. And I wanted to measure, actually, if the participants knew the game, and if he knew the experimental districts that were used in the experiment.

And so the participants answered the questionnaire. And after I did an analysis of correlations between the answers of the participants who had played the game before and the performance on all tasks, and the results showed that there was no effect on any performance. So the participants who were gamers and who used to play the game for themselves did not present any benefit in the performance. It was something that I felt was interesting to measure, in order to ensure that there was no positive effect of previous use of the game.

Awesome. Thank you so much.

ROMAIN: You're welcome. Thank you.

I will just follow up with another question, because I'm really excited about this. You said that the game mechanics and interface were adjusted? The interface was adjusted, in order to simulate this kind of first person view, and taking out any mapping and landmarks. Assistive stuff that's in the game itself.

So did you have support from the developers of the game? In order to equip the environment? Or did you have your own programmer to make modifications? How did this all logistically work?

ROMAIN: So yes, that's a very interesting question. It gives me the occasion to highlight this part of the setting of the material. So what's really interesting, and why I advocate for the use of video games as an easy way to develop a virtual environment, is that we did not have to work with the developers of Grand Theft Auto V. It was -- all the features to set the environment are already implemented in the features of the game, which is called director mode.

The director mode, which you can access from the menu of the game, gives you all useful tools to adapt the environment, head up display, and so on. So just with this mode, it was possible to do almost everything we wanted to do. Of course, to find the experimental districts that would fit the purposes of the experiment, we had to explore the virtual city of Los Santos. We had to look for the right streets with the right landmarks.

But apart from that, it was just like going in... You would have the same things to do, if you went to a real city. And you wanted to look at different streets, to design an experiment. So that was really not difficult to do this.

And after, you can set everything from the time of day to the pedestrian density. Just with the director mode.

Very interesting. Thank you so much.

ROMAIN: You're welcome. Thank you.

THOMAS: One more question from me. This is Thomas. I just got back into the room. So the director mode that you mentioned from Grand Theft Auto V -- that had all the configuration options that you needed to do your research?

Or did you need to request anything else from them?

There was maybe one thing. Sometimes it could have been interesting to edit the environment a bit more. But you can do -- there are some editing features. With the director mode.

You can add several objects. So sometimes it was useful to add these subjects. For example, to avoid an effect of cognitive silencing, we put objects before the names of some stores, which were written in French, and we didn't want the participants to remember the names of these shops. Because it was written in French.

So we used the editing features of the director mode just to put an object before the name of the shop.

THOMAS: Oh, like you could block it out?

ROMAIN: Yes. Actually, it was just to visually neutralize the shop.

THOMAS: That's cool.

ROMAIN: These editing features are quite simple. You can just add some objects, but you can't change the size of the objects you add and things like this. So I would have liked the features to be a bit more precise, but really it's already very well as it is. You have to remember that this is a video game, and this is not a research tool.

So to use it as a research tool with little problems -- it's already very interesting. I think there was a question in the chat. But I just missed it. And I don't know if I can...

THOMAS: Yes, this is actually a critique of Mozilla Hubs. Because I was trying to take a screenshot. Suzanne Lee -- is it possible for you to join via audio and ask your question? That would be great.

Sure. Can you hear me okay?

THOMAS: Yes. We can.

ROMAIN: Thank you.

Awesome. I just wanted to say that was some amazing work that you're doing, and I had a question around the 3D Rudder, and whether that was used alongside the numpad? Or whether you used it separately. And pretty much how did your participants find using the 3D Rudder hardware during this.

ROMAIN: Yes. So the 3D Rudder was used at the same time as the numpad. Because it was used all along the way finding task in the video game. So the participants were all using this 3D Rudder, and what was really interesting for me, and the reason why I wanted to use the 3D Rudder, was that participants could use without moving their hands, which was interesting for me, because half of the participants of the study had to use their hand to perform the interfering task.

So this was really a good way to let the participant move freely in the environment without making them use their hands. Like they would have had to do with a joystick. Also, this is very interesting for studies which explore movement and mobility. Because like I said at the beginning of the talk, there are several types of spatial representations.

One of them being the representation of the routes between different landmarks. And the fact that the participants used their feet to move, even if it's not in the way that they would use their feet in a real street, they have a sensorimotor experience of their trip. So it is closer to a real mobility situation. Than just with the keyboard.

And most participants were really at ease with the device. So the 3D Rudder was really adapted to this study. Before the tasks, all participants took part in a training session, which lasted like 15 minutes, just before the task, just to ensure that they were at ease with the device, and the interfering tasks and the virtual environment. So during the task, there was no problem with the 3D Rudder.

Hey, can anybody hear me? Finally. First off, thank you for the talk. I have a question regarding results.

You've mentioned that when using the survey method, that participants spent more time exploring the environment, compared to the alternative. I would expect the opposite, since they wouldn't have to take in the landmarks when doing so. Do you have an exploration of that particular observation?

ROMAIN: Yes. Actually, they did not spend more time to visually explore the environment. But they did it more. That means that they had to look around them more times than with the landmark assistant.

So that means that when they were using the landmark assistant, the participants did not have to look for landmarks other than the landmarks that were told to them by the assistant. The assistant chose a landmark that was visually close to the participant. So they did not have to look for a landmark that was really far. The landmark was just in front of them.

And the assistant taught them to go along this landmark on the left or on the right. Whereas the survey assistant was really different, and the participant had to take into account the direction of the sun and shadows to ensure that they were in the right direction. So it was more difficult for them, and this is a possible explanation of why they had to look around more.

I see. Thank you.

ROMAIN: You're welcome. Thank you.