Keywords

1 Introduction

Cognitive maps are usually considered as maplike representations of large-scale environments that are stored in human minds. They play an important role in formulating human survey knowledge as opposed to route and graph knowledge in studies of spatial cognition. One of the most well-known evidence of cognitive maps is an individual’s ability of finding novel shortcuts between two locations that have never been traveled between [3]. Beyond that, cognitive maps provide essential spatial knowledge about the environment for travel-planning and decision making. They are crucial in situations when navigation aids are not available, including when those systems fail (software crashes) or it is infeasible to rely on navigation apps (indoor when GPS signal is totally absent). Additionally, without the information in cognitive maps, potential travel routes and destinations could not be utilized during planning and might be viewed as inaccessible.

Humans build and update cognitive maps through interactions with the environments, including the experiences of travel and navigation. Recent studies suggest that spatial knowledge differs within individuals choosing passive versus active travel modes [17]. However, with the evolution of driverless technologies and the promising future such automation will bring to human welfare and the environment, travel modes such as driving different levels of autonomous vehicles are yet to be explored. On the other hand, it is still unknown what effects will the shifting from zero automation to fully-autonomous vehicles bring to individuals as well as to the entire society in the long term. Researchers have agreed that active spatial learning contributes more to survey knowledge acquisition than passive exposure in a novel environment [7]. As human drivers are not required to actively engage in driving at Level 4 to 5 automation [21], the question arises: will the adoption of fully-automated vehicles impact people’s acquisition of spatial knowledge for forming and retaining cognitive maps in the future?

In this paper, we aim to shed light on the possible consequences autonomous driving will bring to humans in terms of the spatial knowledge for cognitive maps by conducting an online survey. To the best of our knowledge, this is the first study that explores the relationships between the use of driverless vehicles and human cognitive maps. As drivers will become more like passengers in a highly or fully automated car [20], we grouped respondents by frequencies of driving as well as riding ground transportation as passengers. The self-report measures of spatial knowledge were leveraged based on Santa Barbara Sense of Direction Scale (SBSOD) [12], which has been considered highly reliable and well-correlated with tasks requiring survey knowledge. This approach allows us to get preliminary insights into the question with a large number of samples, before follow-up lab experiments are conducted, which will benefit from the general insights as well. In particular, we focus on addressing the following research questions:

  • RQ1: What are the differences in spatial survey knowledge for cognitive maps among individuals with different frequencies of driving and riding as passengers?

  • RQ2: What can we learn from these differences in regard to how autonomous vehicles will impact human cognitive maps?

The rest of this paper is structured as follows. We first outline the related work in the field of spatial cognition and automated systems. Then we discuss in detail about the method we used to formulate the questionnaire and conduct the survey. The results of the analysis are presented, followed by discussions on what we could learn from the study. In the end we conclude the paper with a brief summary and our plans for further research.

2 Related Work

2.1 Human Spatial Knowledge and Cognitive Maps

Cognitive maps are map-like, unified mental representations of the spatial environment. They are geometrically-consistent survey knowledge that brains store under a common coordinate system, sometimes known as “global metric embedding” [25]. Research interest in cognitive mapping originated from psychology, but later extended to geography and robotics as well. To build and update cognitive maps, humans acquire and process spatial knowledge about their environment through both direct and indirect experiences.

During spatial navigation, individuals rely on recalling and updating various forms of spatial knowledge besides survey knowledge. They include beacons and landmarks, route knowledge, and graph knowledge [30]. Survey knowledge, which is commonly regarded as the core component in forming cognitive maps, is “maplike” knowledge including metric distances and directions between locations. A person with such knowledge is capable of estimating spatial relations between two locations as well as placing them in a larger representation of the environment. Those abilities are widely deployed by researchers to evaluate an individual’s cognitive map in previous studies.

Humans construct and update cognitive maps over time. This process of spatial learning usually involves the experience of travel and other resources, such as reading maps or having conversations [9]. A number of research has explored the varying performance in building spatial representations among individuals with similar learning experience. For example, [13] found significant differences in participants’ abilities of forming accurate representations under the same phase of learning a new environment. The knowledge was gained through being driven around as passengers in vehicles. Researchers of another study in [23] let people learn actively by walking in a neighborhood, which also resulted in different performance during evaluation of the learned spatial representations. A detailed comparison between active and passive learning in acquiring spatial knowledge was further investigated by [7, 8], where they concluded that both visual and podokinetic information contributed to survey knowledge, while cognitive decision making was the primary component of building graph knowledge, in the context of human spatial learning.

Previous studies showed that different travel experiences contributed to variations in cognitive maps. [17] demonstrated the finding that disparities in spatial knowledge could be partially explained by where and how people travel. They categorized driving and walking as “active” travel modes and compared them with “passive” modes such as public transit and being an auto passenger. Following the similar comparison of active and passive means of transportation, [16] found through user studies with London residents that car users had more complete cognitive maps of the city than people with other modes of transport. However, none of the existing studies focus specifically on the paradigm shift from driving a non-autonomous vehicle to riding inside driverless cars as a passenger. In this paper, we aim to provide indications with regard to the impact of self-driving vehicles.

2.2 Impact of Adopting Automated Systems

In general, the impact of reliance on automation has long been explored and discussed by researchers across multiple domains. With systems requiring less monitoring and intervention, humans exhibit decreased situation awareness, poor skill acquisition and maintenance, as well as delayed reaction to problems and emergencies [19]. It has been discovered that automated systems could hinder operators’ abilities of dealing with new problems, as it is not necessary to understand how to manipulate them [22]. Another downside for automation is the reduced attention human pay to monitoring those systems, which can cause fatal accidents during system failures [10]. One widespread application of automation is automatic navigation, which has been suggested to correlate with degradation in human spatial knowledge. An empirical study conducted by [2] showed that pedestrians who use mobile navigation apps acquire poor spatial knowledge, while [5] provided preliminary evidence on the consequences of using vehicle navigation systems in terms of drivers’ cognitive maps.

Similar issues have also been studied in research on self-driving vehicles. For example, according to an online survey conducted by [27], the introduction of autonomous cars could potentially cause driving skills degradation in the long term, if drivers fail to receive proper training in the first place. To address problems autonomous driving might cause, researchers proposed several approaches: [28] assessed a driver-vehicle interface of cooperative driving that can keep drivers in-the-loop, and [11] evaluated a proposed prototype implementing shared control between drivers and conditionally automated driving systems. In order to help better understand the effect on human spatial cognition, in our study, we explore the possible impact autonomous cars have on human spatial knowledge for cognitive maps.

3 Method

We developed an online questionnaire through Qualtrics survey system based on SBSOD and literature on spatial survey knowledge, aiming to provide general insights into the possible impact of utilizing self-driving vehicles on acquiring spatial knowledge for cognitive maps. The questionnaire contains 25 questions in total, and is divided into three parts: demographics, frequencies of driving and riding as passenger, and spatial knowledge for cognitive maps.

In the first part, participants were asked some general questions including their age, gender, driver’s licence, information about their vehicles (whether they were equipped with semi-autonomous driving systems), main purpose of car use, and their experience with navigation apps. We took advantage of the skip logic during the design, so that participants could save time by not being asked certain questions based on their responses to the previous questions. For example, participants who indicated not having a valid driver’s license were not asked questions on the year they obtained the license or information about their cars.

The second part was to group participants by their reported frequencies of driving and riding as passengers in ground transportation (ride-sharing, bus, shuttle service, light rail, etc.). As more driving tasks will be passed to the vehicles, at level-5 automation no one will be performing driving or monitoring tasks [21], making everyone passengers in fully self-driving cars. With this in mind, we designed the questionnaire to address different groups based on how often they drive and how often they are passengers. We were specifically interested in possible changes in the acquisition of spatial knowledge after adopting fully-automated driving, where individuals drive less and ride more as passengers. Ground transportation was addressed in these questions, since our major focus is on vehicles instead of other travel modes such as airplanes. We asked the number of times participants drive or ride as passengers per day during weekdays and weekends in separate questions.

The last part was consisted of 10 questions on participants’ self-evaluated spatial survey knowledge about the neighborhoods around where they currently live. They were leveraged from SBSOD, a widely used and well-tested scale in the domain of spatial cognition, with the same 7-point Likert scale ranging from “strongly disagree” to “strongly agree”. Some were reworded to closely match the measurements previous studies implemented to assess cognitive maps, such as “relative locations”, “judging distances”, and “point to where I live”. We stated some questions positively and others negatively. A comprehensive list of questions could be found in Table 2. We also added a question on the time period respondents have been living in their current neighborhood.

The questionnaire was distributed on Amazon’s Mechanical Turk (MTurk) from September 2019 to January 2020. Researchers have found that the quality of data acquired on MTurk is similar to data collected from traditional methods [4]. Participants were compensated US$0.01 for finishing the entire questionnaire, which estimated to take about 10 min.

4 Results

Overall, we received 217 completed responses out of 230 recorded questionnaires. Based on the findings from [6], we identified duplicate cases by examining multiple questions on demographics from the same IP address and removed them from our analysis. Respondents who indicated not having a driver’s license but drove vehicles more than zero time a day were considered not paying close attention to the questionnaire and removed from the following analysis. As a result, a total of 204 responses was included in the final analysis.

Among the 204 participants, 74 were female (36.3%) and 130 were male (63.7%). Their ages range from 18 to 68 (M = 34.2, SD = 9.35). Most of the participants indicated having a valid driver’s license (94.6%), and the majority of them reported their primary purpose of driving was commuting to work/school (79.6%), followed by 11% who mainly drove for a living (e.g., taxi drivers). Seven of the drivers had semi-autonomous driving systems equipped on their cars (3.6%), 6 of them were Adaptive Cruise Control and 1 was Tesla AutoPilot. For the experience with navigation systems during driving, more than half respondents stated using navigation sometimes or more than most of the time (55.1%), while only 2 never used the system (1%).

All respondents were grouped into frequent/infrequent drivers and frequent/infrequent passengers according to the calculated daily trips they made on average through driving as well as riding inside vehicles. According to [24], drivers in the US made an average of 2.24 trips per day in 2006, so we identified participants who drove 2 or more trips on average per day as frequent drivers. In regard to passengers, based on public transportation trips data from [1] in 2018 (including the transport mode of demand response), we estimated people in the US took public transportation around 1 trip per day on average. Therefore, respondents indicated riding 1 or more trips per day as passengers were considered as frequent. Table 1 summarizes the numbers of responses by group.

Table 1. Number of responses by group
Full size table

We used IBM SPSS Statistics Version 25 to perform the statistical analysis on the data. Spearman correlation analysis was carried out for the questions in the last part of the questionnaire, showing significant positive correlations in all pairs of positively stated questions (p < 0.05) and significant negative correlations between all pairs consisting of one positively and one negatively stated questions (p < 0.05). Thus, it is likely that respondents were aware of the different phrasing of the questions and answered consistently with their perception of the spatial knowledge they had about the neighborhoods around where they live.

4.1 Single Factor Analysis

A one-way ANOVA was conducted for each item in the last part of the questionnaire to determine the effect of driving frequency on participants’ spatial survey knowledge. Similar analysis was carried out on the same questions with the effect as the frequency of riding vehicles as well. The results are shown in Table 2 and Table 3, with lower values indicating more agreement with the description.

Table 2. Mean and standard deviation of comparing frequent and infrequent drivers, lower value means more agreement with the statement. (*for significance p < 0.05, **for significance p < 0.01)
Full size table

In total, we identified 115 frequent drivers (56.4%) and the rest as infrequent drivers (43.6%) among all respondents. The distribution of gender is 34.8% female in frequent drivers and 38.2% female in infrequent drivers. The mean age of the two groups is comparable: frequent drivers are on average 33.56 years, while the other group is 35.17 (F(1, 202) = 1.49, p = 0.22).

Table 2 shows a significant difference between frequent and infrequent drivers on the question about thinking of the environment in terms of cardinal directions (F(1, 202) = 5.73, p < 0.05). Respondents who drive more reported to have significantly higher tendency of using cardinal directions to mark their environment as north, south, west, east than those with less driving experience. We did not find significant difference in other questions between the two groups of drivers.

Table 3. Mean and standard deviation of comparing frequent and infrequent passengers, lower value means more agreement with the statement. (*for significance p < 0.05, **for significance p < 0.01)
Full size table

For the experience of riding as passengers, we grouped all participants into 76 frequent passengers (37.3%) and 128 infrequent passengers (62.7%). Both group has a similar gender distribution, with 32.9% of frequent passengers being female, and 38.3% female respondents in the other group. However, significant difference was found in the mean age of the two: frequent passengers share a mean age of 32.36, while infrequent riders have the mean of 35.39, which is significantly older than respondents who ride in vehicles more often (F(1, 202) = 5.13, p < 0.05).

From Table 3, we found that frequent and infrequent riders showed significantly different opinions towards most of the questions. Respondents who ride more indicated to get lost easier in the neighborhoods around where they live than those with less riding experience (F(1, 202) = 33.43, p < 0.01), while they significantly agreed more on that they remember routes very well when being passengers (F(1, 202) = 6.65, p < 0.05). In addition, frequent passengers expressed significantly more agreement to the tendency of thinking the environment in cardinal directions (F(1, 202) = 17.42, p < 0.01), and the ability of finding novel shortcuts without the help of navigation apps (F(1, 202) = 5.31, p < 0.05). On the contrary, frequent riders also indicated significant agreement to the statements on finding it hard to find the way using only their mental maps of the environment (F(1, 202) = 17.65, p < 0.01), and their difficulty of pointing to where they live when imagining themselves standing at a location that is not near their homes (F(1, 202) = 4.75, p < 0.05).

Besides frequencies of driving and riding, we also performed one-way ANOVA on the effects of gender, the time span participants live in the current neighborhood, and their previous experience with navigation apps on the spatial survey knowledge participants indicated in the questionnaire. Similar to the widely-reported sex differences of spatial abilities in navigation studies [18], we found females reported significantly lower sense of directions than male participants (F(1, 202) = 4.604, p < 0.05). There is no significant effect on the spatial knowledge in terms of the time period respondents have lived in the neighborhood (F(3, 200) = 1.72, p = 0.16). But a highly significant effect of navigation apps was found in our study (F(4, 188) = 3.59, p < 0.01), where the Tukey post hoc analysis suggests a significant difference between participants who “always” use navigation and those “sometimes” utilize the service while driving (p < 0.05).

4.2 Two Factors Analysis

A two-way ANOVA was conducted to determine the effect of driving frequency and being passengers on the self-reported spatial knowledge. We found a statistically significant interaction between the frequency of being drivers and riding as passengers on the levels of spatial knowledge (F(1, 1) = 6.13, p < 0.05). Table 4 provides an overview of the reported spatial knowledge among the four compared groups. As presented in the table, participants who drive frequently without much experience riding in vehicles reported to have the highest spatial survey knowledge, whereas frequent drivers and frequent passengers indicated having comparably poor spatial knowledge to the respondents that neither drive nor take ground transportation.

Table 4. Mean and standard deviation of self-reported spatial knowledge, lower value indicate better knowledge
Full size table

Simple main effects analysis shows that frequent passengers reported to have significantly worse spatial knowledge than infrequent riders when they are all identified as frequent drivers (F(1, 113) = 4.64, p < 0.05), but no significant differences when they do not drive as much (F(1, 87) = 2.14, p = 0.15). For infrequent passengers, there is a significant difference in terms of the reported spatial survey knowledge between frequent drivers and infrequent drivers (F(1, 126) = 4.97, p < 0.05), but no significant difference was found for participants belonging to frequent passengers (F(1, 74) = 2.12, p = 0.15).

5 Discussion

5.1 General Insights

Our results suggest that in general, the effect of driving frequency on the spatial survey knowledge is dependent on the individual’s experience of riding as passengers. For people who drive often, their experience of being frequent passengers degrades their spatial survey knowledge about the environment, while among infrequent passengers, driving more significantly contributes to better survey knowledge for cognitive maps.

To answer RQ1, we found significant differences in certain aspects that account for spatial survey knowledge between frequent and non-frequent drivers, and among participants who differ in the frequency of riding inside vehicles. More specifically, frequent drivers have more tendency of thinking the environment in terms of cardinal directions than people who drive less often. A similar pattern was discovered by comparing frequent and non-frequent passengers, where the former are more likely to think the environment in north, south, west, and east. One possible explanation could be the variations in practice of wayfinding and navigation, where navigation tools present information in a north-up map [14]. Furthermore, frequent passengers stated their difficulty of finding their way relying solely on their mental map, and their likelihood of getting lost in the nearby neighborhoods is significantly higher than participants with less riding experience. This is consistent with the finding that it is harder for frequent riders to point to their homes at a far-away location, as their mental representation of the environment fails to provide adequate spatial information. Such information does not include route knowledge, as passengers reported to have higher ability of remembering routes when riding in vehicles. However, with poor mental maps, frequent riders could always find novel shortcuts without the aid of navigation apps. This could be due to a lack of relevant shortcutting experience, since passengers are usually passive in navigation and path planning. Thus, they may encounter significantly fewer situations that require finding novel shortcuts by themselves.

In the context of autonomous driving, our results provide a starting point to understand the possible consequences of adopting self-driving technologies in the long term. Our findings suggest that for people who drive frequently and do not to ride as passengers, they might experience a slight degradation in spatial survey knowledge after they ride with self-driving vehicles more as passengers and less as drivers. There are several explanations for this. One is the intuitive idea that passive exposure leads to poor spatial learning than active exploration [7]. As a result, passengers who passively interact with the environment are expected to receive less spatial knowledge if they instead drive and make navigation decisions by themselves. On the other hand, the impact of increased experience with becoming passengers in autonomous cars on human spatial knowledge for cognitive maps depends on individuals’ previous driving experience, according to the interaction effect we discovered in the analysis. Therefore, to further explore the relationships between cognitive maps and the usage of autonomous driving, it is important to take people’s previous experience of driving into account as well.

5.2 Importance of Cognitive Maps

Cognitive maps are considered as an individual’s mental representation of the environment. Although they differ across individuals, the key features of cognitive maps including the relative locations of landmarks and distances between places are useful in both route planning and wayfinding tasks. Beyond spatial cognition, cognitive maps could also benefit designing pleasant spaces and city planning [16]. Recent work on transport demand prediction [15] outlines another potential of cognitive maps as a tool to predict human mobility and traffic planning.

In the future age of autonomous vehicles, humans still need to preserve and update cognitive maps under various situations. During trip planning, potential destinations could be easily considered unreachable without proper knowledge in cognitive maps of the environment. It is also possible to save time and increase efficiency if shortcuts between locations are known in advance. On the other hand, vehicles are not the only mode of travel in everyday life. Even with autonomous vehicles, humans will likely continue use walking, cycling, and other means of transportation that require some level of navigation without full automation. During this process, cognitive maps are extremely helpful in their capacity of taking efficient detours around obstacles and identifying useful shortcuts [26]. There are also situations where navigation assistance fails that could happen at anytime during self-navigation, for example, when GPS signals are absent indoor, or wayfinding apps may crash. With a more complete cognitive map, travelers can feel less lost and have a better chance of finding their way with minimum effort.

5.3 Limitations

Due to the nature of our approach that deploys online questionnaires, our data is completely based on participants’ self-evaluation. Despite the effort of making the survey more understandable and attention checking through combining positively and negatively stated questions, it is possible that some participants did not fully understand the questions asked, or they failed to estimate their spatial knowledge in an objective manner.

Another limitation is considering the time and purpose of this study, we asked questions on scenarios where cognitive maps are applied, instead of actually testing the spatial survey knowledge of the participants using objective measures such as pointing or model-building tasks [29]. This can lead to participants overestimating their actual spatial knowledge, or underestimation could also happen. It is infeasible to request respondents performing specific tasks including pointing to a familiar location or drawing sketch maps, because of the diverse regions respondents of the online survey might live in. To further look into the question, we propose a controlled experiment using a driving simulator in the lab, which will be discussed in the next section.

6 Conclusion

In this paper, we investigate how the frequencies of driving and riding as vehicle passengers effect individuals’ spatial survey knowledge for cognitive maps through an online survey. The findings of this research show a significant interaction effect of driving and riding frequency on the reported spatial knowledge, suggesting that in order to understand the relationships between cognitive maps and the adoption of autonomous vehicles, we need to consider the previous driving experience in addition to the increased role of passengers in self-driving cars. One interesting finding is the subtle decrease of spatial knowledge when comparing frequent drivers without experience as riders with frequent riders who drive less often. Variations in different aspects of cognitive maps were identified between levels of driving frequency and how often participants ride inside vehicles.

The main motivation of this study is to bring up the issue: what long-term impact will autonomous driving bring to human spatial cognition, especially with regard to cognitive maps? We argue that cognitive maps are important to carry out day-to-day activities. By understanding the possible consequences, we will be more prepared to deal with potential changes at an earlier stage, as well as providing valuable resources to designing the interaction with future vehicles. We hope to provide preliminary insights through an online questionnaire, and discuss new research questions on the topic with researchers from multiple domains.

To address some of the limitations of this study, we proposed a controlled experiment in the lab to further look into the question, and we are currently in the process of recruiting participants. We use a driving simulator in a virtual environment to simulate driving at both level 0 and level 5 automation, and evaluate participants’ learned cognitive maps through pointing tasks, model building tasks, and map-sketching tasks. The purpose of the ongoing study is to provide empirical evidence on the differences in spatial survey knowledge individuals acquire of a novel environment after interactions with varying levels of driving automation. Future work include proposing driver-vehicle interfaces that assist with retaining the knowledge for cognitive maps in autonomous cars, and reimagining automobile-human interactions to increase spatial awareness of human passengers.