Towards Mindless Stress Regulation in Advanced Driver Assistance Systems: A Systematic Review

Background: Stress can frequently occur in the driving context. Its cognitive effects can be deleterious and lead to uncomfortable or risky situations. While stress detection in this context is well developed, regulation using dedicated advanced driver-assistance systems (ADAS) is still emergent. Objectives: This systematic review focuses on stress regulation strategies that can be qualified as “subtle” or “mindless”: the technology employed to perform regulation does not interfere with an ongoing task. The review goal is 2-fold: establishing the state of the art on such technological implementation in the driving context and identifying complementary technologies relying on subtle regulation that could be applied in driving. Methods: A systematic review was conducted using search operators previously identified through a concept analysis. The patents and scientific studies selected provide an overview of actual and potential mindless technology implementations. These are then analyzed from a scientific perspective. A classification of results was performed according to the different stages of emotion regulation proposed by the Gross model. Results: A total of 47 publications were retrieved, including 21 patents and 26 studies. Six of the studies investigated mindless stress regulation in the driving context. Patents implemented strategies mostly linked to attentional deployment, while studies tended to investigate response modulation strategies. Conclusions: This review allowed us to identify several ADAS relying on mindless computing technologies to reduce stress and better understand the underlying mechanisms allowing stress reduction. Further studies are necessary to better grasp the effect of mindless technologies on driving safety. However, we have established the feasibility of their implementation as ADAS and proposed directions for future research in this field.


INTRODUCTION
The recent years have seen the rise of advanced driver-assistance systems (ADAS) able to monitor the state of the driver in real time, including or even focusing exclusively on the affective dimension. Recent European projects are funded on this topic 1,2 . The next step would be to regulate this affective dimension (Zepf et al., 2019). Automotive industrials show a growing interest in this topic (see Braun et al., 2020 for a review). Figure 1 illustrates this tendency. The research concerning "emotion" in computer science, one relevant field for this review, is compared to the interest by other disciplines. As displayed in Figure 1, emotion regulation can be considered a new domain compared to the numerous results for emotion recognition. Among the different emotions, stress is arguably one of the most interesting to regulate while driving. Stress can be due to environmental stressors as well as personal factors, such as workrelated issues (Rowden et al., 2011). Moreover, stress can generate other emotions that impact driver performance, e.g., anger, (Shamoa-Nir and Koslowsky, 2010), or provoke distraction and compromise safety (Brake Direct Line, 2011). Selye (1983) defined stress as "A nonspecific response of the body to any demand" (p. 2). In driving, acute stress can lead to deleterious effects on performance. Of note, this state is different from chronic stress, which impacts people daily without requiring the presence of a specific stressor. Acute stress is time limited and emerges when stressors are present. The possible stressors can be very different [e.g., personal threat, time constraint (Klein, 1996), or cognitive workload (Hidalgo-Muñoz et al., 2019)].

Stress and Emotional Regulation
Stress can be physiologically, behaviorally, or psychologically self-regulated (Giannakakis et al., 2019). Under stress, hormones such as noradrenaline or cortisol are released to help the organism adapt to the stressful environment. In high levels of concentration, these hormones can decrease the activity in frontal brain regions (Dehais et al., 2020), regions strongly linked to executive functions and working memory (Motley, 2018), and which support attentional abilities (Dehais et al., 2019). These cognitive functions are involved in event anticipation, the inhibition of distracting stimuli, or mental flexibility (Miyake et al., 2000), which are essential to drive safely.
One of the most accepted models to explain emotional regulation (i.e., the ability to modulate one's own emotions) was proposed by Gross (2014). The author considers the temporal course of emotion emergence to explore five sequential stages where the regulation may occur. These stages are temporally divided depending on whether the emotion is developing or already present. These stages are the following: • situation selection: avoiding certain situations that could elicit stress; • situation modification: modifying the stressful situation by acting on stressors; • attentional deployment: redirecting one's attention away from the stressor; • cognitive change: re-evaluating the stressful situation (reappraisal); • response modulation: acting on experiential, behavioral, or physiological components of the stress response.

Stress While Driving: Emergence Model, Impacts, and Detection Methods
Driving requires a high level of executive functions, such as decision making, cognitive flexibility, and working memory (Mäntylä et al., 2009;Asimakopulos et al., 2012). Studies have pointed out that stress is commonly felt by drivers. For instance, a report based on a survey of 841 UK drivers (Brake Direct Line, 2011) show that 71% of drivers felt stressed and lost concentration due to stress. Another survey conducted in New Zealand reported that 55% of them reported driver anxiety (Taylor, 2018). To understand the origins and consequences of stress, Matthews (2002) proposed a transactional model of driver stress and fatigue. Note that, although this review is focused on stress, there are other factors that can impact driving performance and could be considered to regulate stress. Fatigue is an important one (Crawford, 1961). The transactional model presents stress as the product of the driver's evaluation of the environment, mediated by personal coping abilities. In this model, the term "transaction" refers to the relationship between person and environment. This relationship is modulated through cognitive stress processes, linkable to appraisal theories (Scherer, 1999), that allow the person to evaluate the situation. This evaluation is supported by cognitive functions linked to self-consciousness, memory, motivation, reasoning, and attention. The situational demand can be evaluated as either exceeding or being within the limits of the perceived resources of the driver. The outcome of this transaction is expressed in terms of subjective feelings like stress or fatigue and in terms of road behaviors. This transaction is conceived in a dynamic perspective, meaning that a re-evaluation that considers past outcomes can occur, closing the loop. According to appraisal theory (Scherer, 1999), the evaluation is conducted on several dimensions: pertinence (linked to implication), controllability (linked to possible actions), and significance (linked to beliefs) of the situation for the driver.
Situational factors in driving (strains) can induce cognitive stress. These factors include (but are not limited to) traffic congestion, time constraints, and weather (Rowden et al., 2011). Furthermore, the behavior of other drivers can also be an important stressor (Rasmussen et al., 2000).
Concomitant with the situation, several personal factors, e.g., health level, personality traits, or attitudes toward driving, can influence the perception of one's resources. For instance, neuroticism or trait anxiety can increase the negative evaluation of the situation, thus leading to distress . Furthermore, the occurrence of work-related stress or other life events such as being in the process of a divorce has been shown to be positively correlated with the feeling of stress while driving (Rowden et al., 2011).
As mentioned, stress can reduce cortical activity (Qin et al., 2019), impair executive functions, or even induce attentional tunneling (Dehais et al., 2020). The disruption of these functions can lead to a decrease in situational awareness or to risky behaviors in driving (Healey and Picard, 2005). The question of stress detection and regulation in the car is therefore crucial to ensure safety. However, stress detection should be subtle to avoid disrupting the driving process. Regarding physiology, various stress indicators derived from autonomic activity are well known: heart rate (HR), breathing rate (BR), or their variability (HRV and BRV, respectively) (Taelman et al., 2009;Vlemincx et al., 2012), or electrodermal activity (EDA) (Setz et al., 2009), among others (see Giannakakis et al., 2019 for a review). Table 1 shows several  illustrative examples. Driving behavior measurements can also be used to detect stress: acceleration, speed patterns (Rastgoo et al., 2019), steering wheel usage (Paredes et al., 2018a), etc. Additionally, the methods can be based on environmental measures (context detection) to determine whether a situation is likely to elicit stress. For example, Vhaduri et al. (2014) managed to detect a driver's stress using GPS traces from a smartphone. In the same vein, Dobbins and Fairclough (2018) used contextual data (photographs of the traffic environment) to improve their stress detection algorithm.
All these stress indicators can be combined to achieve better detection (Abou Elassad et al., 2020).
Although ADAS and driving automatization could reduce the prevalence of stressful and risky situations through the management of some situational stressors , this is not a full-fledged solution, as it can lead to new forms of stress. This can be due to the attentional solicitation required by their supervision or due to the reduction of situational awareness (Warm et al., 2008). Actually, in some cases, ADAS can act as stressors on board by increasing the mental workload to handle them, increasing the complexity of the cockpit, or through false alarms (Brookhuis et al., 2001). Besides, in autonomous vehicles, drivers can be engaged in nondriving-related tasks and thoughts (Pfleging et al., 2016), which can elicit internal acute stress in various forms. These elements can impact takeover, a procedure that we can expect in near-future automated vehicles, thus inducing new safety challenges (SAE International, 2018).
Stress regulation methods usually imply cognitive therapies, meditation, or respiratory exercises (Baer, 2003), which could turn into cognitive distraction. An interesting approach to design ADAS aiming to regulate stress can lie in the field of "mindless computing, " defined by Adams et al. (2015) as "a mobile or ubiquitous, persuasive technology designed to subtly influence the behavior of the user without requiring their conscious awareness." Therefore, mindless technologies act on an individual without disturbing him from a main task (i.e., in an unconscious way). As previously illustrated, mindless stress detection can be accomplished using various subtle indicators (e.g., driving context or biosignal remote measurements). Although much less explored, mindless technologies could also be applied for stress regulation, and it is a promising emergent field.

Aim of the Present Review
The goal of this systematic review is to establish the state of the art in the domain of subtle stress regulation in driving. The works selected will allow better understanding on how this regulation can be technically accomplished and its scientific bases. A second objective is to explore new directions for future research on in-car regulation techniques based on mindless computing. Hence, not only patents and research work investigating it in the driving context (actual patents and actual papers) will be considered but also research works that could be applied to driving (potential papers).

Guidelines and recommendations contained in the Preferred
Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) statement (Liberati et al., 2009) have been followed.

Eligibility Criteria
This review includes articles and patents dealing with mindless computing technologies that aim to regulate acute stress. With this in mind, our eligibility criteria were inspired by Adams et al. (2015) regarding the design considerations of mindless computing technologies: • Reflexive technologies: the technologies have to leverage prompt and automated cognitive processes. • Triggering technologies: the desired change of behavior should be achieved without depending on user motivation. Personalization of the trigger per user is possible. • Parallel technologies: the technology has to be subtle. The user does not have to explicitly attend to it and should be able to pursue another activity without being disturbed by the technology.
We also excluded publications that did not consider stress regulation and focused on other emotions (e.g., anger), or that would use invasive techniques (e.g., use of needles/electrostimulation). Furthermore, we excluded publications that request explicit training from the user to be efficient. This means that studies that would be based on prior implicit learning have been included. For the first part of this work (actual applications developed for driving), patents were favored since they could include the most recent and concrete technical development regarding the automotive industry. We also sought experimental studies that investigate stress regulation in a driving context (simulated or naturalistic).
Regarding the second part of our work (potential applications for driving), we based our research on experimental studies and excluded opinion or review papers. We considered studies that focused on adult participants of any age and gender. Studies investigating specific populations as a target to apply the regulation technique on were excluded.

Information Sources
We used several electronic databases to conduct our search. For the actual part, we located patents using Espacenet, 3 a database inventorying worldwide patent. For the papers (actual and potential), we used five electronic databases: Scopus, Web of Science, Association for Computing Machinery Digital Library (ACMDL), PsycInfo, and IEEExplore. We limited our research to English papers and to patents for which an English translation was available on Espacenet. We restricted this search to publication dates from the past decade (2010-2020), since a growing trend of interest toward our research topic seems to appear in this range (Figure 1). When possible, we also limited our search to research fields that would match our review scope to limit the number of false positives.

Search
To conduct a systematic search, we first identified concepts linked to our research topic in both parts of the review. The concepts included for the potential part were the following: "regulation, " "stress, " "mindless, " and "technology." For the concepts in the actual part, the "driving" concept was added to select patents that were specific to the driving context. We chose to exclude the concepts linked to "mindless computing" and "technology" from the patent equation search due to limitations of the search engine and because we stated that emotional regulation while driving manually would, in most cases, be a mindless technique due to the cognitive and attentional requirements of the manual driving context. The screening phases allowed us to include patents that matched this criterion. We excluded patents that focused on nonmindless techniques but included patents that referred to several techniques including mindless and nonmindless techniques.
We then defined various keywords that could be linked to each concept and which were expected to appear in the titles or the abstracts of studies regarding mindless stress regulation. Due to a large number of keywords, we pretested various research equations that could apply to each of the requirements and limitations of the selected database to obtain results that would neither be too restrictive nor too inclusive. Therefore, two main research equations were retained.
For the patents search: [(driver) AND (regulat * OR recov * OR relax * OR modulat * OR manag * OR reduc * ) AND (stress OR anxiety OR emotion)].
Given the search restriction for Espacenet, we chose to use the search token " * " to expand our research to all keywords included in the concepts of "regulation" in the equation for the Potential part. For

Study and Patent Selection
The exclusion process applied strict exclusion criteria throughout three stages: first, we excluded publications based on their titles, then based on their abstracts, and finally based on full texts, when available. Two independent reviewers conducted each step and solved conflicts for the first two steps. For the final selection based on full texts, a third independent reviewer resolved remaining conflicts. The full text of patents was translated when necessary using the tool provided by Espacenet "patent translate." Due to the high variability in the terminology and the diversity of keywords used in the search, a large number of false-positive studies appeared during the review process (0.95% of initial results were included in the first set of papers). We excluded duplicates and authorless publications. Figures 3, 4 detail the PRISMA flowchart of the selection process for the two types of documents (patents and studies) included in the review.

Additional Sources and Selection
To complement the search for scientific publications, additional studies were identified through a backward and forward snowballing approach (Wohlin, 2014) detailed in Figure 4.

Data Collection Process
We classified relevant information from selected full texts in two spreadsheets. The spreadsheets were divided according to publication type: patents that propose techniques directly applicable to a driving context (actual patent, Table 2) and studies that investigate stress regulation technologies in a driving context (actual papers, Table 3) or a fundamental context (potential papers, Table 3).
In the patent spreadsheet, we classified information according to the following variables: (1) regulation strategy identified, (2) patent number, (3) applicant of the patent and year of publication, (4) type of emotion the patent focuses on, (5) main goal of the patent, (6) technique employed for emotion detection if applicable, and (7) regulation technique implemented. In the study spreadsheet, we classified information according to the variables listed below: (1) regulation strategy identified; (2) authors and year of the study; (3) title of the study; (4) presence of stress induction and, if so, the task used; (5) main technique evaluated; (6) experimental conditions; (7) characteristics of the sample, including sample size, average age, or range; (8) metrics to evaluate the efficiency of the technique employed including physiological and subjective measurements; and (9) main outcomes of the technique presented.
No quality assessment, in terms of statistical power, of the contributions was performed.

Classification Process of Regulation Strategies Found
A classification chart (Figure 2) was designed based on the explanations provided by Gross (2014), and each included record went through this chart to determine at what stage it regulates stress is effectively. It was first considered that some techniques can be linked to several regulation strategies: for instance, a technique that uses music to regulate stress could be classified as situation modification (as the global context will be changed by adding the music), but music will only be efficient to regulate once the stage of attentional deployment is reached. In order to draw clear boundaries, we have chosen to include in each strategy the techniques that are effective only once the stage of regulation where the technique is at its maximal efficiency has been reached. However, selected patents often propose several techniques based on several regulation strategies.
For instance, using this chart, a study proposing a regulation strategy relying on biofeedback technology could be classified this way: The technology does not allow avoiding the situation, directly modifying the stressor, nor changing the perception and subsequent evaluation of the situation FIGURE 4 | Adapted preferred reporting items for systematic reviews and meta-analyses (PRISMA) flowchart for study selection. but is indeed based on a change of the physiological stress response.

Patent Selection
The search performed on 03/07/2020 on EspaceNet provided a total of 490 unique records. Of these, 364 were excluded based on their title, mostly because they were not linked to the review field. Among the remaining 126 patents, 53 additional records were excluded after reviewing the abstract. Of the remaining 73 patents, 52 additional records were excluded based on their full description.
Main reasons for full-text exclusion were the following: not being linked to stress regulation (n = 21), not being related to the field (n = 13), not being mindless (n = 7), no full text in English found (n = 5), and being on detection only (n = 6).
The Cohen's kappa score for agreement between the two reviewers for the full text screening phase was κ = 0.66. According to Landis and Koch (1977), this is a strong agreement value.  To reduce the noise emitted by the tires in order to reduce it and prevent stress of the driver due to the noise, and to obtain a frequency close to alpha range of brain waves -A tire design that generates human brain waves by using the monaural beat phenomenon. The offset frequency is formed in the alpha wave band (    Finally, a total of 21 patents were identified for inclusion in the review (for the full patent list, see the Appendix).

Characteristics of Included Patents
It was not possible to determine from the patents whether participants were involved to test the techniques proposed.
Several modalities are employed for regulation. They include audio display such as music or sounds (n = 2), visual display such as lights (n = 5), haptics such as acupressure (n = 1), odors (n = 1), or others modalities such as vehicle control or ambient air changes (n = 2). Several patents employ a combination of modalities, such as audio, visual, and other (n = 2); audio, haptic, and visual (n = 1); visual, audio, and odors (n = 1); audio and other (n = 3); audio and haptic (n = 1); and audio and visual (n = 2). Therefore, strategies identified in patents were situation selection (n = 2), situation modification (n = 3), attentional deployment (n = 6), and response modulation (n = 2). None of the patents are only using a cognitive change-based regulation strategy. Several patents propose at least two regulation strategies, including situation modification and attentional deployment (n = 1); attentional deployment and response modulation (n = 1); attentional deployment and cognitive change (n = 2); situation modification and response modulation (n = 1); situation modification, attentional deployment, and cognitive change (n = 1); situation modification, attentional deployment, and response modulation (n = 1); and situation selection, situation modification, attentional deployment, and cognitive change (n = 1). Figure 6 is an upset plot (Lex et al., 2014) displaying the matrix layout of the combination of regulation strategies in the patents. Each row of the matrix represents a set, and each column displays the intersection between sets of regulation strategies when they exist (lines between the dot). The histogram above the matrix shows the number of patents for each intersection, while the histogram next to the matrix exposes the number of patents that include each regulation strategy.
All of the patents but four (see Table 2) included the possibility of measuring the state of the driver (through various biometrics) or the driving metrics and context.

Study Selection
The combined search in Scopus, Web of Science, ACMDL, PsycInfo, and IEEExplore, performed on 03/07/2020, provided a total of 1,053 unique records. Of these, 974 were excluded because the topic studied was not relevant for the review topic (e.g., Optics or Engineering). Among the remaining 79 documents, 55 additional records were excluded after reviewing the abstract due to being unrelated to stress, or not being mindless; 3 more records were excluded because the full text could not be found. Of the remaining 21 articles, 11 were excluded based on the full paper.
The reasons for exclusion were the following: not being linked to stress regulation (n = 5), being a conceptual paper or no  concrete technique tested (n = 2), not being linked to the field of the review (n = 2), using an invasive technique such as electrostimulation (n = 1), not being mindless (n = 1), and application on a specific population (children, n = 1). One excluded article met more than one exclusion criteria.
The Cohen's kappa score for agreement between the two reviewers for the full text screening phase was κ = 0.62. According to Landis and Koch (1977), this is a strong agreement value.
Due to the low number of included studies, the 10 remaining studies were used as a starter set to apply a backward and forward snowballing procedure, according to guidelines provided by Wholin (2014). Backward snowballing refers to review the references contained in the start set papers, while forward snowballing refers to review records citing the start set papers. Each iteration uses the included studies from the previous iteration as start set. The detail of the three iterations is provided in Figure 4. At the end of the snowballing process, 16 additional papers were included in the review, bringing the total number of included studies to 26. These additional studies met the same criteria as the studies from the initial set.
Five additional papers are international collaborations from: Netherlands and Italy; Canada, China, France, and Sweden; USA and Qatar; USA, Norway, and Germany; and USA and Germany. Papers could be classified as actual papers (n = 6) when the technology was tested in a driving environment; otherwise, they were classified as potential papers (n = 20). The cumulative number of participants in the studies is about 603, including 226 men-the exact number of participants cannot be known since one study does not provide this information. Similarly, five studies do not indicate the number of men involved among the participants.
The population type is mostly composed of young adults that are university students (n = 6 studies), university employees (n = 3 studies), general population (n = 3 studies), drivers in the studies that are conducted in a driving context (n = 6 studies), participants that are subject to anxiety (n = 1 study), or participants with disabilities (n = 1 study) including impaired vision, impaired audition, and autism.
Several modalities are employed in studies to display the intervention on the participants. Modalities include audio display (n = 10), visual display (n = 5), haptics (n = 8), and odors (n = 1). Some studies employ a combination of modalities, such as haptic and visual (n = 1) and haptic and sound (n = 1).
We employed the same classification chart used for patents to determine the regulation strategy employed by studies (Figure 2). Strategies identified in studies are the following: attentional deployment (n = 4) and response modulation (n = 21). None of the studies are using situation selection, situation modification, or cognitive change. One study tested two regulation strategies, including attentional deployment and response modulation. Figure 7 represents the number of studies for each regulation strategy and overlaps.
Data collected in the studies include physiological metrics (n = 20), behavioral metrics (n = 7), and subjective metrics (n = 23). Details of this distribution can be found in Table 3. Figure 8 gives an overview of overlaps for all papers.
Physiological metrics employed are mostly linked to cardiac activity (n = 18), respiratory activity (n = 8), and electrodermal activity (n = 4). One study employs electroencephalography together with cardiac activity. Among the behavioral metrics, two studies employ measurement of performance in the stressful task, while five of the six driving studies ("actual papers") employ measurement of the driving performance using metrics such as acceleration, steering wheel, and lane departure.

DISCUSSION
The Gross emotion regulation model (2014) reveals itself as a pertinent reading grid to classify the selected publications. Therefore, the discussion has been divided into five sections linked to the different Gross emotion regulation stages. Of note, most records do not explicitly provide the regulation stage involved, so a classification was applied as described in Methods. In addition, some publications can fit in more than one section as they evaluate several technologies (see Figures 6, 7). Each section will introduce the corresponding stage; then, the discussion will explain the methods found (Tables 2, 3) involving this stage for: actual patents, thus giving an idea of applications in the automotive industry; actual papers reporting strategies in a driving situation (naturalistic or simulated); and potential papers consisting in scientific contributions tested outside the driving context but which could be suitable due to their intrinsic mindless characteristics and their feasibility in the car.

Situation Selection
This strategy is the first possible stage to regulate stress. It refers to the avoidance of a stressful situation by choosing an alternative situation less likely to elicit stress. Therefore, acting on this component is tantamount to anticipating and avoiding stressors. The application of the strategy is here specific to the driving context.

Actual Patents
The main solution consists in selecting a road that is likely to be less stressful. CN Patent No. 102874259A (2013) proposes to use the GPS to suggest the closest locations to rest. Likewise, JP Patent No. 2011027441A (2011) uses context recognition to determine if there is a lot of traffic surrounding the user's car. If so, the GPS picks a road with less traffic. Similarly, KR Patent No. 20190103521A (2019) provides an optimal road according to the automatically detected state of the driver or when the driver triggers the system. Indeed, the road can be parametrized in advance by the driver himself. This is in line with Gross's (2015) observation that it would be hard to anticipate "how one will feel in different situations." Indeed, this suggests that personalization is necessary according to each user's predisposition to feel distressed in a particular situation. Therefore, to allow the user to (1) select in advance a potentially relaxing road and (2) to trigger the guidance seems particularly pertinent.

Situation Selection: General Discussion
Current apps such as Google Maps (which may provide the road with least traffic) can share the same approach to the patents presented below. Nonetheless, an essential difference exists: the device is aware of the state of the user or the context directly surrounding the car and the possible repercussions on drivers' well-being. Among the six actual papers that have tested stress regulation within the driving context, none of them were identified as exploiting the situation selection strategy. Studies mainly focus on stress response given a particular situation but do not propose to "avoid a situation." Indeed, while driving, it is difficult to find other solutions that fall within situation selection (SS) other than choosing the road with least traffic since other stress sources like weather are not possible to avoid.

Situation Modification
Situation modification refers to altering the stressful situation, without avoiding it. Included in this section are contributions that proposed to act either on the stressful situations or on the stressors. This means that techniques implying a global modification of the context features not related to stressors (e.g., the techniques modifying the cockpit environment by an ambient display) were not included.

Actual Patents
Several patents propose to modify the driving context via the modification of the vehicle's controls. Indeed, CN Patent No. 102874259A (2013), CN Patent No. 109572705A, 2019), and CN Patent No. 209186732U (2019) propose to limit the speed of the vehicle and to control the steering wheel. FR Patent No. 2998159A1 (2012) suggests connecting conventional ADAS (e.g., park assistance, speed limitation) with stress detection. In the autonomous context, JP Patent No. 2016052881A (2016) proposes to adapt the behavior of the vehicle if stress is detected by adjusting the modality of the takeover request that could eventually generate stress.
Three patents are based on very specific driving contexts. KR Patent No. 20110034326A (2011) proposes to modify the structure of the tires to attenuate the noise and to produce monaural beats (this patent will be explained in detail in the response modulation section). JP Patent No. 2018020683A (2018) presents a reduction in stress while crossing a tunnel. The patent detects the side walls and illuminates them. RO Patent No. 127332A2 (2012) proposes to attenuate the dazzlement from incoming vehicles' lights during nighttime.

Situation Modification: General Discussion
Given its prevalence in patents, a question that could arise is: Would the vehicle control be an efficient way to regulate stress? On the one hand, the situation can be modified by limiting or reducing speed, thus becoming less stressful. On the other hand, this countermeasure could entail more stress due to the lack of the feeling of control (Kim and Diamond, 2002). Patents proposing to rely on conventional ADAS to manage the situation when necessary can be linked to some studies suggesting that ADAS such as active park assist, active city stop, or adaptive cruise control could be an efficient way to regulate stress while driving (Chung et al., 2019 for a review).
Situation modification techniques, similarly to situation selection, seem specific to the driving context since it would be hard to interact with personal factors. None of the potential papers nor actual papers investigated situation modification, as its implementation would be very specific to the investigated context.

Attentional Deployment
Attentional deployment refers to an attentional redirection of the user (toward either internal or external focus). The regulation is achieved by diverting the driver from stressors. This attentional redirection can be toward a neutral element or toward an element that can remind relaxing or pleasant memories. To guarantee the mindless approach, only contributions relying on peripheral attention (Bakker and Niemantsverdriet, 2016) were considered. (2018) present classification systems to play the driver's "favorite music" or automatically selected music according to emotional characteristics.

Actual Patents
Breaking from the mold, CN Patent No. 108109228A (2018) proposes to generate an odor to regulate a stressed state. This odor is triggered by the force at which the driver closes the door. CN Patent No. 108628205A (2018) also mentioned odors as an interesting perspective to regulate emotions but does not focus on this technique.

Actual Papers
Exploring regulation in a driving context, Hu et al. (2015) uses a music recommendation system to elicit relaxation. Similarly to some patents presented above, music tracks are classified according to the emotional valence and are delivered to enhance driver's mood. Their results show a reduction in fatigue and negative moods detected through an algorithm.

Potential Papers
Daher et al. (2020) evaluated the efficiency of ambient blue light to reduce stress induced by a mathematical task. The participants reported lower stress levels under this light. While also using lights, Ansems et al. (2011) introduce a photo frame that can light up when distress is detected. The idea is to elicit pleasant or relaxing memories by directing the attention toward a photograph that has significant value. Williams et al. (2015) introduce a wearable scarf that can actuate (vibration, cooling, weight) according to the user emotions. Amores and Maes (2017) tested various odors to elicit relaxation. The odors used were tea tree, peppermint, and rose fragrances and were delivered using a necklace containing the scents that could be triggered manually (on user's request) or automatically (based on biosignals).

Attentional Deployment: General Discussion
The technologies presented in this section should be compatible with driving and be efficient without requiring focused attention. Therefore, the methods are mainly linked to music recommendations and ambient light. Music is well known to reduce stress and can have a beneficial impact on both psychological and physiological aspects (de Witte et al., 2020). Thus, to provide a recommendation system adapted to the driver's mood seems to be a suitable technique. Concerning the use of ambient lights, it has been shown that colors and light can have various impacts on human psychology (Elliot, 2015), although it seems unclear whether specific colors have a specific impact or if it depends on personal factors (Johnson and Toffanin, 2012).
Of note, in the case of complete autonomous vehicles, attentional deployment could present more possibilities, since attention would not have to be always centered on the driving environment (Pfleging et al., 2016). For instance, the technique based on the use of an enlightened photo frame to attract attention (Daher et al., 2020) could be taken further, with for instance multiple photographs or even videos that elicit pleasant memories. Attentional deployment technologies are often complemented with another, such as biofeedback technologies.

Response Modulation
Response modulation is associated with altering the experiential, physiological, or behavioral stress response of the organism. Publications mainly altering the physiological component of user's response were included. This alteration can be done in a passive or active way. The former mostly relies on the deliverance of truthful biofeedback reflecting the physiological state of the user, while the latter tries to obtain a so-called "entrainment effect" to actively align the user response to a specific state (e.g., to a specific breathing or cardiac rhythm) either through guidance, through false biofeedback, or using another alteration. KR Patent No. 101601957B1 (2016) proposes to use monaural beats emitted through speakers. In a similar manner, KR Patent No. 20110034326A (2011) proposes to generate a monaural beat from modification of the tires' structure. Both patents try to generate a frequency close to the alpha range of brain waves to induce a relaxed state (Chaieb et al., 2017). KR Patent No. 20140080727A (2014) proposes to modify the ambient air in the vehicle using the air conditioner and windows. In an interesting way, JP Patent No. 2019194046A (2019) suggests modulating the physiological response by increasing the oxygen and hydrogen levels within the car to induce relaxation. A more diverting technique linked to breath regulation proposes to visually prompt respiratory exercises (FR Patent No. 2998159A1, 2012). MacLean et al. (2013) displayed biofeedback on the wrist of the driver, through a device called "Moodwings." This involves a butterfly that moves its wings according to the stress level, determined via electrodermal activity. This technique is original and well-designed, but the results showed that the participant felt more stressed when the system was activated, although driving performance was improved. Authors explain these results by lack of training. Following the same line, but using "musical biofeedback, " Bhandari et al. (2015) modified a music track by adding white noise varying in intensity depending on the breath rhythm of the driver. The physiological and subjective results showed that their arousal level was significantly lower. Zepf et al. (2020) used a different auditory modality to regulate driver's emotions. In their study, the voice of the "empathic GPS" guidance system was modified to decrease the arousal of the drivers, determined from their electrodermal response. The voice congruent to the state of the driver (in the sense that if the user was stressed, the voice was calmer) was subjectively associated with higher pleasantness, but no difference for stress was found. Such "empathic GPS" can be considered as a kind of biofeedback as the voice adaptation followed EDA.

Actual Papers
Using a haptic modality, an entrainment effect was induced by Paredes et al. (2018b) and Balters et al. (2020) who prompted breathing instructions by using haptic guidance embedded in the seat. Participants of both studies rated this method as subtle and calming, and a significant reduction in breathing rate and augmentation in HRV was observed.

Potential Papers
Lopes and Campos (2019) used a scarf that delivered a sound to induce relaxation. This sound was previously conditioned together with respiratory exercises to induce a relaxation state. Although this method can be linked to cognitive behavioral therapies, it presents the disadvantage of requiring a previous conditioning phase to be efficient. This method was rated as subtle, although it was more relaxing during the conditioning phase. Kim et al. (2018) applied a modulation of music according to the participant's HR. In their study, the music first matched participants HR, and then, a reduction was applied. In the same manner, but using a periodic sound linked to HR, Cuijpers et al. (2019) tried to induce an entrainment effect to regulate the physiological response. Similarly, Bergstrom et al. (2014) modified music tempo according to HR.
Focusing on respiration, Harris et al. (2014) modified music by adding white noise that varied according to participants' BR. Leslie et al. (2019) also tested various music modifications depending on the respiratory signal. Sato and Moriya (2019) used music by controlling the synchronicity with acoustical timing (phrases, pitch change). These results suggest that the degree of synchronization of the music with the physiological response can induce effective changes in the stress response. Fedotchev et al. (2017) proposed an alternative way to modify music following biosignals by using electroencephalogram (EEG). The EEG signal is converted into a flute-like sound that varies in terms of pitch and intensity. The experimentation tests the effect of adding 1 Hz rhythm to this signal. Even though this method could be uncomfortable to apply in a car due to the limits of EEG systems, it presents interesting results, such as a significant increase in well-being, as long as increased brain alpha and beta waves. Williams et al. (2015) used haptic modality to deliver, by means of a scarf, biofeedback to the user in the form of vibrations. Disabled participants found it especially useful. Using also haptic modality, Costa et al. (2016) delivered false biofeedback in the form of vibrations on the participant's wrist. The authors introduced a deceiving condition where the vibration's frequency was always at 60 bpm (lower than the real HR). The participants felt lower levels of anxiety and found the technique as subtle. They replicated their study (Costa et al., 2019) including a condition with 30% reduced HR biofeedback, which induced an HRV decrease, improved performance, and reduced anxiety. A study using similar wrist-worn (Azevedo et al., 2017) found a consistent pattern of results with a reduced biofeedback, since the arousal level (measured by electrodermal activity) and selfreported anxiety were lower. However, a replication study (van der Velden and Lakens, 2020) failed to replicate this effect. Using a similar device delivering a heartbeat-like vibration at 120 or 60 bpm, Choi and Ishii (2020) showed that the heart returned faster to a resting state after physical exercise when 60 bpm vibrations were delivered. Zhou et al. (2020) tested the effect of HR biofeedback using a vibration little box that the participant had to hold. Their results showed an increase in HRV and a reduction in HR. Ban et al. (2018) employed respiratory biofeedback using a cushion placed on the participant, which actuated at 7 or 15 bpm. They obtained an entrainment effect on BR.
Using visual modality, Hamon et al. (2018) designed a biofeedback flower that could adjust its petal colors according to HR and BR. Their preliminary results showed that the device failed to act as an ambient device and did not provide any significant results. Similarly, Yu et al. (2018) investigated the effect of a biofeedback ambient light changing according to the user's HRV. The ambient change was either toward cold hue (shades of blue) or warm hue (shades of yellow) light. Their results showed a better effect of the cold hue light to modulate the user's HRV toward a relaxed state. Cold light was rated as more relaxing and less distracting than the warm light.

Response Modulation: General Discussion
Response modulation is a promising strategy that is not much proposed in patents. There are many ways to modulate physiological response.
Regarding biofeedback, there is a strong link between emotions and body perception. Damasio (2006) theorized the somatic markers' hypothesis, which postulates that patterns of physiological activations linked to a particular emotion are integrated in a cortical level and impact cognitive processes. The degree of awareness of one's bodily processes is referred to as "interoceptive awareness" or "interoceptive sensitivity." Damasio's hypothesis suggests that people with high interoceptive awareness could use this bodily information as a guidance for cognition and for emotion regulation. Indeed, studies points out that this interoceptive awareness plays a major role in regulating emotions due to the feedback loop existing between the bodily information and the emotion processing (Füstös et al., 2013). Biofeedback is precisely a technology that acts on interoceptive awareness by enhancing it (Ceunen et al., 2016), thus facilitating emotion regulation (Schoeller et al., 2019).
Entrainment seems to be a promising method. Entrainment can be defined as "the process by which independent rhythmical systems interact with each other" (Clayton, 2012). Using entrainment to elicit a particular emotion is a method well developed in the film industry through music, especially in horror movies (Winters, 2008). Several modalities can be used and physiological components be targeted to obtain an entrainment effect. Either way, studies investigated the entrainment effect mostly in the domain of music (van der Velden and Lakens, 2020), as this modality seems particularly efficient. Bergstrom et al. (2014) hypothesized that music presents the advantage to continuously draw some attention, while other modalities such as sonification does not present particular significance and, therefore, could be easily ignored by the brain. In addition, the music can by itself elicit relaxation or positive emotions.

General Discussion
Next, we would like to give a critical view and remark on some relevant aspects of the selected contributions. First, the proportion of each regulation strategy differs substantially between patents and actual or potential papers. Even when most experimental studies focus on response modulation strategies, these strategies are overlooked in patents (Figures 6, 7), which tend to favor attentional deployment based. This fact can be a bit surprising since we would assume that patent strategies have been tested previously in experimental studies to guarantee their effectiveness. However, no recent papers were found on the same applications. Most of the actual papers showed the feasibility of their implementation in a controlled laboratory driving task. Arguably, the next step would be further research using ecological settings. This would allow better insight into the real impact of this method, not only on emotion regulation but also on driver distraction before proposing prototypes.
Usually, it is difficult to assess the validity of technologies described in patents, as they are not peer reviewed and are only meant to illustrate the technical feasibility of the proposed implementation. Moreover, a lack of theoretical fundamentals could be observed in some patents and in particular concerning definitions of the emotions used. This could make some propositions very expectative and original but also groundless.
While we considered that mindless attentional deployment is feasible using techniques relying on peripheral attention (such as music or lights), our sample provided limited evidence of their real effectiveness. On the other hand, several techniques that are attentionally demanding (based on cognitive change) could fit in the context of autonomous driving. Indeed, in autonomous vehicles, usual driving activity is replaced by a supervision task, which may change the stress regulation strategies involved. Either way, some original examples can be found in the literature. For instance, Paredes and Chan (2011) proposed to send an SMS message to alert the closest contacts in a user's social network if stress is detected.
When observed as a globality, some of these results may seem contradictory, and researchers on this domain are still reluctant to apply these techniques. For instance, MacLean et al. (2013) found an increase in stress when using biofeedback, showing that interoceptive awareness could also elicit further stress. Besides, most of the studies reported can be qualified as pilot studies, with a restricted number of participants and, sometimes, subjective measurements to evaluate efficiency (see Figure 8), leading to various biases, including social desirability.
Regarding sensorial modalities, haptic modalities were used mostly in studies (to induce response modulation), while patents preferred audio or visual, where more scientific evidence about their effect exists. Touch is an interesting avenue to investigate in the field of emotion regulation, as it can be implemented in various forms (MacLean, 2009). Furthermore, recent studies demonstrate the positive effects of haptic stimulation on wellbeing (McDaniel and Panchanathan, 2020). As for music, patents mostly use it for attentional deployment, while potential papers mostly use it for response modulation. With this aim in mind, music attributes are modified according to the state of the user, to perform biofeedback, or to generate an entrainment effect.
For visual purposes, based on the present review, Figure 9A details the possible implementations of the mindless technologies tested in the studies, whereas Figure 9B displays the techniques presented in the patents. According to these figures, and considering that patents tend to implement multiple systems at once, one could wonder whether an accumulation of implementation would always be safe or disruptive. A "hysteresis effect" disrupting the driver when using multiple mindless strategies could appear. It would be relevant to investigate this point in future studies, in addition to personalization tailored to each participant, knowing that stress emergence greatly varies across individuals.

LIMITATIONS
As any systematic review, the present work has tackled with the divergences in indexation of the studies. This issue is even more challenging in emerging fields, such as mindless computing (as defined by Adams et al., 2015), where theoretical or technological bases are not completely framed and an important list of keywords have to be used. Indeed, to be sure that few suitable papers fell out of the research, an important number of keywords was considered from the beginning, leading to an initial set of papers consisting of 1,053 results, from which <1% remained. The numerous keywords used have homonyms in very diverse topics (e.g., cellular biology or material science). This outcome can justify the use of a snowballing search strategy to guarantee an exhaustive review on the topic.
An original point of the present review is the inclusion of patents. They can present the current technologies that are being developed in the automotive industry. However, several difficulties were encountered. Most of the patent description versions in English have been translated by an automatic tool provided by Espacenet, making some definitions incomplete or hard to understand properly. Furthermore, this literature is not necessarily strict in definitions of emotions, making it sometimes difficult to disentangle between the actual emotions regulated in patents. Finally, experimental results are scarcely presented in patents, and the main outcomes in terms of efficiency or pertinence are difficult to quantify and compare to classical experimental protocols. Despite such difficulties, we encourage the inclusion of this type of contributions in reviews dealing with driving, since they present modern systems that will be incorporated with ADAS.
Concerning cognitive change, we chose to put this regulation stage aside in the discussion. We considered that the techniques collected in the present review based on this regulation stage were not mindless. As the results showed, several patents (n = 4) proposed this strategy, among others. Cognitive change is linked to modifying the appraisal of the situation (i.e., the way that an individual evaluates the situation). Proposed techniques suggest to prompt the driver to regulate his feelings or to drive safely using various mediums including vocal alarms (CN Patent No. 109572705A, 2019;CN Patent No. 102874259A, 2013), informative lights (CN Patent No. 110393540A, 2019), or a combination of modalities such as haptics, visual displays, and sounds (CN Patent No. 110403617A, 2019). We chose to put this regulation stage aside in the discussion. Furthermore, to prompt the user to remain calm or drive safely could, in some cases, induce even more stress (Silvia, 2002).
Regarding subjective data assessing the intervention, it is suitable to collect participant insights, feelings, and attitudes toward the intervention. However, we have to be cautious while concluding, since individual differences in populations can bias some results and no further information is provided about the participants (e.g., personality, technophobia. . . ); also, the small sample size of some studies prevent us from generalizing conclusions. Moreover, some results may be contradictory between the subjective evaluation and physiological metrics, indicating that complex dynamics in emotion regulation have to be further explored in some specific cases.
Finally, note that, even when Gross model is a suitable approach to classify emotion regulation strategies, Gross underlined some possible overlap between strategies and specifically between situation selection and situation modification. Thus, such classification is laborious to conduct a posteriori, since some of the techniques investigated were not described following this approach.

CONCLUSIONS AND PERSPECTIVES
The establishment of this review allowed us to combine substantial literature published in the last 10 years about an emergent topic. Mindless stress regulation technologies can take various forms and modalities and can rely on several low cognitive processes, as we have illustrated based on the Gross model (2014). This review presented not only how such technologies could help improve safety and comfort in driving but also the technical feasibility of their implementation in a car.
The growing interest in mindless technologies to regulate emotional states should see an increase in the number of studies on this subject in the coming years. In the driving context, several studies already focus on emotional regulation (for reviews, see Chung et al., 2019;Braun et al., 2020) without mentioning the mindless aspect explicitly. Indeed, mindless regulation could apply to more emotions than stress, with the added difficulty of differentiating these emotions. This could be achieved using, for instance, machine learning. This tendency is also observed in patents that also consider other emotions (e.g., KR Patent No. 101601957B1, 2016).
While such implementations start to see the light of the day (see section 6 from Braun et al., 2020), further inspiration from mindless computing could work toward an ubiquitous perspective in parallel with the development of calm technologies (Weiser and Brown, 1996). This would ensure not only safety on the road but also a better integration of the human being into a world overwhelmed by technology and information.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author/s.

AUTHOR CONTRIBUTIONS
AB was responsible for records collection and wrote the first draft of the manuscript. AB and AH-M organized the database, performed the statistical analysis, and wrote sections of the manuscript. All authors contributed to the conception and design of the study, revision, read, and approved the submitted version.

FUNDING
The work benefited from the support of University Gustave Eiffel, as part of the unifying Project Voyageur Virtuel: simulating human beings in their environment to anticipate the impacts of future mobility.