A Systematic Scoping Review of Human-Dog Interactions in Virtual and Augmented Reality: The Use of Virtual Dog Models and Immersive Equipment

Virtual reality is beneficial from a research and education perspective as it allows the assessment of participants in situations that would otherwise be ethically and practically difficult or impossible to study in the real world. This is especially the case where the assessment of human behaviour in the presence of stimuli (e.g. an aggressive dog) is being measured which could potentially constitute a risk in a real-world environment (e.g. a dog bite). Given that the dog is the most popular companion animal species, to date there is limited research that identifies and reviews the use of virtual and augmented reality directly relating to human-dog interactions. Furthermore, there also appears to be no review of the equipment and dog model specifications, such as dog breed and behaviours, which are currently used in these studies. As a result, this systematic scoping review searched ten databases to assess the current use and specifications of dog models which directly focused on human-dog interactions. Ten articles were identified. Six related to assessment or treatment of dog fear/phobia (cynophobia), three included multiple animal phobias, including dogs, and one article investigated the human and virtual dog interactions whilst walking. Six articles used a single breed (German Shepherd, Beagle, Doberman, and Rottweiler). Both the breed and behaviours displayed lacked justification and were often not evidence based. Specific measurements of model quality (e.g., polygons/vertices) were reported in only two articles which may affect repeatability and make comparisons between studies difficult. The virtual reality equipment (e.g. CAVE, head mounted display) and navigation methods (e.g. joystick, mouse, room scale walking) used varied between studies. In conclusion, there is a need for the accurate development and representation, including appearance and behaviours, of dog models in virtual and augmented reality. This is of high importance especially as most of the research covered in this review was conducted with the aim to treat the fear or phobia of dogs.


INTRODUCTION
Pet ownership in the United Kingdom is popular: as of 2020, 59% (17 million) of households owned a pet animal, the most popular species being dogs (33% of households; 12.5 million dogs) (PFMA, 2021). Companion and service/therapy/assistance dogs are suggested to provide a range of physical benefits (e.g., increased exercise and physical activity) and psychological benefits (e.g., reduced loneliness and depression, aids in social facilitation) to owners (see reviews by Sachs-Ericsson et al., 2002;Friedman and Krause-Parello, 2018;Gee and Mueller, 2019;Wells, 2019). Given our affinity and interactions with animals, it is therefore not surprising that they have been incorporated via entertainment and gaming technology into virtual pets for commercial purposes.
Over the past 30 years, pets have been replicated by technology including virtual (2D) and robotic pets. These can be either "realistic" or "unrealistic". Realistic pets are based on the appearance and/or behaviour of a real animal, e.g., Nintendo dogs (a virtual pet dog); AIBO (artificial intelligence robot, a robotic dog), and Lakaigo (a robotic dog imitating the locomotion of a real dog). Unrealistic pets do not fully resemble real-life animals but may have similar characteristics, e.g. Furby (a robotic pet); (Laureano-Cruces and Rodriguez-Garcia., 2012; Bylieva et al., 2020;Rativa et al., 2019;Peng et al., 2020). The traditional market for virtual pets, whether implemented as quickly as games or robots, is mainly children. Children use virtual pets for the purposes of: 1) entertainment; 2) learning how to take care of a pet (e.g., walking, feeding, etc., where the pet deteriorates in the absence of care), without the cost associated with real pet ownership; 3) companionship (Luh et al., 2015). However, virtual dogs (e.g., Nintendo dogs) can stimulate emotion and emotional attachment in users (e.g., Weiss et al. (2009) found that children made an emotional attachment with a robotic dog, AIBO) (Laureano-Cruces and Rodriguez-Garcia, 2012;Bylievia et al., 2020), but invariably do not offer the same level of companionship to that of a real pet might provide (Chesney and Lawson, 2007). Comparing social affordances between a stuffed dog and a virtual dog, the stuffed dog was associated with friendship and the virtual dog being associated with entertainment (Aguiar and Taylor, 2015). More recently, Lin et al. (2017) conducted a survey of 774 individuals who played games that included a virtual companion (e.g., Nintendo dogs) and found the main reason for playing was because the individual could not own a real pet (e.g., due to allergies) and virtual companions were deemed a form of emotional support.
In addition to entertainment, virtual dogs have a use in public health and education. Research has been undertaken into the use of virtual dogs for children as a means of increasing breakfast (Byrne et al., 2012) and fruit and vegetable consumption (Ahn et al., 2016) and promoting physical activity (Ruckenstein, 2010;Ahn et al., 2015;Hahn et al., 2020), increasing attitudes and empathy (Tsai and Kaufman, 2014), reducing obesity (Johnsen et al., 2014) and promoting effort making behaviours in learning (Chen et al., 2011). More recently, virtual animals have also been incorporated into mobile gaming apps (e.g., Pokémon Go) and have been found to be beneficial for human physical and psychological health. For example, Kogan et al. (2017) found that Pokémon Go usage increased the time spent with family members, walking their own 'real' dog, and exercising, as well as reducing anxiety levels.
As a result of recent technological advances, increased availability and the significant reduction in cost of equipment, the use of Virtual Reality in research has increased (Slater, 2018). The term "virtual reality" (VR) refers to a simulated threedimensional environment in which a user can be psychologically immersed through VR or AR (Augmented Reality) technology [such as an HMD (Head Mounted Display) or CAVE (Cave Automatic Virtual Environment)], and interact with the environment, through visual, auditory and haptic feedback (Virtual Reality Society, 2017;Johnston, 2018). VR provides a range of benefits such as user immersion and presence in the environment, the ability to potentially interact with a virtual object (such as a pet), the ability to elicit an increased degree of emotion, and the viewing area is much greater compared to 2D formats and is often, but not always, controlled by natural user movement (Lin et al., 2017). However, the degree of immersion, presence, perceptions and interactions in VR may be influenced by a variety of factors such as equipment, user's knowledge and experience, virtual environment, model development and appearance/quality/ realism (e.g., the "Uncanny Valley" as previously seen using realistic and unrealistic images of cats and dogs) (Yamada et al., 2013;Lin et al., 2017;Schwind et al., 2018).
There has been development of VR and AR applications for public entertainment. For example, in the VR game "The Lab-Postcards", released in 2016 by the Valve Corporation, a user can interact with a virtual robotic dog (fetch-bot) including haptic feedback upon contact with the dog and throwing a stick which the dog retrieves (Lin et al., 2017). More recently, as with Nintendo dogs in 2005, an AR mobile application dog "Dex" has recently been developed where users can walk, feed, play and look after their pet dog in AR (see Labrodex Studios, 2019).
More specifically, virtual animals may be of use in addressing public health outcomes directly related to contact with animals. For example, hospital admissions in England as a result of dog bites are increasing (Tulloch et al., 2021a) causing significant physical injury and interventions to prevent these occurring are required. Dog bites can also result in ASD (acute stress disorder) or PTSD (post-traumatic stress disorder) (Peters et al., 2004;Ji et al., 2010). VR animals developed for research and treatment of human participants exist. For example, the use of VR and/or AR for animal phobias, in the form of exposure therapy, is well established and includes a range of species such as spiders (Miloff et al., 2016;Tardif et al., 2019), cockroaches (Botella et al., 2010), dogs (Farrell et al., 2021), multiple small animals (Quero et al., 2014;Suso-Riber et al., 2019) and animals in general (zoophobia) (Suárez et al., 2017). Additionally, software companies also provide animals models for health care professionals for the treatment of various phobias (dogs, cats, snakes, spiders) [e.g. see InVirtuo (http://invirtuo.com/)].
The use of VR, in animal simulations has animal and human welfare implications. It may often be more ethical (i.e., no live animals used) and practical (i.e., one has control over a virtual stimuli/environment). In addition, it is a more affordable alternative to the use of live animals whilst allowing for repeated treatments (Farrell et al., 2021). Examples, where this is the case, include, animal-assisted therapy (Ratschen and Sheldon, 2019) (e.g., the Dolphin swim club https://thedolphinswimclub.com/), dog phobia treatments (Farrell et al., 2021) and animal dissections (Lalley et al., 2010).
Despite the latter benefits, to the authors' knowledge, there has been no scoping review on the current use, efficacy, advantages and disadvantages of the use of dog models in VR and AR. Here we focus specifically on a scoping review of direct human interactions with VR and AR dog models and the consideration and representation of the models physical appearance (i.e., breed) and behaviours displayed. The accurate representation of dog models and their behaviours is important, especially where they are used for injury prevention (e.g., education) and/or post-injury mental health treatment (e.g., phobia treatment).
Dog bites are often described as being "unprovoked" (Love and Overall, 2001), however, this is often not the case as evidence indicates that dogs show a range of behaviours before a dog bite occurs indicating stress, ranging from subtle "appeasement" signals (e.g., lip licking, yawning) that individuals may be less aware of to those that are more obvious (e.g., growling, showing teeth, barking) (Shepherd, 2009;Owczarczak-Garstecka et al., 2018). Therefore, the accurate representation of evidence-based dog behaviours is important from a public health viewpoint. Further, to ensure that the successful treatment of dog phobia occurs an individuals' understanding and recognition of dog behaviour is important (e.g., when to and when not to approach a dog in the real world based on behavioral signals). Furthermore, in the context of dog bites and aggression, the public media is often negatively biased towards specific dog breeds (e.g., bull breeds) (see review Kikuchi and Oxley, 2017) and this may influence public opinion. Therefore, exploration of breeds chosen and their contexts in VR and AR is important to evaluate.
If effective use of VR animal models is to be applied to realworld situations, an evidence-based approach is needed. Therefore, this review aims to: 1) Explore the scope of the field in which VR/AR dog models have been used in research with the focus directly on humandog interactions. 2) Describe the representation of virtual dog models (e.g., appearance/breed) and dogs behaviour including evidencebased development and fidelity.
3) Identify what equipment is used and if/how these differ between studies. 4) Describe the main findings of the research and measures used, both objective and subjective, to assess the human-dog interaction and other measures used in VR.

METHODS
This scoping review adhered to PRISMA (Preferred Reporting of Items for Systematic Reviews and Meta-Analyses) guidelines and methodology (Moher et al., 2009).

Identification of Relevant Studies and Search Criteria
Literature from a 30-year period (January 1990-September 2020) was reviewed due to the rise in the popularity of VR from the 1990s and the invention of CAVE (Cave Automatic Virtual Environment) in 1992 (Cruz-Neira et al., 1992). Data collection occurred on the 9th and 10th of October 2020. Due to the multidisciplinary nature of research articles using VR and AR dog models, ten databases were searched, covering psychology (APA), veterinary science (CABI direct), medical and veterinary (Cochrane library, PubMed, Medline), technology, computing, and engineering (IEEE, ProQuest) fields, in addition to the large databases; Scopus, Web of Science, and Google Scholar. In addition, to database searches, references from relevant articles were identified by reviewing these manually.
The search terms were used to identify relevant articles using the article title, abstract and/or keywords are given in Table 1.
Peer-reviewed journal articles and conference articles were included in the search findings but not editorials, commentaries, reviews ( Table 2). Conference articles were included due to the recent emergence of this area of research and several relevant conference articles specifically focusing on human interactions with a VR or AR dog model (e.g., Hnoohom and Nateeraitaiwa, 2017;Norouzi et al., 2019).

Behavioural Dog Models
Articles included in the review are displayed in Figure 1. All articles involved dog models which displayed some form of behaviour and focused on direct interaction between the human and virtual dog. The first category of articles, for exclusion from this study, consisted of indirect VR dog model use; the dog model was not part of the main purpose of the study. Examples include, haptic forces used for rehabilitation through the use of simulated dog walking (Sorrento et al., 2018), used to facilitate the study (e.g., leads or assists the users to an area as part of a non-dog related study/task (e.g., Hung et al., 2018)) or study conditions (e.g., a red robot dog that barked to distract the user (Rewkowski et al., 2019)). Articles were excluded if they were in 2D due to the reported disadvantages when compared to 3D VR including reduced levels of presence, immersion, and spatial navigation success rates (Slobounov et al., 2015;Minns et al., 2018). Articles with the use of mobile phones were only included if they consisted of 3D VR/AR with an HMD as they are likely to provide a similar VR experience (e.g., stereoscopic vision, enclosed eyes). The second category, for inclusion in this study, was direct VR dog model use; the dog model was a key part of the study with direct focus and involvement of, and/or interaction with, the dog model (e.g., phobia treatment) ( Figure 1).

RESULTS
In total ten articles were found to directly research, or propose future research, human interactions with virtual dog models using a VR or AR set up. Despite the initial 30-year inclusion period, all articles were published from 2008 onwards ( Table 3).
Nine articles included some form of results from participants [mean sample size = 13.2 (range: 6-32)]. One article described the development of VR animal models (including dogs) for future use to treat phobic participants but did not report research with participants (Maglaya et al., 2019).

Areas of Research and Measures
Nine out of the ten articles specifically focused on the topic of the development of a VR dog model to stimulate emotions or the proposed or actual treatment of individuals who were fearful or had a phobia of dogs (cynophobia) (6/9) or multiple animal phobias (i.e., zoophobia) which included cynophobia (3/9). One article targeted non-phobic individuals to investigate the proximity to and collision between an AR dog model and a human who was walking the dog.
Nine studies recorded some form of subjective measurement, with the most commonly used being the Subjective Units of Distress Scale, some form of presence measurement (e.g., Igroup questionnaire) and a subjective Behavioural Assessment Test. One study recorded biological/physiological measurements including skin conductance (Taffou et al., 2013). Another article briefly mentioned that measurements of heart rate, anxiety and sweating were recorded but no further details were provided (Suárez et al., 2017) (Table 3).

Main Findings
Research articles mainly focused on the evocation of fear and the treatment of fear and phobias through VR dog models. It was evident that the dog models resulted in an increase in fear, distress, anxiety, and behavioural responses. Audio, where recorded, in the form of dog vocalisations (e.g., growling, barking) also appeared to increase fearfulness of the dog. Of those studies which specifically used the dog model as part of a dog fear or phobia treatment, these often result in reduced fear or phobia ( Table 3). For example, in one article 75% of children were deemed as recovered 1 month after treatment (Farrell et al., 2021).

Equipment
Equipment varied from four studies using a AR/VR HMD (e.g., Oculus Rift) and five articles using a projection screen (single or multiple screens (e.g., CAVE/BARCO Ispace/Blue room)). Out of the nine articles where the user navigation/control method was stated, six used a hand controller (e.g., mouse, joystick, game controller, remote control), one article a therapist controlled the movement through a tablet, one article there was room scale movement for the user and one article it was unclear the if the user navigated or moved their head only (i.e., 3DOF or 6DOF) ( Table 4).

Breed, Coat Colour and Behaviour
Seven articles stated the breed of the dog model used which included six studies using a single breed [German Shepherd, Beagle, Doberman (3), Rottweiler] and one study which included videos of multiple breeds (Cocker Spaniel, Labrador x Kelpie, Rottweiler x Border Collie, Cavoodle, Japanese Spitz). Where a single breed was used, in some cases different colours and textures of the models were included (see Table 5). There was a lack of justification and/or scientific evidence for the dog behaviours displayed and were often predefined prior to purchase of the model. The number of behaviours displayed often varied between studies and limited detail about the behaviours was provided (see Table 5).

Dog Model Quality
The quality of the virtual dog models in terms of polygon or vertices count was not mentioned in any article. In one case there TABLE 1 | Search terms used for title, abstract or key words. Acronyms being used for Augmented Reality (AR) and Virtual Reality (VR) were originally included, but due to the broad alternative use of AR (e.g., AR protein/gene expression, androgen receptor, allergic rhinitis; allelic ratio, anterior right) searches were conducted separately and initially reviewed for each database but no new articles were identified.

Search terms
("Virtual Reality" OR "Virtual Environment" OR VR OR "Mixed Reality" OR "Augmented Reality") AND ("Companion animal" OR "companion pet" OR "pet animal" OR pet OR pets OR dog OR dogs OR canine OR cynophobi a ) a Asterisk indicates plural terms (e.g., cynophobic or cynophobia). Mobile phones/tablets that are used on their own without an HMD.

Literature focus
All articles which include a VR representation of a live pet dog that displays behaviours and is the focus of the article Human-dog interactions is not the focus of the study. VR robotic dog models, anatomical models, 2D dog models, real dogs and/or non-dog animal models. Software/technical development with no reference to VR.
Frontiers in Virtual Reality | www.frontiersin.org March 2022 | Volume 3 | Article 782023 was a web link to a pre-defined dog model which highlighted the number of polygons via an external website ( Table 5). In one study investigating multiple phobias, the dog model was described in very little detail and therefore unlikely to be replicated in future research (Maskey et al., 2019). Another study used 360-degree video footage of real dogs in conjunction with a VR headset and separate assessments with the use of real dogs (Farrell et al., 2021).

DISCUSSION
The aim of this review was to identify and assess research that directly involved the use of human interactions with dog models in VR and AR. To the authors knowledge this is the first scoping review specifically identifying and assessing human interactions with VR and AR dog models, model quality, behaviours displayed, and equipment used. Findings from this review highlight that although research using VR is well established, the development and use of VR and AR dog models for the purpose of human-dog interaction assessment is in its infancy.
The use of VR dog models as a form of exposure therapy had positive effects. However, there was variation in the study sample size, VR equipment used and the behaviours displayed by the virtual dog, which tended to lack an evidence-based approach to the development of a canine model in relation to canine behaviour.

Equipment
There were several different VR HMD's and screen-based systems identified. Changes and advances in technology are inevitable. Furthermore, as technology improves other forms of HMD's become outdated and are no longer used which highlight the importance of stating technical specifications of all equipment used in research with VR models. This should include: • VR equipment (HMD/Screen/CAVE) specifications: Navigation method, whether the VR HMD is 3DOF or 6DOF, HMD specifications (resolution, refresh rate, field of view, tethered or wireless), tracking (outside in or inside  • State Trait anxiety Inventory (STAI) Spielberger et al. (1983) Task: Participants were asked to explore the area to find a green frog which was visual and produced sound and found in the surroundings of the dogs Sig. Higher (p < 0.01) STAI scores between after VR exposure Behavioural: Behaviour ratings of the user in the presence of a virtual dog (scale 1-6 (1-dog not noticed to 6dog noticed and flight or freeze)) Virtual dogs evoked a verbal and behavioural reactions (rating median score range 2-5 when exposed to 4 virtual dogs) Paper  Hnoohom and Nateeraitaiwa (2017) (n = 10) • Four-point Likert scale ("Few, average, much, very much"): Realism of application, "dreadfulness" of the dogs in three levels, sound, animation, and distance between the player avatar and dogs Behavioural: Not stated A dog model and environment were developed Conference paper (5M/5F) 50% of participants rated the freestanding dog in the back yard the most "dreadful" followed by the dog in a cage and the dog in the house 30% rated the following statement as "much" and 60% "average". "Hearing the dog sounds made us more fearful" VR ". . .whether VR OST results in clinically significant improvement for children with a specific phobia of dogs using a controlled, multiple baseline case series design where participants are randomly assigned to 2-, 3-or 4weeks baselines, followed by the VR OST and a 1 month follow-up." (p.4) Multiple Baseline (2, 3, 4 weeks) case Series • Anxiety Disorders Interview Schedule: Parent (ADIS-P) Silverman and Albano,  VR "the objective of this project is to provide a reasonable alternative for treating various types of Zoophobias, using virtual reality" (p.1) Preliminary/pilot • "Laboratory tests were performed with the experimental group using virtual reality and traditional therapy with the control group. In each patient, five sessions and two levels of complexity were performed." (p.5) Task: Five sessions and two levels per person-No further detail provided. "The clinical status of the patients involved in the tests had symptoms of high heart rate, numbness, excessive sweating and anxiety" (p.5) "After the session of the fifth practiced patients treated with VR, although the symptoms did not disappear completely, an 80% decrease in anxiety, sweating and heart rate was observed in all cases; While patients in the control group treated with traditional therapy, they had a 35% reduction for the same symptoms." "In this study and development, VR will be used as a tool to aid psychologist and psychiatrists in assessing and treating the different fear levels of patients"(p1.39)

Prototype
• Not applicable-prototype Task: Not applicable -prototype Proposed usage for multiple animal phobias "Cynophobia: The patient will be situated inside a house. Lower levels of experience will involve sounds coming from a dog. The next level will be a shadow of a dog outside the window. The next level will be a dog on a leash slowly getting closer to the patient until the patient can touch the dog" (p.140) Paper (  • ADIS-P Silverman and Albano, (1996) Customed scenes designed based on an individual's phobia. The four sessions occurred in hierarchical order from lowest severity to most intense but only if low levels of anxiety were reported. CBT and relaxation methods were used during each VR session (such as challenging thoughts) In comparison to the control group, treatment groups had significantly improved on target behaviour ratings from baseline to 2 weeks (p = 0.021) and baseline to 6 weeks after the exposure session (p = 0.007) Alongside visual and audio feedback, haptic feedback in VR is important as it can enhance user immersion as it allows simulated physical interaction, and feedback, between a user and virtual or a combination of real and virtual objects within the virtual environment (Wang et al., 2019). For example, Carlin et al. (1997) conducted a case study of an individual with a spider phobia and found that touching a real toy spider, whilst viewing a VR spider, provoked a strong emotional response. In the present review, no articles indicated that they used haptic feedback as part of the VR setup. This could be due to the type of studies that were conducted as the majority focused on the treatment of phobia and therefore the contact with a dog may be unlikely. In contrast, the use of haptics may be of use in a dog phobia context especially for patients who are gradually exposed and become comfortable with the presence of dogs eventually coming into "contact" with the dog. The use of bespoke VR setups and varying navigation  Rottweiler "This paper selected the model we use that suitable and realistic with the scene" "Rottweiler is fierce" " 26 animations can apply to this work."

Not stated
Behaviours: 26 animations (not stated). The VR task consisted of three levels: 1. Dog sleeping in a living room and when participants approach it the dog sits up and starts panting 2. Standing inside a cage in the back garden. When a user gets within close proximity the dog turns and growls at the user, if the user gets closer the dog will "attack" and bark 3. The dog is standing in the back garden and behaviours are the same as level 2. However, when a user is within closer proximity the dog runs at the user and eventually leaps and attacks the user Vocals: Barking, growling, panting Residential area-1) Living room of a house, 2) back garden in a cage and 3) outside the gate of the house (Continued on following page) Frontiers in Virtual Reality | www.frontiersin.org March 2022 | Volume 3 | Article 782023 All six breeds used in VR video "Each dog was selected based on providing a variation of breeds and sizes to maximize variability."

4K 360°video viewed in a VR HMD
Behaviours: There were ten levels including: "1 Dog and assistant walks into and sits on the opposite side of the room (on leash); 2 Subject moves closer to dog (on leash); 3 Subject moves closer to dog (on leash); 4 Subject moves directly next to dog (on leash); 5 Subject back to original side of room, assistant and dog standing up walking 1m forward (on leash); 6 Assistant and dog standing up walking 1m forward from previous position (on leash); 7 Assistant and dog standing up walking 1m forward from previous position (on leash); 8 Dog walking side to side and around camera (on leash); 9 Dog walking/ running towards subject (off leash and assistant in room); 10 Dog without assistant in room and no leash" (p7.) "A large room" was used for the VR video treatment and post treatment. (limited information provided) • Doberman Post VR assessment with a real dog-dogs varied and no breeds were stated Vocals: Not stated Multiple phobias

Suárez et al. (2017) German Shepherd German Shepherd
Pre-defined behaviour on purchase "Each of these models is really well made" Behaviours: "walk, run and sit" Vocals: Not stated "3D House model with three floors and furniture" Maglaya et al.
Not stated Not stated Not stated Not stated Behaviours: Minimal detail ["The Patient will be situated inside a house. Lower levels of the experience will involve sounds coming from a dog. The next level will be a shadow of the dog and gradually revealing a dog outside the window. The next level will be a dog on a leash slowly getting closer to the patient until the patient can touch the dog." (p.140)] "The Patient will be situated inside a house." Maskey et al. methods (e.g., mouse/joystick) by individual laboratories may have also played a role in the lack of haptic feedback used as separate development may have been needed. Having said this, the use of realistic haptic feedback in VR is complex and commercial VR controllers are limited to various basic forms of vibrations (Wang et al., 2019;Yin et al., 2021). Further research exploring the use of basic and more complex forms of haptic feedback (see review by Yin et al. (2021) for the current and future use of haptics in AR and VR) in human-dog interaction studies in AR/VR would be beneficial, especially in dog phobia and educational research.
In the present review only one article used AR. More research is needed on the use of AR dog models as it provides increased ecological validity compared to VR and interaction with a users own hands rather than virtual hands (Suso-ribera et al., 2019).

Research Studies
The majority of articles focused on the assessment and treatment of humans with a fear or phobia of dogs or animal related phobias. For example, Farrell et al. (2021) found that the majority of participants (75%) were deemed to have recovered 1 month after a one-session treatment, but the sample was small (n = 8). This technology could be beneficial in future clinical real-world applications. Recent hospital data indicates that NHS waiting times in England are an important public concern (The Kings Fund, 2021). There has also been a significant increase in demand for mental health services which has been exacerbated by the COVID-19 Pandemic (NHS Providers, 2021). In addition, the rate of hospital attendance due to dog bites has reported to have increased during COVID-19 lockdowns, likely due to the increased contact between humans and dogs (Dixon and Mistry, 2020;Tulloch et al., 2021b). This could result in an increased rate of dog bite victims seeking mental health advice and treatment (such as for PTSD or ASD). However, mental health interventions such as exposure therapy is deemed a nonurgent treatment. Therefore, further research into the role of AR and VR technology which could assist mental health practitioners or even replace the involvement by professionals is needed.
Exposure therapy could be an opportune moment for the education of individuals about appropriate and inappropriate behaviour in the presence of dogs and general dog behaviour. Yet, only a single paper mentioned, although briefly, that the researchers incorporated education about dog behaviour and safe interactions with a real dog (Farrell et al., 2021;p.7). This highlights the potential for future research using VR and AR dog models as a form of educational intervention, either stand alone or alongside phobia treatment, for both children and adults, regarding appropriate behaviour around dogs and recognition of specific dog behavioral signals. Further exploration is needed into the impact that experiences with AR and VR dog models and associated educational applications have on the potential for participant behaviour change. As previously highlighted by Schwebel et al. (2012) dog bite prevention education in the form of online software may increase knowledge but does not result in behaviour change.
Often VR dog models are developed for an individual or multiple studies by the same organisation/research group and therefore there is little systematic re-use of dog models. Having different dog simulations makes comparisons difficult as each simulation may have different effects on human users, depending on how accurate the models appearance and behaviour is. Similar issues have previously been highlighted in research involving virtual human avatars (Mountford et al., 2016). Further, little reference to the quality of the model (e.g., high or low polygons) was provided. Judging the quality of dog models is important due to the potential impact it has on a user's behaviour towards and interpretation of the dog. Previous research has highlighted that the impact of model quality and design (i.e., anthropomorphic features, naturalness, stylisation) could relate to the perceived realism of virtual animals (Schwind et al., 2018). For example, Schwind et al. (2018) note that if a virtual animals appearance deviates from its natural appearance (e.g., human facial expressions), or movement, then this can result in negative perceptions (e.g., eerie sensation/uncanny valley) of the virtual animals and may have the potential to affect interactions with them. In contrast one study, used a VR HMD (Oculus Rift) to view 360 degree videos of real dogs with positive results (Farrell et al., 2021). Initially this method appears to overcome issues associated with the need to design accurate and realistic models. However, this format of VR has several practical limitations. For example, firstly, interactions with dogs in the video is not possible; secondly, initial video footage is required with various dog breeds, behaviours, space and permission to film the footage is required. Thirdly, additional ethical approval is needed for both the use of animals, especially where a dog may be display aggressive behaviours, and human participants (Swobodzinski et al., 2021).

Dog Breed
Several articles chose specific breeds such as Rottweilers or Dobermans (Viaud-Delmon, 2008;Hnoohom and Nateeraraitaiwa, 2017). In some cases, breed choice was justified, for example, Viaud-Delmon (2008) conducted the screening of nine different breeds, and based on ten participants, found that the Doberman was the animated dog model which provoked the most negative emotional response. However, the latter study did not state if participants had any previous experience with dogs or were involved with a dog related incident such as a bite. Further research would be useful to ascertain the difference between individual perception based on limited or no experience of dogs and those who are phobic of specific breeds due to a dog related incident.
Furthermore, other research does not appear to justify the choice of breed or chooses a breed based on likely biased perceptions of the breed; for example, Hnoohom and Nateeraraitaiwa (2017) used a virtual reality dog model based on a Rottweiler breed and refers to the dog as a "fierce dog". Similarly, an online company advertising the treatment for the fear of dogs through VR also states, "One of the most commonly feared dogs, Rottweiler, often considered dangerous" (Psious, 2018). Similar inflammatory language (e.g. "ferocious" and "vicious") has been previously reported for Rottweilers and German Shepherds in medical literature (see Arluke et al., 2018, p.216).
Choice of specific breeds could have been influenced by external factors such as the news media which often focus on specific breeds (Kikuchi and Oxley, 2017) or breeds, such as Rottweilers, German shepherds and Dobermans, frequently used as guard and police dogs (Podberscek, 1994;Meade, 2006). A recent survey of veterinarians in the United States regarded the Rottweiler and German Shepherd as breeds which poses a high risk of biting and evoke a negative emotional response if an unfamiliar adult dog, which was off the lead, ran up to them (Kogan et al., 2019). Although it is likely that some breeds may be perceived as more aggressive or fearful than others, it is important to highlight that all dogs have the potential to bite and can be due to multiple factors such as management, health status, genetics, and environment (including human and dog behaviour) (Haug, 2008). The role of dog model physical characteristics and the impact it has on human perception and behaviour is an area that requires further research, for example the effects of skull (brachycephalic, mesocephalic and dolichocephalic) and ear shape, tail length, coat colour and type, size (toy, small, large, giant) and weight (underweight or overweight).

Coat Colour
The coat colour of the dogs was briefly discussed. Suied et al. (2013) found that participants were more fearful of a dark coloured dog in comparison to a white or brown. However, given the same Doberman model was used, the reaction of participants could have been in relation to the most realistic dog model in terms of both breed and natural colour, as Dobermans are stereotypically known and associated in roles and the media with black coats and less often brown or not at all with white coats. Further research would be useful into the impact that coat colour has on human behaviour and participants perceptions; especially as black dog syndrome (also known as big black dog syndrome) appears to be frequently mentioned online despite there being little evidence to support this phenomenon (Woodward et al., 2012;Sinski et al., 2016). In previous research, breed specific differences and size have been found to be more influential factors than the coat colour of dogs (Woodward et al., 2012;Sinski et al., 2016). From a research perspective, VR is a useful tool in this respect as size and colour can be controlled and changed with relative ease, whereas multiple similar-looking dogs would be required in real life scenarios to test these variables.

Dog Behaviours
The dog models used in this review appeared to display generic behaviour with limited evidence of behaviours being based on canine behavioural science research or expert feedback. It was evident that behaviours were frequently predefined based on models that were purchased. This could be due to the type of research that the dog models were being used for (i.e., dog phobias) and therefore it was perceived that a dog model which displays basic behaviours such as walking, sitting, barking, jumping were required. Alternatively, models that can be purchased with predefined behaviours can be preferable as less time is needed for development. However, accurate behaviour representation is important to consider, especially in the case of dog phobic participants. The display of subtle (e.g., growling, barking) and more intense (e.g., running towards, lunging or attacking (Hnoohom and Nateeraitaiwa, 2017)) behaviours towards participants is likely to be required for realistic treatment but also may cause significant stress and needs careful consideration in this context.
Realistic behaviours can be included in a form of exposure therapy and range from relaxed, play to fear and agonistic behaviours. It is important to note that dog behaviour can be complex and could be easily misinterpreted by an untrained individual. For example, appeasement signals (also known as calming signals) may include behaviours such as lip licking, yawning, and paw raises, indicating stress and discomfort which are often misinterpreted (Shepherd, 2009) and were not included in the reviewed articles. Similarly, theories about dog behaviours and their meaning can vary such as in the case of dominance of dogs towards humans (Westgarth, 2016). This highlights the importance of collaboration between animal behaviour experts and VR/AR developers. Often this type of collaboration appears to be lacking presumably due to the need for large amount of animation and technical development of models or the reliance on predefined models.
Finally, the importance of messaging also needs consideration, even if hypothetical, within the virtual environment especially regarding the treatment and management of animals. For example, Hnoohom and Nateeraraitaiwa (2017) display a virtual dog within a cage which, if in reality, would be considered a serious welfare concern in many countries.
In conclusion, this review highlights the current limited use of dog models in VR and AR. The small number of reviewed articles generally were also limited by small sample sizes and the results need to be interpreted with caution. This review also only included English articles. Despite this there was some evidence to indicate that the use of VR to treat dog phobias is effective and holds much potential, especially including the assessment of participants physiological parameters (heart rate, skin conductance, eye tracking, etc). Of the studies found, there is a lack of emphasis placed on the dog model's behaviour, breed and quality. Future developments and research need to consider appearance (e.g., breed and unbiased basis for this), canine behaviour (based on up-to-date evidence-based research and canine behavioural expert review) and quality of dog models. We also recommend that the detail of the dog model is reported including the sources or development of the model, quality (i.e., polygons/tris/vertices), and behaviours displayed. Future collaboration between canine behavioural experts and VR and AR developers would be beneficial for an accurate and realistic representation of dogs in virtual reality.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article, further inquiries can be directed to the corresponding author. reviewed all articles and wrote the first draft of the manuscript. CW, GM, KS provided feedback, discussion and comments on drafts of the manuscript. All authors discussed and contributed to the manuscript.

FUNDING
JAO is a PhD student funded by a Dogs Trust Canine Welfare Grant.