Paper deep dive
Toward Inclusive Avatar Design with Limb Differences Through Artificial Intelligence
Fernanda Miyuki Yamada, João Paulo Gois, Hiroki Takahashi
Intelligence
Status: succeeded | Model: Gemma-4-26B-A4B | Prompt: intel-v1 | Confidence: 91%
Last extracted: 7/18/2026, 2:42:25 PM
Summary
This paper reviews the challenges and opportunities in designing inclusive 3D avatars for users with limb differences and amputations. It highlights that current XR systems largely support normative bodies, leaving users with morphological variations underrepresented. The authors analyze existing guidelines for inclusive design, critique current applications that often rely on prosthetic modeling rather than structural variation, and propose Artificial Intelligence (AI) as a key solution. Specific AI methods like AJAHR and DreamAble are discussed as steps toward accurate reconstruction and text-based generation, though limitations in dataset diversity, rigid parametric models, and animation pipelines remain.
Entities (12)
Relation Signals (8)
3D Avatar → supports → Normative Bodies
confidence 95% · Most 3D avatar systems only support normative bodies
3D Avatar → failstorepresent → Limb Differences
confidence 94% · do not accurately depict people with limb differences, amputations, or other morphological variations
Artificial Intelligence → enables → Inclusive Avatar Generation
confidence 92% · This paper positions artificial intelligence as a promising path to overcoming these limitations and advancing inclusive 3D avatar generation
DreamAble → generates → 3D Avatars
confidence 90% · DreamAble... generates 3D avatars with upper limb differences from natural language descriptions
AJAHR → reconstructs → 3D Human Meshes
confidence 90% · AJAHR... reconstructs 3D human meshes from images while accounting for missing limbs
SMPL → assumes → Normative Bodies
confidence 88% · Although these models ensure anatomical consistency, their design assumes normative bodies
Mindscape → features → Prosthetic Devices
confidence 85% · Mindscape 2 allows players to create 3D avatars that use wheelchairs or prosthetic devices
ParaJecripe → raisesawarenessof → Adapted Sports
Cypher Suggestions (0)
No Cypher suggestions yet.
Abstract
Abstract:As extended reality becomes more popular for social interaction and entertainment, 3D avatars must represent the full diversity of body types. Most 3D avatar systems only support normative bodies and do not accurately depict people with limb differences, amputations, or other morphological variations. This paper reviews emerging technical approaches for inclusive 3D avatar customization for this group and current guidelines that promote respectful and accurate representation. We highlight persistent challenges, including the scarcity of diverse datasets and the limitations in animation for non-normative anatomies. This paper positions artificial intelligence as a promising path to overcoming these limitations and advancing inclusive 3D avatar generation.
Tags
Links
- Source: https://arxiv.org/abs/2607.11512v1
- Canonical: https://arxiv.org/abs/2607.11512v1
Trouble viewing inline? Open PDF directly →
Full Text
33,580 characters extracted from source content.
Expand or collapse full text
Toward Inclusive Avatar Design with Limb Differences Through Artificial Intelligence Fernanda Miyuki Yamada 1 , Jo ̃ao Paulo Gois 2 , and Hiroki Takahashi 1 1 The University of Electro-Communications, Chofu, Tokyo, Japan 2 Federal University of ABC, Santo Andr ́e, S ̃ao Paulo, Brazil Abstract As extended reality becomes more popular for social interaction and entertainment, 3D avatars must represent the full diversity of body types. Most 3D avatar systems only support normative bodies and do not accurately depict people with limb differences, amputations, or other morphological variations. This paper reviews emerging technical approaches for inclusive 3D avatar customization for this group and current guidelines that promote respectful and accurate representation. We highlight persistent challenges, including the scarcity of diverse datasets and the limitations in animation for non-normative anatomies. This paper positions artificial intelligence as a promising path to overcoming these limitations and advancing inclusive 3D avatar generation. Introduction In extended reality (XR) applications, 3D avatars mediate the user interaction with dig- ital environments, shaping self-representation and overall experience. Despite recent ad- vances in modeling and animation, most 3D © 2026 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses, in any current or future media, in- cluding reprinting/republishing this material for ad- vertising or promotional purposes, creating new col- lective works, for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works. DOI: 10.1109/MCSE.2026.3685086 avatar customization systems continue to de- fault to normative bodies. This lack of mor- phological diversity affects mainly users with visible physical disabilities, such as limb dif- ferences, amputations, or other anatomical variations. Such users are often forced to choose between misrepresenting themselves through normative 3D avatars or forgoing au- thentic self-representation entirely. Conse- quently, the inability to create accurate self- representations in XR can reinforce feelings of exclusion present in physical spaces. Inclusive 3D avatars offer more than just self-representation.They can impact ar- eas such as virtual sports, rehabilitation, and education.For example, while digi- tal games frequently simulate sports for nor- 1 arXiv:2607.11512v1 [cs.HC] 13 Jul 2026 mative bodies, there is a notable lack of virtual environments inspired by adapted sports. This gap highlights the potential of inclusive 3D avatars to increase the visibil- ity of Paralympic athletes, challenging tradi- tional assumptions about ability and promot- ing greater appreciation for the performance and dedication of athletes with disabilities. Achieving these outcomes remains chal- lenging due to the scarcity of 3D anatomi- cal representations that capture diverse body types and the lack of established design stan- dards to guide developers. Recent progress in artificial intelligence (AI) provides promising means to address these limitations. Advances in generative modeling, computer vision, and physics-based simulation have enabled acces- sible 3D avatar generation and improved rep- resentation of morphological diversity in vir- tual environments. This paper presents an overview of emerg- ing approaches for inclusive 3D avatar gen- eration, with a focus on users with limb dif- ferences, amputations, and other morpholog- ical variations. Refer to the Key Terminology box for definitions of terms used throughout the paper. We advocate for digital adapted sports as a domain where such 3D avatars can have a high social and cultural impact. We emphasize the interplay between techni- cal innovation and ethical design, reviewing existing applications and systems that ad- dress anatomical diversity, dataset consider- ations, and design guidelines that support di- verse representation. Finally, we discuss fu- ture directions in which AI may play a critical role in enabling accessible, socially responsi- ble XR applications. Self-representationandCus- tomization Systems Self-representation refers to how the user or player is embodied and visually represented within a digital environment. Systems that let players represent themselves, instead of only controlling a predefined protagonist, can strongly influence how interactive content is experienced. Despite this importance, many games still provide only narrow support for self-representation in practice. A large share of action and adventure titles use fixed pro- tagonists defined by the narrative. When cus- tomization exists, it often covers only surface elements such as clothing and accessories. This pattern appears in titles such as Until Dawn, Death Stranding, and The Last of Us. Character customization systems allow players to change the appearance of the on- screen character, instead of using a fixed pro- tagonist. Market indicators show that cus- tomizable avatars are increasingly common in commercial games. Golando [1] noted that 7 of the top 13 best-selling video game titles as of February 2023, approximately 54%, use a “Customizable Avatar” model rather than a fixed protagonist. Similarly, metadata from Steam, the largest PC gaming platform, iden- tifies more than 13,550 active titles explic- itly tagged with “Character Customization” 1 . At first glance, these numbers seem to sup- port the idea that modern games increasingly enable self-representation. However, charac- 1 SteamDB, “Character Customization” tag. Availableat: https://steamdb.info/tags/. Accessed February 9th, 2026. 2 Key Terminology • Limb difference: a congenital or acquired condition in which one or more limbs are absent or partially formed. The term refers to variations in limb presence and morphology and does not imply any specific level of functionality. • Amputation: surgical or traumatic removal of a limb or part of a limb. This term refers specifically to acquired cases and is distinct from congenital limb differences. • Morphological variations: differences in body structure, proportions, or configuration relative to the standardized human model commonly presented in medical contexts. These include, but are not limited to, limb differences and variations in skeletal morphology. • Non-normative anatomy: anatomical presentation characterized by morphological variations, referring to bodies that differ from the conventional anatomical reference mod- els used in medical contexts. • Prosthetic/Assistive device: a wearable device that replaces or supplements a missing or impaired body part or function. It may be designed to restore lost capabilities, as in prosthetic devices, or to enhance and assist existing functions, as in assistive devices. ter customization does not necessarily mean true self-representation. Many systems offer only cosmetic adjustments, with little sup- port for body-shape editing. As a result, users may interact with a customized char- acter that still does not reflect their identity or self-image. The gap is larger for people with limb dif- ferences and other morphological variations. Some games feature playable characters with visible limb differences without prosthetics, such as titles in the Devil May Cry and Guilty Gear series. These examples show that non-normative body structures can be integrated into gameplay without compro- mising the functionality. However, such char- acters are predefined by the developers and offered as fixed protagonists, rather than cus- tomizable options. Therefore, these designs support inclusion at the narrative level but do not enable true self-representation for people with limb differences or other morphological variations. Moreover, a content analysis of 108 game trailers released between 2006 and 2016 found that, although disability is occa- sionally depicted in background non-playable characters, only 1% of the content included playable characters with disabilities [2]. A similar pattern appears in XR systems, where users are typically represented through 3D avatars, and embodiment plays a central role in interaction and presence. Surveys of social virtual reality platforms report that customization support is limited. Handley et 3 al. [3] analyzed 44 social virtual reality plat- forms and found that only 27.3% of systems in their catalog offered customizable features. Options for amputations or limb differences are rarely available. A growing body of XR research addresses limb differences in rehabil- itation and clinical contexts. This sustained research activity indicates that 3D avatar representations with limb differences are not a niche or isolated topic, but rather an ac- tive and expanding area of XR research with demonstrated practical applications. Guidelines in Avatar Design Although the importance of inclusive 3D avatars is increasingly recognized, translating this goal into practice remains challenging. Some physical disabilities are easier to in- corporate within standard frameworks, par- ticularly when they do not alter the under- lying anatomy or skeletal structure of the 3D model.For instance, hearing devices, wheelchairs, or lower limb prosthetics can of- ten be added as separate items. This discrep- ancy creates an unfortunate hierarchy of rep- resentation, where some disabilities are more “visible” in digital spaces than others. Addressing this gap requires not only tech- nical innovation but also a structured un- derstanding of inclusive design practices. In this context, clear guidelines for 3D avatar design ensure that users with limb differ- ences are represented accurately and respect- fully in XR applications. These guidelines are often developed through user-centered re- search, where people with disabilities, design- ers, and accessibility experts collaborate to identify what works and what causes exclu- sion. Inclusive 3D avatar guidelines emphasize giving users flexible and granular control over their virtual representation. Zhang et al. [4] suggest using continuous controls, like sliders or text prompts, so that users can describe their unique body differences naturally. This includes adjusting body features such as limb length, symmetry, and joint movement using smooth and flexible controls. User preferences for 3D avatar representa- tion can vary depending on social and per- sonal considerations. Park & Kim [5] found that some users prefer to hide their disabili- ties to enjoy activities they cannot perform in real life and avoid stigma, while others value showing their disabilities to raise aware- ness and facilitate communication. Regard- less of these differences, participants agreed that customization systems must offer diverse options. Together, these studies underscore that inclusive 3D avatar customization systems have to: (1) allow users to customize the appearance of their 3D avatars to support self-representation; (2) provide granular con- trol over the representation of physical dif- ferences and assistive devices; (3) support context-dependent representations, allowing users to represent themselves with or with- out prosthetics, and reflecting personal pref- erence and situational context; and (4) inte- grate user-centered design and validation. From a technical perspective, the most challenging and important aspect is allow- ing users to represent themselves without 4 prosthetics. As discussed at the beginning of this section, physical disabilities that do not require modifications to the underlying skeleton can be added relatively easily to the design of 3D avatars.Modeling ab- sent limbs requires deviation from norma- tive body templates, affecting not only the mesh but also the skeleton, rigging, and ani- mation pipelines. Consequently, people with limb differences who choose not to use pros- thetics or cannot use them often remain the most “invisible” in virtual spaces. When limb differences are represented using prosthetics, the system defaults to normative body tem- plates, thus only simulating the appearance of a disability rather than promoting au- thentic representation. The stakes are high, as accurate and realistic representations re- quire not only careful 3D modeling but also fully adaptive animation pipelines that pre- serve natural motion without defaulting to normative body assumptions.In the fol- lowing, we analyze current 3D applications and avatar customization systems to assess whether these guidelines are implemented or overlooked in the representation of limb dif- ferences. Current Applications and Limi- tations Considering the principles of inclusive 3D avatar design outlined in the previous section, it is notable that many systems that incorpo- rate 3D avatars with limb differences or am- putations still focus primarily on represent- ing users with prosthetics. Although pros- thetic modeling is helpful in clinical contexts, its default inclusion in 3D avatar customiza- tion can hinder authentic self-representation in social settings where users may prefer to represent themselves without prosthetics. Even for applications that focus on inclu- sion, as in digital adapted sport games, 3D avatars with limb differences are not accu- rately represented and often rely on pros- thetic models. Mindscape 2 allows players to create 3D avatars that use wheelchairs or prosthetic devices as visible forms of inclu- sion. The Pegasus Dream Tour 3 allows play- ers to take part in virtual sports events using 3D avatars with prosthetics and adaptive de- vices. Another example is ParaJecripe [6], a game developed to raise awareness of adapted sports. It includes 3D avatars based on real Paralympic athletes, some of whom have limb differences and do not rely on prosthetics, but customization is limited to clothing and ac- cessories. As shown in Figure 1, ParaJecripe features avatars inspired by real Paralympic athletes, some of whom have limb differences, and are represented without prosthetics. Only a few 3D avatar systems support people with limb differences and amputees without prosthetics, offering limited options for appearance customization.Stracke et al. [7] present a prototype for a virtual reality simulation of lower limb amputation, where 2 Mindscape,2024.Availableat: https://paralympics.org.uk/articles/ mindscape-by-paralympicsgb-and-deloitte/. Accessed February 9th, 2026. 3 The Pegasus Dream Tour, 2021. Available at: https://pegasus-dream.com/. Accessed February 9th, 2026. 5 Figure 1: Parajecripe modalities for adapted sports [6]. tracked movements are shown through a 3D avatar, whose template is created with the Ready Player Me system. The customiza- tion options are limited to choosing gender, height, and amputation type, which must be single-sided and either above or below the knee. The Computer Assisted Limb Assess- ment (CALA) system by Prahm et al. [8], building on MakeHuman and Blender, in- troduces continuous sliders for altering in- dividual body segments such as the hands, forearms, upper arms, and shoulders. These sliders allow extreme and asymmetric mod- ifications, ranging from thin to thick limbs, as well as complete upper limb amputation. This system partially follows the guidelines for disability representation in 3D avatars, especially regarding the recommendation for granular continuous control over the repre- sentation of physical differences. However, the process is not fully automated and still re- quires some level of specialized knowledge to manually adjust multiple parameters through a 3D modeling interface, such as navigating camera angles and manipulating meshes. The literature shows that existing ap- proaches tend to fall into two main directions: (1) medical applications, where customiza- tion focuses on accurately modeling disabil- ities for clinical or rehabilitative purposes, and (2) self-representation systems, which emphasize visual customization of 3D avatar appearance but often overlook the diversity of disabilities, typically relying on prosthetic additions rather than structural variation. This divide underscores the lack of methods that integrate both anatomical accuracy and expressive customization, thus not fulfilling the inclusive 3D avatar guidelines. In the following, we review some AI-based methods that aim to address these limitations. Artificial Intelligence and Mor- phological Editting Current AI-based 3D avatar generation frameworks refine models by rendering them from multiple camera viewpoints and us- ing diffusion models to assess and improve realism.Diffusion models are a genera- tive method that constructs data by itera- tively denoising a random signal, approxi- mating the data distribution through a re- verse stochastic process learned via neu- ral networks.Parametric models such as Skinned Multi-Person Linear Model (SMPL) and SMPL eXpressive (SMPL-X) [9] guide this process, providing a structured represen- tation of the human body through parame- ters controlling pose, shape, and sometimes facial or muscle details. Although these mod- els ensure anatomical consistency, their de- sign assumes normative bodies, hindering the representation of limb differences or signifi- cant asymmetries. To address this limitation, recent research has focused on modifying parametric models to handle non-normative 6 bodies. Cho et al.propose the Amputated Joint Aware 3D Human Mesh Recovery (AJAHR) [10], which reconstructs 3D hu- man meshes from images while accounting for missing limbs. The method uses the SMPL model kinematic tree without modifying pre- trained parameters. Amputations are repre- sented by setting the pose values of the af- fected joint and its descendants to zero, caus- ing the associated vertices to collapse near the joint and simulate limb absence. A joint regressor then processes these vertices to pro- duce anatomically consistent 3D joint posi- tions that reflect the hierarchical effects of missing limbs. A Vision Transformer predicts the pose values from images, with separate branches for classifying the amputation sta- tus of each major limb. Global body rotation, shape, and camera position are predicted in- dependently and combined with pose values to reconstruct the final mesh. Figure 2 shows the 3D meshes extracted from photographs of individuals with limb differences. Figure 2: 3D mesh recovery based on images by AJAHR [10], licensed under C BY-SA 4.0. Yamada et al. propose DreamAble [11], a framework that generates 3D avatars with upper limb differences from natural language descriptions. The authors justify this focus by observing that upper limb prosthetics are less standardized, more context-dependent, and less commonly used than lower limb pros- thetics. DreamAble allows users to specify not only their limb differences but also their appearance, clothing, and style. Instead of relying on image input or rigid-body parame- ters, it interprets textual descriptions of limb differences and adapts the 3D avatar accord- ingly. The method uses the SMPL-X model kinematic tree to remove the corresponding keypoints from an OpenPose representation of the skeleton, which is later used for skele- tal guidance for a diffusion model. The au- thors introduce a skeleton-aware loss func- tion that penalizes anatomical inconsisten- cies by comparing generated 3D avatar sil- houettes against the adapted skeleton struc- ture. Figure 3 illustrates results produced by DreamAble. Figure 3a shows the generated 3D avatars in a canonical pose. Figures 3b and 3c provide close-up views, highlighting limb termination and cloth positioning de- tails, respectively. Both methods take meaningful steps to- ward more inclusive 3D avatar generation, but address the inclusive guidelines in dif- ferent ways. AJAHR focuses on anatomical accuracy, offering a technically precise recon- struction of limb differences through a mod- ified joint representation.However, it re- mains limited to reconstructing existing hu- man meshes and does not provide appear- ance customization. DreamAble allows users 7 (a) Full-body avatars in canonical pose. (b) Rendering of limb termination. (c) Adaptive handling of cloth around the limb. Figure 3: 3D avatars generated by Dream- Able [11]. to describe their limb differences and appear- ance through natural language, but it is lim- ited to upper limb differences.A shared limitation of both approaches is the lack of continuous control over limb length, as they only support limb differences resembling joint-level amputations. These approaches also overlook user-centered validation and community engagement. Emerging Challenges and Op- portunities Advances in 3D modeling, animation, and AI are enabling developers to generate 3D avatars that are increasingly diverse and foster inclusive self-representation. Recent guidelines emphasize the importance of cus- tomizable 3D avatars that can accurately re- flect the physical characteristics of diverse users. In the following, we discuss emerging trends and outline key open problems that must be addressed in the field. • Clothing and appearance cus- tomization: For individuals with limb differences, realistic clothing simulation is essential, as garments must drape naturally over the limb. Customization should extend to prosthetics, letting users modify or design assistive devices and save different versions of the same avatar. AI-based 3D avatar generation approaches,especiallytext-prompt- based methods, represent a promising path to provide flexible and accessible customization. • Limitation in rigid parametric models: Since parametric body mod- els are built upon fixed skeletal joints and hierarchical connections, current AI- based systems [10, 11] are limited in rep- resenting limb differences that deviate from conventional joint-level structures. Addressing this limitation requires ei- ther moving away from rigid paramet- ric structures or extending them to han- dle variable limb configurations through editing mechanisms. • Editing and representing congeni- tal conditions: Current AI-based sys- tems are limited by their inability to 8 model the anatomical diversity of con- genital limb conditions that may occur at the ends of limbs. Modeling these complex variations requires the integra- tion of multiple specialized classifica- tion systems, such as the Oberg-Manske- Tonkin system for the upper limbs, dif- ferent types of macrodactyly, and ra- dial longitudinal deficiency. Future re- search could adopt a two-step method, where the first stage employs AI-based 3D avatar generation to create a rough body model that captures limb differ- ences up to the anatomical level where the congenital variation occurs. The sec- ond stage would then refine the gran- ular details around the ends of the limbs through an editing tool similar to the CALA system [8]. This refinement process could be further automated by learning from diverse datasets of limb variations. • Challenges in motion capture and rigging: Creating adaptive skeletons and rigs that accommodate diverse anatomies while preserving realistic mo- tion remains challenging. Current mo- tion datasets, such as Motion-X [12], are designed for normative body types, revealing a gap in tools for animating non-normative bodies. In particular, for lower limb differences, directly applying motions from existing datasets can pro- duce uncanny results due to shifts in bal- ance caused by asymmetry.Address- ing this challenge will require integrating parameterized body models and motion retargeting methods that adapt to dif- ferences in limb length, joint structure, and mobility. Future research should de- velop motion datasets that specifically target people with limb differences. Be- yond applications in AI-based 3D avatar animation, such datasets could provide benchmarks for non-normative motion and facilitate the study of compensatory strategies. • Addressing visual data scarcity and diversity: Real-world datasets on limb differences remain limited and inconsis- tent, making it difficult to learn from few and unbalanced samples. Advances in the generation of data can help expand data diversity and enhance model gener- alization. For example, Cho et al. [10] in- troduce A3D, a synthetic dataset of var- ied amputee poses designed primarily for training image-based generative models, such as diffusion models, to create realis- tic images of non-normative bodies. Un- like motion datasets, which encode tem- poral dynamics and skeletal movements, datasets like A3D focus on visual repre- sentation, supporting tasks such as neu- ral rendering for 3D avatar generation and appearance editing. • Standardization, ethical guidelines, and user-centered validation: Cur- rent research on inclusive 3D avatar gen- eration lacks standardized protocols for evaluation, making comparisons across studies difficult.Most works rely on subjective assessments of visual quality 9 or anatomical plausibility, overlooking measures of inclusivity and representa- tion accuracy. To address this gap, re- search should incorporate feedback from users, particularly those with disabili- ties.This feedback can be achieved by establishing participatory design in AI-based frameworks, implementing us- ability testing protocols, and validating methods, while also adhering to guide- lines established in the literature. • Integration in XR and sports simu- lations: Incorporating 3D avatars that account for diverse anatomies into de- ployed XR experiences presents inten- sified versions of the challenges dis- cussed earlier. For instance, when de- veloping digital adapted sport games, prosthetic variability, motion adapta- tion, and scarcity of data representing diverse disabilities become even more pronounced. As upper limb prosthetics vary widely in form and function, mo- tion capture and animation should be adapted to each sport modality. Chal- lenges in training data also arise, as dif- fusion models struggle in sports where limbs are obscured by water or present motion blur. If any of these aspects are not properly addressed, the digital rep- resentation cannot accurately reflect the original event, undermining the goal of inclusivity. The scarcity of media on Paralympic athletes amplifies these dif- ficulties, exposing ethical considerations around privacy, consent, and fair repre- sentation. Conclusion Although current AI-based 3D avatar gen- eration approaches face considerable techni- cal and ethical challenges, ongoing innova- tions offer promising means to enable respect- ful and representative avatars. Recent ad- vances in the field permit users with joint- level or limb-specific variations to customize their 3D avatars to accurately reflect their physical characteristics. However, users who do not conform to normative body structures or have complex congenital conditions may encounter limitations due to rigid kinematic tree of parametric models. Future research should prioritize the de- velopment of adaptive editing tools that al- low users to customize their avatars to re- flect their preferences. Integrating principles of responsible AI development is also impor- tant to mitigate biases and prevent misrep- resentation. Participatory design involving users with disabilities is essential for validat- ing avatar systems and ensuring they meet diverse needs. Advancing this area requires close collabo- ration between technical development and an understanding of human experience. A cen- tral question that remains is whether existing AI-based 3D modeling workflows designed for normative bodies can be adapted to enable genuine individual customization, or whether entirely new paradigms must be developed to achieve true inclusivity. 10 Takeaway and Future Directions Recent advances in AI-based approaches show that inclusive 3D avatar design for peo- ple with limb differences is no longer a distant goal. The next priority for the community is to move beyond proof-of-concept systems and commit to delivering deployed XR applica- tions. As a first step to achieve this goal, re- searchers should develop methods that jointly address anatomical accuracy and expressive customization, while also establishing stan- dardized evaluation protocols that incorpo- rate user-centered validation. The next step is for developers of XR applications to ex- tend the animation and rigging pipelines to include non-normative anatomies.The fi- nal step is for XR platforms to take an ac- tive role in enhancing the visibility of in- dividuals with limb differences, particularly in adapted sports simulations, and to part- ner with Paralympic athletes and disabil- ity organizations. Table 1 summarizes how each group can adopt inclusive practices and guidelines. Acknowledgments This Project was conducted with the sup- port of the Industrial Technology Innovation Program (20023347 Development of Graph- based Intelligent Metaverse Engine for Im- mersive Content-sharing Service) funded by the Ministry of Trade, Industry & Energy of the Republic of Korea. References [1] T. Golando, “Nerf This!Navigating the Accessibility and Inclusivity of Video Games Through Expressive Arts Thera- pies: A Literature Review,” Expressive Therapies Capstone Theses, 2023. [2] J. Shell, “What Do We See: An Inves- tigation Into the representation of dis- ability in video games,” arXiv preprint arXiv:2103.17100, 2021. [3] R. Handley, B. Guerra, R. Goli, and D. Zytko, “Designing social VR: a col- lection of design choices across commer- cial and research applications,” arXiv preprint arXiv:2201.02253, 2022. [4] K. Zhang,E. G. S. Spencer Jr, A. Manikandan, A. Li, A. Li, Y. Yao, and Y. Zhao, “Inclusive avatar guide- lines for people with disabilities: Sup- porting disability representation in so- cial virtual reality,” in Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems, 2025, p. 1–26. [5] J. Park and S. Kim, “How do people with physical disabilities want to con- struct virtual identities with avatars?” Frontiers in Psychology, vol. 13, p. 1– 13, 2022. [6] G. B. Domingos, A. L. Brand ̃ao, A. G. Szykman, and J. P. Gois, “Production and post-production phases of the game scrum for the development of an adapted sports digital game,” in Proceedings of 11 the XV Brazilian Symposium on Games and Digital Entertainment, SBGAMES, vol. 17, 2017, p. 782–788. [7] L. Stracke, S. Burbach, R. Jakob, R. Br ̈uck, and T. J. Eiler, “Development of a Standalone VR Application for the Simulation of Lower Extremity Ampu- tations,” in International Conference on e-Health and Bioengineering. Springer, 2023, p. 426–433. [8] C. Prahm, K. Bauer, A. Sturma, L. Hruby, A. Pittermann, and O. Asz- mann, “3D body image perception and pain visualization tool for upper limb amputees,” in 2019 IEEE 7th Inter- national Conference on Serious Games and Applications for Health (SeGAH). IEEE, 2019, p. 1–5. [9] G. Pavlakos, V. Choutas, N. Ghorbani, T. Bolkart, A. A. Osman, D. Tzionas, and M. J. Black, “Expressive body cap- ture: 3D hands, face, and body from a single image,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2019, p. 10 975–10 985. [10] H. Cho, G. Choi, and J. Choi, “AJAHR: Amputated Joint Aware 3D Human Mesh Recovery,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2025, p. 7925–7935. [11] F. M. Yamada, J. P. Gois, and H. Taka- hashi, “A Framework for Text-Guided 3D Avatar Generation with Upper Limb Differences,” in IEEE Interna- tional Conference on Artificial Intelli- gence and Extended Reality (AIxVR), 2026. [12] J. Lin, A. Zeng, S. Lu, Y. Cai, R. Zhang, H. Wang, and L. Zhang, “Motion-X: A large-scale 3D expressive whole-body human motion dataset,” Advances in Neural Information Processing Systems, vol. 36, 2024. 12 CommunityRecommended Actions Researchers • Expand synthetic image datasets like A3D to broader disability profiles to address data scarcity • Extend or replace rigid 3D parametric body models to represent people with limb differences • Build motion capture datasets that feature people with limb differences for automatic rigging • Incorporate localized fine-grained modification of limbs for representing congenital conditions • Develop evaluation protocols that measure representation accuracy, not visual quality alone • Establish shared benchmarks for inclusive 3D avatar generation for comparison across methods • Embed participatory design with people with limb differences to ensure respectful representation Developers • Introduce AI-based 3D avatar generation to foster self-representation of people with limb differences • Build interfaces to edit limb differences in 3D avatars and save versions with or without prosthetics • Extend animation pipelines to account for variation in limb length and mobility range • Ensure clothing simulations handle natural draping for 3D avatars with limb differ- ences XR Plat- forms • Develop adapted sports games as both entertainment and awareness tools for general audiences • Partner with Paralympic athletes and disability organizations to ensure respectful representation • Collaborate with researchers to fund the production of motion capture data in adapted sports • Govern adapted sport content with policies on consent, privacy, and fair representa- tion • Treat incomplete self-representation as an accessibility failure, not a secondary com- pliance issue Table 1: Actions to support inclusive 3D avatar design across research, development, and XR platforms. 13