Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Sari Sandbox: A Virtual Retail Store Environment for Embodied AI Agents

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Sari Sandbox is a photorealistic virtual convenience store with 250 interactive items, three layouts, a Python API, and a dataset of human VR demonstrations, built to benchmark embodied AI shopping agents directly against human shoppers.

desk verdict A genuinely useful retail simulator fills a real gap, but the manuscript as provided is missing the benchmark results and dataset details that its abstract promises, so the central claim is unverified. read the letter →

arxiv 2508.00400 v1 pith:6G5IUKBB submitted 2025-08-01 cs.CV

Janika Deborah Gajo , Gerarld Paul Merales , Jerome Escarcha , Brenden Ashley Molina , Gian Nartea , Emmanuel G. Maminta , Juan Carlos Roldan , Rowel O. Atienza This is my paper
classification cs.CV
keywords embodiedAIretailsimulationvirtualrealityvision-languagemodelbenchmarkinggrocerystorehumandemonstrationsinteractiveenvironment
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper introduces Sari Sandbox, a photorealistic simulation of small convenience stores built to give embodied AI agents a place to shop. The environment contains over 250 interactive grocery items across three store layouts, a Python API for agent control, and virtual-reality support for human participants. Alongside it, the paper publishes SariBench, a benchmark of shopping tasks at three difficulty levels with annotated human demonstrations, and reports a baseline vision-language-model agent that completes the easy tasks without fine-tuning. The aim is to let researchers compare AI shopping agents directly with human shoppers under controlled, repeatable conditions.

What carries the argument

The central object is the Sari Sandbox environment itself and its paired benchmark SariBench. The environment is built on a real-time game engine with photorealistic 3D product models, rigid-body physics with box colliders, level-of-detail and culling for performance, and procedurally generated expiration dates and price tags. Interaction is mediated by two interfaces: a Python API exposing navigation and manipulation actions to agents, and a VR hand-interaction system with teleportation and haptic feedback for humans. SariBench defines tasks at three difficulty levels (easy, average, difficult) that combine perception, navigation, manipulation, memory, decision-making, and comprehension, and records human VR demonstrations as the baseline. An embodied agent loops through associative learning, planning, and execution, using a large vision-language model to generate action sequences, an OCR module to read labels, and a depth estimator for distance estimation.

What would settle it

Run the identical SariBench tasks with human participants in a physical convenience store arranged to match one of the three layouts, with the same product set and price labels, and compare task completion times, error rates, and search paths against the VR human demonstrations; if physical-store performance diverges sharply from VR performance, the simulation's difficulty calibration does not transfer to real shopping.

Watch

Extended reading notes

Core claim

Sari Sandbox is a high-fidelity retail simulation that closes a gap: existing embodied-agent simulators focus on household navigation and manipulation, not grocery shopping. The environment includes 250 interactable products in 11 categories, three store layouts based on surveys of real local stores, dynamic price and expiration labels, functional self-checkout with ray-cast barcode scanning, and both API and VR interfaces. SariBench records human demonstrations of easy, average, and difficult shopping tasks, and a baseline vision-language-model agent using a reasoning-and-acting loop with semantic and episodic memory completes the easy tasks, providing a first numeric comparison between human and agent performance in a retail setting.

Load-bearing premise

The benchmark's external validity depends on the assumption that simplified simulated interactions—rigid-body products, box colliders, ray-cast scanning, and teleport-based VR navigation—do not change the difficulty or nature of shopping tasks compared with physical stores; the paper does not validate this against real-world shopping behavior.

Editorial extensions

If this is right

  • Embodied agents can now be evaluated on retail shopping tasks against human demonstrations recorded in the same simulated store, removing the need to run physical stores for benchmarking.
  • The three store layouts and randomized product placement allow controlled experiments on how layout and organization affect search, navigation, and decision-making.
  • SariBench's graded task difficulties give agent developers a progression from simple pick-and-place to comparative reasoning tasks that require reading labels and making choices.
  • The off-the-shelf vision-language agent baseline shows that easy retail tasks are already tractable without fine-tuning, giving future work a concrete performance bar to beat.
  • Performance profiling of the environment identifies rendering and physics bottlenecks, informing optimization targets such as stable frame rates on mid-range hardware.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the benchmark holds, it could be used to study how shelf layout and product grouping affect search times in a controlled way, extending to retail design questions the paper only mentions.
  • The human thought-process flowcharts suggest a template for hybrid agent architectures that interleave reasoning and action; the paper extracts them but does not formalize them into the agent.
  • The ray-cast barcode scanning with orientation checks simulates a real perceptual skill that could transfer to physical robot checkout, but transfer is untested.
  • A natural extension is to add dynamic store states such as restocking, misplaced items, and crowds to test robustness; the paper lists dynamic additions as future work.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper presents Sari Sandbox, a Unity-based photorealistic virtual retail environment with over 250 interactive grocery items across three store layouts, a Python API for agent control, VR-based human interaction, and SariBench, a proposed benchmark of human demonstrations of shopping tasks of varying difficulty. It also describes a VLM-powered embodied agent that uses off-the-shelf components (Gemini 2.5 Pro, PaddleOCR, Depth-Anything-V2) for easy retrieval tasks. The abstract and introduction promise quantitative benchmarks and human-agent comparisons, but the visible manuscript text does not contain a Section 4 with results; the main text jumps from human thought-process flowcharts to the conclusion, and the conclusion defers benchmarking across difficulties to future work.

Significance. If the environment and benchmark were fully documented with results, the contribution would fill a genuine gap in retail-specific embodied AI simulators and provide a useful testbed for comparing AI agents to human shoppers. Strengths include the detailed environment design (product variety, barcode scanning mechanism, VR comfort features), the provision of an open-source code repository, and the use of off-the-shelf models without fine-tuning, which makes the agent baseline cheap to reproduce. The central claim, however, is not currently supported because the promised benchmark results and dataset statistics are absent.

major comments (3)
  1. [Section 4 / Introduction (last paragraph)] The introduction promises 'performance profiling and comparisons between humans and agents' and the abstract claims that the paper 'concludes with benchmarks, performance analysis,' but no Section 4 appears in the manuscript. There are no quantitative results: no task completion rates, no time-to-completion statistics, no learning curves, and no human-agent comparison metrics. Since the central claim of the paper is that Sari Sandbox enables benchmarking embodied agents against human performance, this omission is load-bearing and must be addressed before the paper can be considered complete.
  2. [Section 3.5 (SariBench)] SariBench is introduced as 'a dataset of annotated human demonstrations,' but the manuscript reports no dataset statistics: number of recruited participants, number of demonstrations per task, distribution across the three difficulty levels, annotation schema, or dataset release location. The only supporting material is a set of qualitative flowcharts (Figures 13 and 14) of human thought processes. Without quantitative description, the dataset cannot be assessed for coverage, balance, or usability as a benchmark baseline.
  3. [Section 5 (Conclusion) and Supplementary S2] The embodied agent presented in the supplementary material is explicitly restricted to 'easy' SariBench tasks and excludes checkout and multi-item manipulation, and Section 5 lists 'broadening embodied agent evaluation across all SariBench difficulties' as future work. This contradicts the abstract's statement that the sandbox 'provides baselines against human performance,' because no human comparison on the full benchmark is reported. The paper should either report the promised human-agent comparisons or clearly reposition its claims to describe an environment-and-dataset contribution without stating that the benchmark has been executed.
minor comments (5)
  1. [Section 1] The sentence 'These environments enables us to systematically test hypotheses' has a subject-verb agreement error; change 'enables' to 'enable'.
  2. [Section 3.5] The phrase 'no existing studies have established a baseline for such scenarios' is grammatically awkward; consider rewording.
  3. [Supplementary S1] The supplementary text repeatedly references 'Sec. 4.2' (e.g., 'As discussed in Sec. 4.2'), but the main text has no Section 4; this cross-reference is inconsistent with the visible structure.
  4. [References] Reference [37] lists the author as 'P. Authors' with a placeholder; the actual PaddleOCR author list should be provided.
  5. [Figures 13 and 14] The flowcharts labeled 'General thought process for average tasks' and 'General thought process for difficult tasks' are inserted between Section 3.5 and Section 5 without any accompanying discussion in the visible text; either add a section analyzing these flowcharts or move them into the supplementary material.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the missing performance results are an evidence gap, not a circular derivation.

full rationale

The paper does not contain a derivation chain that reduces to its own inputs. SariBench human demonstrations are collected from volunteer VR participants independently of the agent and of any fitted parameters, so they serve as an external baseline rather than a constructed prediction. The embodied agent relies on off-the-shelf models (Gemini 2.5 Pro, PaddleOCR, Depth-Anything-V2) without fine-tuning, meaning its behavior is not statistically forced by the environment's design. The environment's design choices are grounded in real-store surveys and external references, not in the benchmark outcomes. The only same-author citation, Atienza et al. VRex [35], supports generic VR comfort features alongside an external citation [36] and is not load-bearing for the paper's central claims. No self-definitional equation, no fitted-input-called-prediction step, and no imported uniqueness theorem appears. The paper's substantive weakness is evidential rather than circular: the main-text Section 4 performance analysis and quantitative SariBench statistics are absent, and the conclusion admits that embodied-agent evaluation across all SariBench difficulties is future work. Missing results affect verification and completeness, but they do not make the presented environment or benchmark logically circular.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

No free parameters or invented entities appear in this environment paper. The axioms listed are domain assumptions about the fidelity and transferability of the simulation, which the paper does not experimentally validate.

assumptions (4)
  • domain assumption Unity's rigid-body physics and box colliders adequately simulate product manipulation for benchmark purposes.
    Products are simulated as rigid bodies with only box colliders (Section 3.3), which may not capture deformable or fine-grained interactions, but is assumed sufficient for the benchmark tasks.
  • domain assumption Off-the-shelf Gemini 2.5 Pro, PaddleOCR, and Depth-Anything-V2 provide sufficient perception and reasoning in the simulated store.
    The agent design in Supplementary S2 and S3 relies on these external tools' performance in synthetic scenes, with no fine-tuning to the environment.
  • domain assumption Human VR task performance is a meaningful baseline for embodied agents.
    The benchmark compares human VR demonstrations with agent performance, assuming the VR interface does not distort task difficulty (e.g., tunneling vignette, teleportation, simplified hand controls).
  • domain assumption The three store layouts are representative of real local convenience stores.
    Layouts are based on surveys (Section 3.3), but no quantitative validation of representativeness is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Sari Sandbox: A Virtual Retail Store Environment for Embodied AI Agents." pith.science (2026). https://pith.science/paper/6G5IUKBB

@misc{pith2026250800400,
  author       = {Pith},
  title        = {Pith review of: Sari Sandbox: A Virtual Retail Store Environment for Embodied AI Agents},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6G5IUKBB}},
  note         = {Machine review of arXiv:2508.00400}
}
read the original abstract

We present Sari Sandbox, a high-fidelity, photorealistic 3D retail store simulation for benchmarking embodied agents against human performance in shopping tasks. Addressing a gap in retail-specific sim environments for embodied agent training, Sari Sandbox features over 250 interactive grocery items across three store configurations, controlled via an API. It supports both virtual reality (VR) for human interaction and a vision language model (VLM)-powered embodied agent. We also introduce SariBench, a dataset of annotated human demonstrations across varied task difficulties. Our sandbox enables embodied agents to navigate, inspect, and manipulate retail items, providing baselines against human performance. We conclude with benchmarks, performance analysis, and recommendations for enhancing realism and scalability. The source code can be accessed via https://github.com/upeee/sari-sandbox-env.

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. RetailSMV: Exocentric vs. Egocentric Adaptation of Foundation Video World Models in Retail

    cs.CV 2026-07 unverdicted novelty 5.0 of 10

    Exocentric-only LoRA adaptation of Cosmos3-Nano on a new synchronized retail video dataset matches or exceeds combined ego+exo training on most held-out metrics.

Reference graph

Works this paper leans on

44 extracted references · 34 canonical work pages · cited by 1 Pith paper

  1. [1]

    Deep understanding of shopper behaviours and interactions using rgb-d vision,

    M. Paolanti, R. Pietrini, A. Mancini, E. Frontoni, and P. Zingaretti, “Deep understanding of shopper behaviours and interactions using rgb-d vision,” Machine Vision and Applications, vol. 31, no. 7, p. 66, 2020. [Online]. Available: 8 https://doi.org/10.1007/s00138-020-01118-w 2

    work page doi:10.1007/s00138-020-01118-w 2020

  2. [2]

    Kroger and nvidia to reinvent the shopping experience through state-of-the-art, ai-enabled applications and services,

    The Kroger Co. and NVIDIA Corporation, “Kroger and nvidia to reinvent the shopping experience through state-of-the-art, ai-enabled applications and services,” Press release, NVIDIA Newsroom, Mar. 2022, describes a joint AI lab and demonstration center using NVIDIA Omniverse digital twins at Kroger HQ in Cincinnati. [Online]. Available: https://nvidianews....

    work page 2022

  3. [3]

    Store digital twin: Giving associates “superpowers

    Lowe’s Innovation Labs, “Store digital twin: Giving associates “superpowers” to better serve customers,” Project page, Lowe’s Innovation Labs website, Sep. 2022, developed with NVIDIA Omniverse and Magic Leap 2; live in two pilot stores; includes AR restocking support, X-ray vision, and in-store simulation tools. [Online]. Available: http://lowesinnovatio...

    work page 2022

  4. [4]

    A Study on the Challenges of Using Robotics Simulators for Testing

    A. Afzal, D. S. Katz, C. Le Goues, and C. S. Timperley, “A study on the challenges of using robotics simulators for testing,” arXiv preprint arXiv:2004.07368 , Apr. 2020, survey of 82 robotics developers, identifying 10 major barriers to simulation use in testing and CI pipelines. [Online]. Available: https://arxiv.org/abs/2004.07368 2

    work page Pith review arXiv 2004

  5. [5]

    Emerging trends in realistic robotic simulations: A comprehensive systematic literature review,

    S. M. Kargar, B. Yordanov, C. Harvey, and A. Asadipour, “Emerging trends in realistic robotic simulations: A comprehensive systematic literature review,” IEEE Ac- cess, vol. 12, pp. 1–26, May 2024, systematic review of ROS-enabled simulators, game-engine platforms, and AI-enhanced realistic scenario replication. [Online]. Avail- able: https://ieeexplore.i...

    work page arXiv 2024

  6. [6]

    Future Convenience Store Challenge 2024 – Post-Event Report,

    World Robot Summit Executive Committee, “Future Convenience Store Challenge 2024 – Post-Event Report,” INTEX Osaka, November 13–15, 2024, 2024, includes online rulebook and Gazebo models. [Online]. Available: https://worldrobotsummit.org/en/wrs2025/fcsc/ 2

    work page 2024

  7. [7]

    A survey of robotic navigation and manipulation with physics simulators in the era of embodied ai,

    L. H. K. Wong, X. Kang, K. Bai, and J. Zhang, “A survey of robotic navigation and manipulation with physics simulators in the era of embodied ai,”arXiv preprint arXiv:2505.01458,

    arXiv

  8. [8]

    Habi- tat 3.0: A co-habitat for humans, avatars and robots,

    X. Puig, E. Undersander, A. Szot, M. Dallaire C ˆot´e, T.-Y . Yang, R. Partsey, R. Desai, A. W. Clegg, M. Hlav ´aˇc, S. Y . Min, V . V ondruˇs, T. Gervet, V .-P. Berg `es, J. M. Turner, O. Maksymets, Z. Kira, M. Kalakrishnan, J. Malik, D. S. Chaplot, U. Jain, D. Batra, A. Rai, and R. Mottaghi, “Habi- tat 3.0: A co-habitat for humans, avatars and robots,” 2023. 2

    work page 2023

Show all 44 references
  1. [9]

    Ai2-thor: An interactive 3d environment for visual ai,

    E. Kolve, R. Mottaghi, W. Han, E. VanderBilt, L. Weihs, A. Herrasti, M. Deitke, K. Ehsani, D. Gordon, Y . Zhu, A. Kembhavi, A. Gupta, and A. Farhadi, “Ai2-thor: An interactive 3d environment for visual ai,” arXiv preprint arXiv:1712.05474, Dec. 2017, published Dec 14, 2017. 2

    2017 arXiv

  2. [10]

    Matter- port3d: Learning from rgb-d data in indoor environments,

    A. Chang, A. Dai, T. Funkhouser, M. Halber, M. Nießner, M. Savva, S. Song, A. Zeng, and Y . Zhang, “Matter- port3d: Learning from rgb-d data in indoor environments,” in Proceedings of the International Conference on 3D Vision (3DV). IEEE, Oct. 2017, pp. 667–676. 2

    2017

  3. [11]

    Threedworld: A platform for in- teractive multi-modal physical simulation,

    C. Gan, J. Schwartz, S. Alter, D. Mrowca, M. Schrimpf, J. Traer, J. De Freitas, J. Kubilius, A. Bhandwaldar, N. Haber, M. Sano, K. Kim, E. Wang, M. Lingelbach, A. Curtis, K. Feigelis, D. M. Bear, D. Gutfreund, D. Cox, A. Torralba, J. J. DiCarlo, J. B. Tenenbaum, J. H. McDer- m...

    2021

  4. [12]

    Future convenience store challenge (fcsc), world robot summit 2025,

    “Future convenience store challenge (fcsc), world robot summit 2025,” https://worldrobotsummit.org/en/wrs2025/ fcsc/fcsc2024/, World Robot Summit Executive Committee and METI, July 2025, competition stages held November 13–15, 2024 (INTEX Osaka) and July 13–19, 2025 (EXPO Cent...

    2025

  5. [13]

    Room- across-room: Multilingual vision-and-language navigation with dense spatiotemporal grounding,

    A. Ku, P. Anderson, R. Patel, E. Ie, and J. Baldridge, “Room- across-room: Multilingual vision-and-language navigation with dense spatiotemporal grounding,” in Proceedings of the 2020 Conference on Empirical Methods in Natural Lan- guage Processing (EMNLP), 2020, pp. 4392–4412. 2

    2020

  6. [14]

    Ion: Instance-level object navigation,

    W. Li, X. Song, Y . Bai, S. Zhang, and S. Jiang, “Ion: Instance-level object navigation,” in Proceedings of the 29th ACM international conference on multimedia , 2021, pp. 4343–4352. 2

    2021

  7. [15]

    Humanoidbench: Simulated humanoid benchmark for whole-body locomotion and manipulation,

    C. Sferrazza, D.-M. Huang, X. Lin, Y . Lee, and P. Abbeel, “Humanoidbench: Simulated humanoid benchmark for whole-body locomotion and manipulation,” arXiv preprint arXiv:2403.10506, 2024. 2

    2024 arXiv

  8. [16]

    Mujoco: A physics en- gine for model-based control,

    E. Todorov, T. Erez, and Y . Tassa, “Mujoco: A physics en- gine for model-based control,” in IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2012, pp. 5026–5033. 2

    2012

  9. [17]

    Alfred: A bench- mark for interpreting grounded instructions for everyday tasks,

    M. Shridhar, J. Thomason, D. Gordon, Y . Bisk, W. Han, R. Mottaghi, L. Zettlemoyer, and D. Fox, “Alfred: A bench- mark for interpreting grounded instructions for everyday tasks,” in Proceedings of the IEEE/CVF conference on com- puter vision and pattern recognition , 2020, pp....

    2020

  10. [18]

    Isaac sim,

    NVIDIA, “Isaac sim,” https://developer.nvidia.com/isaac/ sim, n.d. 2

  11. [19]

    Nvidia omniverse,

    ——, “Nvidia omniverse,” https://www.nvidia.com/en-us/ omniverse/, accessed: May 22, 2025. 2

    2025

  12. [20]

    igibson 1.0: a simulation environment for in- teractive tasks in large realistic scenes,

    B. Shen, F. Xia, C. Li, R. Mart ´ın-Mart´ın, L. Fan, G. Wang, C. P ´erez-D’Arpino, S. Buch, S. Srivastava, L. P. Tchapmi, M. E. Tchapmi, K. Vainio, J. Wong, L. Fei-Fei, and S. Savarese, “igibson 1.0: a simulation environment for in- teractive tasks in large realistic scenes,” ...

    2021

  13. [21]

    igibson 2.0: Object-centric simulation for robot learning of everyday household tasks,

    C. Li, F. Xia, R. Mart ´ın-Mart´ın, M. Lingelbach, S. Srivas- tava, B. Shen, K. Vainio, C. Gokmen, G. Dharan, T. Jain, A. Kurenkov, C. Liu, H. Gweon, J. Wu, L. Fei-Fei, and S. Savarese, “igibson 2.0: Object-centric simulation for robot learning of everyday household tasks,” in...

    2021

  14. [22]

    A survey of embodied ai: From simulators to research tasks,

    J. Duan, S. Yu, H. L. Tan, H. Zhu, and C. Tan, “A survey of embodied ai: From simulators to research tasks,” IEEE Transactions on Emerging Topics in Computational 9 Intelligence, vol. 6, no. 2, pp. 230–244, 2022, accessed: May 22, 2025. [Online]. Available: https://ieeexplore....

    2022

  15. [23]

    Visual lan- guage maps for robot navigation,

    C. Huang, O. Mees, A. Zeng, and W. Burgard, “Visual lan- guage maps for robot navigation,” in Proceedings of the IEEE International Conference on Robotics and Automation (ICRA), London, UK, 2023. 3

    2023

  16. [24]

    LM- nav: Robotic navigation with large pre-trained models of language, vision, and action,

    D. Shah, B. Osinski, B. Ichter, and S. Levine, “LM- nav: Robotic navigation with large pre-trained models of language, vision, and action,” in 6th Annual Conference on Robot Learning , 2022. [Online]. Available: https: //openreview.net/forum?id=UW5A3SweAH 3

    2022

  17. [25]

    Advances and challenges in foundation agents: From brain-inspired intelligence to evolutionary, collaborative, and safe systems,

    B. Liu, X. Li, J. Zhang, J. Wang, T. He, S. Hong, H. Liu, S. Zhang, K. Song, K. Zhu et al., “Advances and challenges in foundation agents: From brain-inspired intelligence to evolutionary, collaborative, and safe systems,”arXiv preprint arXiv:2504.01990, 2025. 3

    2025 arXiv

  18. [26]

    V oyager: An open-ended em- bodied agent with large language models,

    G. Wang, Y . Xie, Y . Jiang, A. Mandlekar, C. Xiao, Y . Zhu, L. Fan, and A. Anandkumar, “V oyager: An open-ended em- bodied agent with large language models,” arXiv preprint arXiv:2305.16291, 2023. 3

    2023 arXiv

  19. [27]

    Embodied agent interface: Benchmarking llms for embodied decision making,

    M. Li, S. Zhao, Q. Wang, K. Wang, Y . Zhou, S. Srivastava, C. Gokmen, T. Lee, E. L. Li, R. Zhang et al., “Embodied agent interface: Benchmarking llms for embodied decision making,” Advances in Neural Information Processing Sys- tems, vol. 37, pp. 100 428–100 534, 2024. 3

    2024

  20. [28]

    React: Synergizing reasoning and acting in language models,

    S. Yao, J. Zhao, D. Yu, N. Du, I. Shafran, K. Narasimhan, and Y . Cao, “React: Synergizing reasoning and acting in language models,” in International Conference on Learning Representations (ICLR), 2023. 3

    2023

  21. [29]

    Stateact: State tracking and rea- soning for acting and planning with large language models,

    N. Rozanov and M. Rei, “Stateact: State tracking and rea- soning for acting and planning with large language models,” arXiv preprint arXiv:2410.02810, 2024. 3

    2024 arXiv

  22. [30]

    Alfworld: Aligning text and embod- ied environments for interactive learning,

    M. Shridhar, X. Yuan, M.-A. C ˆot´e, Y . Bisk, A. Trischler, and M. Hausknecht, “Alfworld: Aligning text and embod- ied environments for interactive learning,” arXiv preprint arXiv:2010.03768, 2020. 3

    2010 arXiv

  23. [31]

    Cogni- tive architectures for language agents,

    T. Sumers, S. Yao, K. Narasimhan, and T. Griffiths, “Cogni- tive architectures for language agents,” Transactions on Ma- chine Learning Research, 2023. 3

    2023

  24. [32]

    A survey on the memory mecha- nism of large language model based agents,

    Z. Zhang, X. Bo, C. Ma, R. Li, X. Chen, Q. Dai, J. Zhu, Z. Dong, and J.-R. Wen, “A survey on the memory mecha- nism of large language model based agents,” arXiv preprint arXiv:2404.13501, 2024. 3

    2024 arXiv

  25. [33]

    Mem- orybank: Enhancing large language models with long-term memory,

    W. Zhong, L. Guo, Q. Gao, H. Ye, and Y . Wang, “Mem- orybank: Enhancing large language models with long-term memory,” in Proceedings of the AAAI Conference on Artifi- cial Intelligence, vol. 38, no. 17, 2024, pp. 19 724–19 731. 3

    2024

  26. [34]

    Ret- llm: Towards a general read-write memory for large lan- guage models,

    A. Modarressi, A. Imani, M. Fayyaz, and H. Sch ¨utze, “Ret- llm: Towards a general read-write memory for large lan- guage models,” arXiv preprint arXiv:2305.14322, 2023. 3

    2023 arXiv

  27. [35]

    Vrex: A framework for immersive virtual reality experiences,

    R. Atienza, R. Blonna, M. Tan, V . Tan, and A. Mora, “Vrex: A framework for immersive virtual reality experiences,” in 2018 IEEE Region Ten Symposium (TENSYMP), 07 2018. 5, 6

    2018

  28. [36]

    A testbed for studying cy- bersickness and its mitigation in immersive virtual reality,

    D. Calandra and F. Lamberti, “A testbed for studying cy- bersickness and its mitigation in immersive virtual reality,” IEEE transactions on visualization and computer graphics , vol. PP, 08 2024. 5

    2024

  29. [37]

    Paddleocr, awesome multilingual ocr toolkits based on paddlepaddle

    P. Authors, “Paddleocr, awesome multilingual ocr toolkits based on paddlepaddle.” https://github.com/PaddlePaddle/ PaddleOCR, 2020. 7, 1, 4

    2020

  30. [38]

    Benchmarking performance analysis of opti- cal character recognition techniques,

    M. A. N. Hadi, M. Gul, M. Khan, G. N. Alwakid, and N. Z. Jhanjhi, “Benchmarking performance analysis of opti- cal character recognition techniques,” in 2024 26th Interna- tional Multi-Topic Conference (INMIC) , Karachi, Pakistan, 2024, pp. 1–6. 7

    2024

  31. [39]

    A comparison study on op- tical character recognition models in mathematical equations and in any language,

    S. A. Francis and M. Sangeetha, “A comparison study on op- tical character recognition models in mathematical equations and in any language,” Results in Control and Optimization , vol. 18, p. 100532, 2025. 7

    2025

  32. [40]

    Norvig and S

    P. Norvig and S. J. Russell, Artificial Intelligence: A Modern Approach. Pearson, 2016. 1

    2016

  33. [41]

    Mmmu: A massive multi-discipline multimodal understanding and rea- soning benchmark for expert agi,

    X. Yue, Y . Ni, K. Zhang, T. Zheng, R. Liu, G. Zhang, S. Stevens, D. Jiang, W. Ren, Y . Sun et al. , “Mmmu: A massive multi-discipline multimodal understanding and rea- soning benchmark for expert agi,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern R...

    2024

  34. [42]

    Gemini: A family of highly capable multimodal models,

    G. Team, R. Anil et al. , “Gemini: A family of highly capable multimodal models,” 2025, arXiv preprint arXiv:2312.11805. [Online]. Available: https://arxiv.org/ abs/2312.11805 1

    2025 arXiv

  35. [43]

    A formal basis for the heuristic determination of minimum cost paths,

    P. E. Hart, N. J. Nilsson, and B. Raphael, “A formal basis for the heuristic determination of minimum cost paths,” IEEE transactions on Systems Science and Cybernetics , vol. 4, no. 2, pp. 100–107, 1968. 1

    1968

  36. [44]

    Depth anything v2,

    L. Yang, B. Kang, Z. Huang, Z. Zhao, X. Xu, J. Feng, and H. Zhao, “Depth anything v2,”Advances in Neural Informa- tion Processing Systems , vol. 37, pp. 21 875–21 911, 2024. 4 10 Sari Sandbox: A Virtual Retail Store Environment for Embodied AI Agents Supplementary Material S1....

    2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.