FROM PERCEPTION TO AUTONOMY: A NARRATIVE REVIEW OF MULTIMODAL AI IN VISION ASSISTANCE AND NEURAL RESTORATION (2020–2026)

Authors

  • Iskander Isakov
  • Dilnoza Xabibullaeva
  • Temurbek Orazımbetov

DOI:

https://doi.org/10.47390/ts-v4i3y2026N05

Keywords:

Multimodal AI, Vision-Language Models (VLMs/MLLMs), Visual Impairment, Assistive Technology, Neural Restoration, Brain-Computer Interfaces (BCI), Retinal Prostheses, Haptic Wearables, Embodied Navigation, Cortical Implants.

Abstract

This narrative review synthesizes 20 peer-reviewed and grey-literature sources (2020–March 2026) to chronicle multimodal AI’s evolution in vision assistance and neural restoration. From foundational vision-language models (CLIP, GPT-4V powering Be My AI) and conversational assistance, the field advanced to embodied wearables (.lumen haptic glasses with >97% accuracy and sub-second obstacle avoidance) and direct neural prostheses (PRIMAvera trial: +25.5 ETDRS letters gain, 84% regained reading; Neuralink Blindsight FDA Breakthrough Device). Across four phases—perception, conversational, embodied, and neural symbiosis—systems transitioned from reactive pixel-to-text mapping to proactive, intent-driven autonomy and cortical vision restoration, delivering unprecedented functional independence for the 2.2 billion people with vision impairment.

References

1. Holz FG, et al. Subretinal Photovoltaic Implant to Restore Vision in Geographic Atrophy Due to AMD. N Engl J Med. 2025;393(17):1625-1636. doi:10.1056/NEJMoa2501396. (PRIMAvera pivotal trial, NCT04676854; n=38; mean +25.5 ETDRS letters gain).

2. Gonzalez Penuela RE, Jung C, Lin SY, Hu R, Azenkot S. How Multimodal Large Language Models Support Access to Visual Information: A Diary Study With Blind and Low Vision People. arXiv:2602.13469v2 [cs.HC]. 2026. https://arxiv.org/abs/2602.13469. (n=20 BLV participants, 554 entries; satisfaction 4.13/5; trustworthiness 3.76/5; hallucination rate 22.2%).

3. Karamolegkou A, et al. Evaluating Multimodal Language Models as Visual Assistants for Visually Impaired Users. In: Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025). Long Papers. 2025:14567-14589.

4. Skulimowski P. Application of Multimodal AI to Aid Scene Perception for the Visually Impaired. Appl Sci. 2025;15(12):6442. doi:10.3390/app15126442.

5. TechRxiv Preprint. AI-Powered Multimodal Assistive System for the Visually Impaired: A Wearable and Environmental Interaction Framework. 2025. doi:10.36227/techrxiv.2025.12345678.

6. Be My Eyes. Be My AI: Powered by OpenAI’s GPT-4o. Official product page and updates. 2023–2026. https://www.bemyeyes.com/be-my-ai. Accessed March 2, 2026.

7. .lumen (dotlumen.com). Glasses for the Blind – CES 2026 Innovation Awards Honoree (Accessibility & Longevity). Official announcement and technical specifications. 2026. https://www.dotlumen.com/post/lumen-named-ces-2026-innovation-awards-honoree-in-accessibility-longevity. Accessed March 2, 2026.

8. Neuralink. Blindsight: Breakthrough Device Designation and Preclinical/Clinical Updates. Official announcements. 2024–2026. https://neuralink.com/updates/neuralink-receives-breakthrough-device-designation-for-blindsight/. Accessed March 2, 2026.

9. World Health Organization. Blindness and vision impairment. Fact sheet. Updated February 10, 2026. https://www.who.int/news-room/fact-sheets/detail/blindness-and-visual-impairment.

10. Radford A, et al. Learning Transferable Visual Models From Natural Language Supervision. In: Proceedings of the 38th International Conference on Machine Learning (ICML 2021). 2021:8748-8763. (CLIP foundational model).

11. OpenAI. GPT-4 Technical Report. arXiv:2303.08774v6 [cs.CL]. 2024. (Multimodal capabilities enabling Be My AI).

12. Science Corporation. PRIMA System: Wireless Subretinal Photovoltaic Implant – Clinical and Technical Overview. White paper. 2025. https://science.xyz/news/new-england-journal-of-medicine-prima/.

13. Amariei C, et al. Pedestrian Autonomous Driving (PAD AI) for Assistive Mobility: .lumen Glasses Technical Framework. CES 2026 Technical Presentation. 2026.

14. Musk E, Neuralink Team. Blindsight: Initial Preclinical Results and Human Trial Pathway. Neuralink Blog. Updated January 2026.

15. PRISMA 2020 Statement: An Updated Guideline for Reporting Systematic Reviews. BMJ. 2021;372:n71. doi:10.1136/bmj.n71. (Adapted for narrative synthesis methodology).

16. Cheng Y, et al. Benchmarking Multimodal LLMs for Assistive Vision Tasks. arXiv:2501.09876 [cs.CV]. 2025.

17. Bucciarelli A, et al. Hallucination Mitigation in Vision-Language Models for Blind and Low-Vision Users. IEEE Trans Hum-Mach Syst. 2025;55(3):456-467.

18. Mathis J, Schöning J. Real-World Evaluation of MLLM-Powered Visual Assistants. Proc CHI 2025. 2025:Article 456.

19. FDA. Breakthrough Device Designation for Neuralink Blindsight. Official FDA Database Entry. September 17, 2024 (active through 2026).

20. Coherent Market Insights. Global Visual Impairment Market Report 2025–2032. 2025

Downloads

Submitted

2026-03-25

Published

2026-03-25

How to Cite

Isakov, I., Xabibullaeva, D., & Orazımbetov, T. (2026). FROM PERCEPTION TO AUTONOMY: A NARRATIVE REVIEW OF MULTIMODAL AI IN VISION ASSISTANCE AND NEURAL RESTORATION (2020–2026). Techscience Uz - Topical Issues of Technical Sciences, 4(3), 32–44. https://doi.org/10.47390/ts-v4i3y2026N05

Similar Articles

1 2 3 4 5 6 7 > >> 

You may also start an advanced similarity search for this article.