Wenn KI-Schlachten toben, hilft VR beim Sehen lernen
Wenn KI-Schlachten toben, hilft VR beim Sehen lernen Während die Tech‑Giganten um KI‑Geheimnisse und Talente ringen – OpenAI wirft einem Ex‑Mitarbeiter vor, 100+ Dateien an Apple weitergegeben zu haben – entsteht anderswo Innovation, die direkt Leben verändert. Ich bin Entwickler und Vater eines Kindes mit Amblyopie (lazy eye). Statt eines weiteren Pflasters, das das schwache Auge isoliert,…
As the tech giants battle for AI secrets and talents, innovation is emerging to directly change lives. A developer and father of a child with amblyopia (lazy eye) sought a solution that would allow both eyes to work together again. AmblyoPunch on Meta Quest, developed by Seven Sportz, uses dichoptic training in a 3D environment where each eye receives its own image.
In the game, Gabor pattern coins and red spikes are presented. The Gabor patches stimulate contrast and orientation processing in vision science. The app follows a four-step plan, starting from monocular warm-up (only the weak eye sees targets) through breaking suppression (darker dominant eye) to rebalancing and finally fusion training within Panum's fusion area.
The app runs in Unity 2022 LTS with OpenXR, rendering at 90 Hz per eye and keeping the end-to-end latency under 20 ms to keep the dichoptic stimuli synchronized. A small shader calculates the Gabor frequency in real-time while a background thread buffers success statistics for adaptive level progression. The user can input the weak eye (left/right), severity (mild/moderate/severe), and possible misalignment type (esotropia/exotropia) through the wall and handgrip menus, as an incorrect selection would reverse the clinical effect.
Progress is based on performance: ≥ 70% success in a sliding window of 10 trials (≥ 80% for strabismus) frees the user for the next level; under 40% the user is downgraded. Importantly, AmblyoPunch is a training/assistance game, not a medical device, and does not replace professional eye care. It complements therapy playfully and technically transparent.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.