MIT's AromaGen turns text into smell: 71% blind-test accuracy, open-sourced
pliang279 · x · 2026-11-13
MIT Media Lab open-sourced AromaGen, a wearable olfactory interface mapping free-form text, speech, and images into timed scent sequences over a programmable odorant palette, with human-in-the-loop refinement. In blind tests across 50 substances in 12 categories, participants identified AI-generated smells at 71% accuracy—more than double the 33% chance baseline—with realism rated 4.65/7. Qualitative interviews showed users exploring abstract concepts like "catharsis" and recreating memories: one participant's "Sister's hug" scent brought her to tears. The repo includes the AI composition backend, BLE hardware stack, and user study data under an MIT license.
Related event: Stanford Team Unveils AromaGen: AI System for Real-Time Scent Transmission(6 posts)→
More from Embodied
- Apple details Secure Exclaves, a silicon architecture built for ambient AI sensing — BenBajarin · 2026-11-13
- Open robot learning book adds planning chapter covering CEM and loop control — GlenBerseth · 2026-11-13
- Gazebo community meeting: how OTTO by Rockwell validates industrial AMRs — ___Mufasaa · 2026-11-13
- 2.8 hours of inference for 30 seconds of walking: GPT-6 Astra as a robot policy tested — Willing_Amphibian825 · 2026-11-13
- Origami Robotics debuts Gen2 full direct-drive robot hand at CoRL dexterity night — viddivj · 2026-11-13
- Fable 5.2 hits 64.9% on RoboDojo; can frontier models replace dedicated robot policies? — Buffalo4516 · 2026-11-13