EssilorLuxottica's ARGO smart glasses run YOLOv11 on-device for real-time obstacle detection

Researchers from EssilorLuxottica's Smart Eyewear Lab and Politecnico di Milano have unveiled ARGO, a fully sensorized smart eyewear platform designed for on-device machine learning. Unlike cloud-dependent solutions, ARGO leverages the STM32N6 microcontroller's integrated Neural Processing Unit to minimize latency and preserve user privacy by processing all data locally. The platform's core innovation is the deployment of an optimized YOLOv11 model for real-time urban obstacle recognition, enhanced with a novel Head-wise Parallel Attention (HPA) architecture that ensures efficient execution on the NPU while maintaining the original computational logic.

ARGO integrates a multimodal sensor suite including RGB cameras, Time-of-Flight sensors, microphones, and ambient sensors. The model, trained on the Walking On The Road (WOTR) dataset, achieves an mAP50-95 of 24 under strict memory constraints with a footprint of just 2.483 MB. The system delivers 10 frames per second with a continuous autonomy of approximately 113 minutes on a 200 mAh battery. These results demonstrate that high-performance, privacy-preserving assistive devices are feasible, and highlight the growing need for tightly integrated, multidisciplinary co-design approaches in edge AI.