Project Oxygen & Ideo-LabIDEO LAB Dashboard 2026

Edge AI hors-ligne — du modùle au vol

Quantisation & optimisation ‱ Apprentissage hors-ligne ‱ Adaptation locale limitĂ©e ‱ Outils de dĂ©ploiement (TensorRT, OpenVINO, PyTorch Mobile, Edge Impulse).

Jetson/ARM/FPGA INT8 ‱ pruning ‱ distillation Few-shot ‱ continual (bornĂ©) MLOps edge ‱ HIL/SIL
Latence p→a (P95)< 80 ms
Pipelines zero-copy, batch=1, NMS GPU.
Budget puissance IA≀ 7–12 W
INT8, duty-cycle capteurs, clocks dynamiques.
Taille modùle< 30–80 MB
Pruning + distillation + ZIP/CRC signé.
Robustesse driftDrift < 1%/m
SLAM VIO tight-coupled, loop-closure.