AARK NOVA Pick
MVP-4 detector bridge
Predict picking failure before robot motion.
Local-first intelligence for warehouse robots: real vision detection, target selection, grasp ranking, risk scoring, and robot plan preview before hardware moves.
Public showroom only. Production runtime is designed for no cloud dependency and customer-private data stays local.
Inputimage or sample scene
Decisionpick / ask human
World modelfuture prediction + value
Boundaryhardware control disabled
Try it: choose a demo scenario, then run the local decision loop to see generated actions, predicted futures, value scores, and the robot plan preview.
MVP-4 real vision bridge: upload a bin image to run the fine-tuned YOLO detector first, then send detected objects into the world-model decision loop. If the detector model is unavailable, the app falls back to clearly marked pseudo-vision.
Demo scenario
Product boundary: this demo does not control a real robot and does not use customer data. Customer footage, SKU data, calibration, robot control, and safety rules stay local/private in production.