The foundation: a multi-layered AI architecture with state prediction & real-time control
- A lightweight latent world model for next-physical-state prediction, paired with Craftnet, our VLA model for trajectory control combined with a System 0 for real-time, tactile-based grasp control, power North's autonomous work.
- CraftNet, our "reflex system" for contact-rich manipulation, gives the hand precise, continuously-adjusted force control: enough pressure to shape and hold the ice cream, but not enough to crush the cup.
- Critically, DQ's signature Blizzard upside-down flip requires tactile-driven control that no teleoperation device can achieve today — North has to operate autonomously to perform the move.
A meaningful step toward making robots productive in changing environments
- The same underlying model handles a cup, a spoon, a machine handle, an autoclave door, and metal rings.
- The model adapts to real-world variability — different patterns of customer movement, deformable cups that change shape under grip, changing lighting, and more.
- North represents a meaningful step toward robotic generalization, pointing to a future where robots can complete full shifts across a wide range of real-world service environments.
Where to find your Sharpa North-made ice-cream
- Location: DQ store, Shanghai
- Timeline: Launching in August
Media & partnership enquiries: pr@sharpa.com