HYPRLABS

HYPRLABS develops HYPR ACTV, an end-to-end learning architecture for autonomous machines. The system begins with paired camera observations, vehicle telemetry and human control actions, then uses supervised interventions and selected novel experiences to refine a shared perception-and-control model during real-world operation.

Its current proving platform is a two-vehicle test fleet equipped with five cameras and NVIDIA Orin AGX compute. HYPR reports a 20-minute autonomous urban drive in San Francisco after training on about 1,600 hours of distilled driving exposure, with 12 watts used for neural inference. The company is applying this runtime-learning pipeline toward a domain-general autonomy stack and plans to introduce its first purpose-built robot in 2026.