Mobileye

Mobileye, founded in 1999 and headquartered in Jerusalem, Israel, develops perception and decision-making systems spanning the full stack from silicon to cloud. At the hardware layer, the company designs the EyeQ system-on-chip family - purpose-built for vision processing - with over 230 million units shipped globally through 2025, embedded in production vehicles from multiple OEMs. The hardware/software boundary is central to their approach: each EyeQ generation integrates accelerators tuned for neural-network inference on camera data, while higher-level planning and fusion run on paired application processors.

The software stack is layered around a camera-first perception pipeline augmented by radar and lidar fusion where required. Their REM (Road Experience Management) crowdsourced mapping system aggregates data from deployed vehicles to build and update a global HD map, closing the loop between fleet-scale sensing and localization. For autonomous driving, the Mobileye Drive platform implements a True Redundancy architecture that runs independent perception subsystems in parallel - one vision-centric, one lidar/radar-centric - so that either can maintain safe operation if the other degrades. This design choice directly addresses the validation challenge: by keeping the subsystems decoupled, each can be tested and certified on its own terms.

Beyond passenger vehicles, Mobileye has extended its perception and mapping stack into humanoid robotics and physical AI, applying the same sensor-fusion and real-time inference techniques to non-automotive embodied systems. The engineering culture emphasizes safety as a property that must be designed in from the outset, not verified after the fact, with individual contributors given latitude to own subsystem decisions across computer vision, machine learning, mapping, and embedded hardware domains.