MiniEye provides a modular perception SDK that fuses camera, lidar, and radar data to deliver real‑time 3D object detection, tracking, and semantic segmentation with sub‑0.1 m accuracy and <30 ms latency on automotive‑grade hardware. The platform includes weather‑adaptive models, ISO 26262 ASIL‑D safety validation tools, and OTA update capabilities, enabling OEMs and Tier‑1 suppliers to integrate robust autonomous‑driving perception without extensive R&D.
Funding
Funding not disclosed
Founders
Product
Problem
Current autonomous driving stacks struggle with reliable 3D perception in diverse weather and lighting conditions, leading to gaps in object detection, lane understanding, and sensor redundancy. Integrating data from multiple sensors (camera, lidar, radar) often requires custom fusion pipelines that are costly to develop and maintain. These limitations impede the deployment of Level 3+ autonomous functions at scale.
Solution
MiniEye delivers a modular perception platform that combines high‑resolution 3D object detection, semantic segmentation, and sensor‑fusion algorithms optimized for automotive-grade hardware. The software stack ingests raw data from cameras, lidar and radar, applies calibrated‑free alignment, and produces a unified environmental model in real time (< 30 ms latency). Built‑in weather‑robustness modules maintain detection accuracy under rain, fog, and low‑light scenarios. The platform is delivered as an SDK with pre‑validated safety cases, enabling OEMs and Tier‑1 suppliers to integrate advanced perception without extensive R&D. Continuous over‑the‑air updates ensure algorithmic improvements and compliance with evolving safety standards such as ISO 26262.
Target Audience
Primary customers are automotive OEMs and Tier‑1 suppliers developing Level 3 and higher autonomous driving systems, as well as robotics firms requiring robust multi‑sensor perception for navigation.
Features
- Multi‑sensor fusion engine that merges camera, lidar, and radar streams using deep‑learning‑based cross‑modal registration
- Real‑time 3D object detection and tracking with < 0.1 m positional error at 100 m range
- Weather‑adaptive perception models trained on synthetic and real‑world datasets for rain, fog, and night operation
- Semantic segmentation of drivable space and lane markings with pixel‑level accuracy > 95 %
- Automotive‑grade inference pipeline optimized for NVIDIA DRIVE, Qualcomm Snapdragon and custom SoCs, delivering < 30 ms end‑to‑end latency
- Built‑in safety validation toolkit with ISO 26262 ASIL‑D compliance documentation and fault‑injection testing
- OTA update framework for seamless model upgrades and calibration‑free sensor onboarding
- API‑first design with ROS2 and AUTOSAR interfaces for rapid integration into existing vehicle architectures