Bee Edge AI
Edge Platform for Custom Workloads
Last updated
Edge Platform for Custom Workloads
Last updated
At its core, the Bee was designed with a modular, plug-in architecture, making it fast and easy to deploy workloads. Starting today, we’re unlocking three powerful capabilities for developers.
You can now deploy your own computer vision models directly on the Bee, running alongside the native Hivemapper Map AI stack. This allows you to tailor object detection to your specific use case—whether it’s identifying utility infrastructure like transformers and poles, or industry-specific features like inspection panels, parking meters, or construction signage.
Models are deployed via OTA (over-the-air) updates using Bee Maps’ geo-targeted infrastructure. You can push models to specific devices by region—at the country, state, or metro level—optimizing inference and data coverage based on your operational needs.
The Bee natively uses a single, edge-optimized YOLO model that performs real-time detection and classification on all frames captured by its 12.3MP camera. Running on a 5.1 TOPS-capable System-on-Module (SoM), the Bee is optimized for low-latency inference, even in offline or bandwidth-constrained environments.
How to Deploy a Custom Model
• We collaborate with partners to train objects using their data.
• Your model is deployed to a small subset of Bee devices to validate accuracy and recall
• Once approved, it’s rolled out to your designated geographies via an OTA update to Bee devices.
Bee - Native Mapping Detections (eg speed signs, traffic lights, etc.)
Your own proprietary dataset for these objects
Your Custom Object Detections
Your own proprietary dataset for these objects
You can track changes to specific infrastructure objects—like speed limit signs—by uploading a reference dataset in GeoJSON format. Each object should conform to our feature spec, for example:
Bee Maps automatically partitions and distributes these objects to devices most likely to encounter them. As Bees detect and upload fresh observations, your GeoJSON file is updated. You can subscribe to real-time changes via API to track updates as they occur.
All data generated from your custom edge detections and change detection workflows—along with optional supporting imagery—is automatically streamed from the Bee to your Bee Maps developer account. This uses the device’s built-in LTE and WiFi offload channels, with no additional integration work required.
Simply configure your endpoint, and the Bee will begin transmitting structured visual data in real time, ready for ingestion into your systems or workflows.
Contact us (hi@hivemapper.com) to become a Bee developer.
