Omniseer

Omniseer is an edge-to-cloud embodied AI project centered on open-vocabulary perception on a ROCK 5B+ mobile robot.

The implemented robot-side path captures camera frames, preprocesses them with RGA, runs YOLO-World on the Rockchip NPU, and publishes typed detections and performance telemetry through ROS 2. Simulation, firmware, real-hardware bringup, a gRPC gateway, and on-demand SRT preview provide the supporting platform.

The active portfolio deliverable is narrower than autonomous object search:

• select semantic target classes
• run and measure onboard inference
• record detections, telemetry, and useful evidence
• review an experiment on a laptop and, later, through a cloud-hosted workflow

Native runtime class updates, structured run bundles, cloud synchronization, and a hosted review dashboard are planned. They are not described as implemented. Autonomous seek and capture are explicitly deferred.

Start with:

• Scripts Front Door for the supported local command surface
• Edge-to-Cloud Perception for the active deliverable
• System Architecture for runtime boundaries and implementation status
• Vision Pipeline for the native hot path
• CI/CD Overview for automated verification and its limits
• Robot Gateway and Preview Streaming for operator diagnostics