Loading content…
QBS Co.
Engineering proof-of-concept integrating real-time computer vision object detection with unmanned surface vessels, 4-camera panoramic perception, 360° LiDAR depth fusion for range estimation, drone visual integration, and low-latency Flutter telemetry over MQTT.
PythonNVIDIA JetsonTensorRT360 LiDARSensor FusionMQTTFlutterComputer VisionObject DetectionDrone IntegrationMission PlanningEdge AICUDAPyTorchAutonomous Vessels
NAVIROX was a high-stakes engineering proof-of-concept developed at QBS Co. from February to July 2026, built to answer one question: can you give an unmanned surface vessel the ability to see, measure, and think in real time?
What It Was A multimodal perception system for autonomous unmanned vessels — combining cameras, LiDAR, and drone feeds into a single edge-compute intelligence layer capable of detecting obstacles, estimating distances, and feeding spatial data into mission planning engines.
4-Camera Array + 360° LiDAR Fusion Four surround cameras gave the vessel panoramic visual coverage. A synchronized 360° LiDAR sensor added real depth — computing precise distance, range, and bearing to maritime obstacles, other surface vessels, and navigational markers. The two modalities were fused so every visual detection came with an accurate spatial measurement, not just a bounding box.
Edge AI on NVIDIA Jetson All inference ran locally on embedded NVIDIA Jetson hardware — no cloud, no latency. Object detection models were optimized with TensorRT for FP16 acceleration, sustaining 30+ FPS across all cameras under strict power and thermal constraints. Edge-first by design.
Drone Integration → Mission Planning Aerial drone feeds were pulled into the same perception pipeline. Detections from the drone — objects, vessels, coastlines — were translated into spatial parameters and injected directly into the vessel's mission planning and navigation trajectory engine. The drone became a scout; the vessel responded autonomously.
MQTT Telemetry + Flutter Cockpit All live data — vessel velocity, LiDAR point clouds, detection alerts, and spatial coordinates — was streamed over lightweight MQTT broker channels to a custom Flutter tablet and mobile application. Field operators could monitor the vessel's full situational awareness in real time from anywhere.
The POC was field-validated with working multimodal perception, accurate 50-metre obstacle distance estimation, drone-to-vessel mission coordination, and sub-second telemetry responsiveness.
Unmanned surface vessels operating in maritime environments face severe perception challenges: single-camera setups lack depth accuracy in open water, while standalone LiDAR struggles to semantically classify obstacles. Constrained edge compute on embedded hardware requires sub-30ms perception latency for safe navigation.
Engineered an edge multimodal perception pipeline on NVIDIA Jetson using TensorRT-accelerated deep learning detectors in Python. Fused 2D detections from 4 surround cameras with spatial depth from 360° LiDAR returns, integrated drone aerial detections into mission trajectory planning, and streamed telemetry over MQTT to a Flutter operator cockpit.
Built the core perception daemon in Python with TensorRT and CUDA optimizations on NVIDIA Jetson. Synchronized timestamped frames across 4 USB/MIPI cameras and mapped 360° LiDAR range bins to camera bounding frustums for distance estimation. Connected aerial drone detection streams to the vessel's mission planner, and published JSON telemetry packets over MQTT topics (QoS 1) consumed by a custom Flutter BLoC telemetry app.
Successfully demonstrated working multimodal perception in field trials: real-time 4-camera surround detection at 30+ FPS on Jetson, accurate obstacle distance estimation up to 50 meters via 360° LiDAR fusion, coordinated drone-to-vessel mission parameterization, and responsive sub-second telemetry updates in the Flutter app.
Edge multimodal perception and telemetry architecture on NVIDIA Jetson fusing 4 surround cameras, 360° LiDAR, and drone feeds with an MQTT broker and Flutter telemetry cockpit.
Surround Vision & Aerial Ingestion
Captures and synchronizes 4 onboard surround cameras plus wireless video telemetry from reconnaissance drones.
PythonOpenCVRTSP4-Camera Array
Edge Deep Learning & TensorRT Engine
Executes accelerated object detection and semantic classification directly on NVIDIA Jetson hardware.
NVIDIA JetsonTensorRTCUDAPyTorchPython
360° LiDAR Spatial Fusion
Projects 360° LiDAR point-cloud range returns onto 2D camera bounding boxes to calculate exact obstacle distances and bearings.
360 LiDARSensor FusionPoint CloudNumPy
Mission Planning & Drone Integration
Translates visual and LiDAR detections from vessel and drone into spatial avoidance parameters for autonomous mission trajectory planning.
Mission PlanningDrone IntegrationAutonomous Navigation
Telemetry & Operator Cockpit
Dispatches high-frequency vessel telemetry and obstacle alerts over MQTT to a cross-platform Flutter tablet application.
MQTTFlutterBLoCDartCross-Platform
