Powering Next-Generation AI-Powered Drones
Customer
Gremsy (in collaboration with Lantronix and Teledyne FLIR OEM)
Application
Drone / UAV Edge AI Compute
Industry
Unmanned Aerial Vehicles (UAV)
Qualcomm Technology
Dragonwing QRB5165
Qualcomm Partner Network Offering
Open-Q 5165RB
Overview
Gremsy develops advanced gimbal and camera systems for drones used in public safety, defense, inspections, and surveillance. For its next-generation drone platform and Lynx ISR (Intelligence, Surveillance, and Reconnaissance) camera payloads, Gremsy needed onboard compute that could support real-time AI, thermal-visible sensor fusion, and mission-ready payload integration.
Gremsy worked with Lantronix to use the Open-Q 5165RB System on Module, based on the Dragonwing QRB5165. The platform is structured to give Gremsy a compact, reusable compute foundation for AI-enabled drone payloads without depending on cloud processing.
Challenge: Enabling real-time, onboard AI processing for UAV payloads within strict SWaP and defense-compliance constraints
Modern UAV payloads for defense and public safety need to process data in flight, not after landing. The compute platform must run AI inference, synchronize multiple sensors, and support reliable field operation while staying within strict size, weight, and power (SWaP) limits. General-purpose CPUs and cloud AI are generally not designed to meet these demands; the compute has to be purpose-built and onboard.
Gremsy needed a compute platform that could deliver:
- Real-time onboard AI: Detect and track objects in flight, without relying on the cloud.
- Multi-sensor fusion: Combine thermal and visible camera feeds into one synchronized view.
- Field-ready reliability: Perform consistently in harsh conditions over long deployments.
- Low SWaP: High performance in a small, lightweight, power-efficient package.
- Defense compliance: Meet NDAA and TAA requirements with long-term component availability.
- Fast time to market: A production-ready platform that scales across multiple drone models.
The Solution: Lantronix Open-Q SOMs built on Qualcomm Dragonwing QRB5165 SoCs delivered NDAA/TAA-compliant, low-SWaP AI compute for real-time UAV sensor fusion
Gremsy deployed the Dragonwing QRB5165, integrated into Lantronix Open-Q™ SOMs, as the compute foundation for its drone platform and Lynx ISR payload. Both SoCs share a 7 nm heterogeneous architecture that are engineered to distribute AI, vision, video, and connectivity workloads across dedicated engines simultaneously. Lantronix is constructed to package each SoC into a production-ready 50 mm x 29 mm module with 8 GB LPDDR5, 128 GB UFS, Wi-Fi 6, Bluetooth 5.1, and full I/O, helping give Gremsy a fast path to deployment without custom silicon design.
The solution includes:
- 15 TOPS on-device AI inference: The 5th-gen Qualcomm® AI Engine (Qualcomm® Hexagon™ Tensor Accelerator, NPU 230, Qualcomm® Hexagon™ 698 DSP) helps enable real-time detection, tracking, and classification in flight, with a dedicated Engine for Video Analytics (EVA) handling computer vision at reduced latency.
- Multi-camera ISR sensor fusion: The Qualcomm Spectra™ 480 dual 14-bit ISP is built to support 7 concurrent cameras on Dragonwing QRB5165 or 24 video streaming cameras on Dragonwing QCS8250, with hardware-synchronized thermal and visible streams, 8K video encode/decode, and 64 MP capture.
- Heterogeneous compute for concurrent workloads: Qualcomm® Kryo™ 585 CPU, Qualcomm® Adreno™ 650 GPU, Hexagon DSP, EVA, and NPU are constructed to run AI inference, video processing, gimbal control, and communications simultaneously without bottlenecks.
- Military-grade reliability in a compact module: The 7 nm SoC operates at -25°C to +85°C with advanced performance-per-watt. Wi-Fi 6, Bluetooth 5.1, PCIe Gen3, dual USB 3.1, and MIPI CSI/DSI are all on the 50 mm x 29 mm module, aiming to minimize payload footprint and maximize mission endurance.
- NDAA/TAA-compliant silicon with decade-long lifecycle: Dragonwing QRB5165 meets NDAA Section 889 and TAA requirements. Dragonwing 5165RB is supported through February 2036 under Qualcomm's IoT Product Longevity Program.
- Scalable, reusable platform architecture: A unified Qualcomm software stack (Android/Linux with ROS2) and consistent SoC architecture across both SOMs are designed to allow Gremsy to deploy the same compute platform across multiple drone models and payload configurations.
Outcome
- Real-time situational awareness at the edge: Up to 15 TOPS on-device AI is designed to deliver in-flight detection, tracking, and analytics without cloud dependency or latency.
- Advanced ISR imaging intelligence: Hardware-synchronized thermal and visible sensor fusion are built to support high-fidelity inspection, surveillance, and drone-as-first-responder missions.
- Accelerated drone platform development: Validated Lantronix SOM hardware and unified Qualcomm® software stack help reduce integration risk and shorten development cycles across UAV models.
- Full NDAA/TAA compliance: NDAA/TAA compliance support helps Gremsy address procurement requirements for defense and public safety UAV programs.
- Scalable across multiple UAV platforms: Consistent SoC architecture and software stack help enable reuse across drone models and payload configurations, aiming to reduce per-program development investment.
