What is AirPath®
The Real-Time On-Device AI Platform for Robotic Intelligence
To unlock the next generation of robotics,
machines must process, reason, and respond in real-time — without relying on the cloud.
AirPath® is a high-performance on-device AI platform built to power humanoids, service robots, AI home agents, and intelligent mobility systems.
More than just model compression,
AirPath® delivers a complete integration of SoC architecture, OS, firmware, AI runtime, and middleware,
setting a new standard for embedded intelligence at the edge.
AirPath® enables real-time, scalable AI that brings the age of hybrid and hyperintelligent robotics into reality.
The Rise of On-Device & Hybrid AI ,
Why AirPath® Matters Now
As AI applications move closer to the edge, the demand for high-performance, real-time inference is rapidly transforming the robotics industry. Traditional cloud-dependent systems are no longer sufficient to support the intelligence and autonomy required in next-generation robots.
AirPath® is a purpose-built on-device AI platform that combines:
– High-speed tensor SoC and NPU integration
– ML workload acceleration through advanced AI programming
– A flexible and scalable software architecture
– Optimized firmware and embedded OS support
– Lightweight, high-performance AI models
These capabilities converge to provide a ready-to-deploy development environment for building intelligent robots, humanoids, smart mobility systems, LLM-powered home agents, and industrial AI machines.
More than just a development kit, AirPath® is an end-to-end AI solution stack — the brain for advanced robotics, enabling tight integration, edge-level inference, and hybrid intelligence across the AI spectrum.
On-device AI cannot be achieved through model optimization alone.
A high-performance on-device AI pipeline requires deep understanding of the SoC architecture and stack, combined with kernel and firmware-level optimization.
End-to-End Pipe Line for On-device AI
From Silicon to Intelligence — Building the Brain for AI Devices
At Integrit, we go beyond lightweight AI model compression.
We design the full on-device AI pipeline — from silicon to intelligence.
Our development process spans every critical layer of the embedded AI stack, integrating SoC architectures, firmware systems, runtime frameworks, and multimodal AI models into one unified ecosystem.
01
Next-Generation AI SoC Integration
Integrit’s BSP platform brings together Qualcomm Dragonwing™ and NVIDIA Orin™ SoCs, combining AP, NPU, and TPU resources into a high-efficiency edge AI environment. Our expertise in SoC software stacks, hardware resource control, and low-power design enables the creation of unified, high-performance BSPs optimized for AI inference at the edge.
02
OS & Firmware Development
We design embedded firmware and kernel-level integration for Linux, Ubuntu, Android, and ROS2 environments. This includes sensor driver synchronization, LiDAR and multi-camera control, and memory resource management — ensuring a stable, real-time AI execution environment across complex robotic and IoT systems
03
AI Runtime & Middleware Frameworks
Integrit optimizes the operation of global AI runtime libraries such as TensorRT, SNPE, ONNX, and TVM, building an AI middleware architecture tailored for each SoC and operating system. This ensures maximized performance, lower latency, and energy-efficient AI inference across devices running multimodal workloads.
04
Lightweight AI Model Integration
Finally, we integrate optimized and compressed AI models directly into the device runtime. Our on-device LLM, VLA, and multimodal AI architecture enables real-time tensor computation, adaptive reasoning, and multimodal perception — all executed locally, without reliance on the cloud. This allows multiple AI models to coexist and cooperate seamlessly within a single embedded platform.
An End-to-End AI Solution for High-Performance Robot Brains
Integrit builds the full brain.
From chip-level kernel drivers to multimodal runtime,
What is commonly called “on-device AI” today is often just cloud-trained models compressed through public quantization or lightweight conversion frameworks. Yet real on-device AI—running multiple models concurrently, under strict compute and thermal limits, on robots that require deterministic real-time inference—demands a fundamentally deeper engineering approach.
Integrit delivers this by architecting and optimizing the complete execution stack: tensor SoC core scheduling, memory and bandwidth orchestration, firmware-level tuning, and real-time control integration, creating a truly operational and intelligent embedded brain.
From AI model development to hardware execution, AirPath® provides a unified workflow—
an all-in-one development environment that enables model development, deployment, execution, integration, and scalable expansion on a single platform.
Echo System for On Device AI Develpoment
From Silicon to Intelligence — Building the Brain for AI Devices
AirPath® delivers a fully integrated, SoC-aware AI execution pipeline that spans the entire lifecycle of embedded AI deployment—covering model generation, quantization and lightweight optimization, hardware-level graph mapping, firmware driver binding, and deterministic real-time control. All processes are engineered to align with the underlying SoC architecture, accelerator topology, and hardware interface subsystems.
By tightly coupling model execution with low-level hardware operations, AirPath® provides synchronized control across heterogeneous compute units, sensor arrays, and multimedia pipelines. This includes coordinated driver-layer integration (I2C, SPI, GPIO, PCIe, V4L2, GStreamer), timing-accurate multi-sensor fusion, and runtime orchestration within the SoC’s AI, DSP, GPU, and ISP domains.
Through this unified SoC–firmware–AI stack, developers can rapidly prototype, optimize, and validate AI workloads and deploy them seamlessly to AirPath®-enabled devices—achieving predictable real-time performance and stable operation in resource-constrained edge and robotic environments.
01
Model Design & Training
Choose architectures such as YOLO, ResNet, etc. Train models in cloud or local environments Design models optimized for target device performance
02
Model Optimization & Quantization
8-bit / 16-bit quantization Structural pruning and speed optimization Prepare models for embedded execution
03
Framework Conversion & Packaging
Support for TensorFlow (.pb), PyTorch (.pt), ONNX (.onnx), Caffe2 Convert to Qualcomm SNPE (.dlc) ROS/ROS2 wrapping for robotic integration
04
AI Runtime & Neural Network Parsing
Support SNPE, PyTorch, TF-Lite, ONNX Runtime TVM-based model parsing Integrit's SynaAI model parser INN IR for SoC-specific graph transformation
05
Firmware & OS Integration
Linux/Ubuntu-based system environment Hardware accelerators optimized: OpenCL, NEON, HTP Core, AIC Compiler Real-time processing and multi-sensor synchronization
06
On-Device SoC Execution
Utilize heterogeneous compute: CPU, GPU, HTP, AIC, LPAI Execution scheduling and memory optimization Achieve both high performance and power efficiency
07
Robotics & System Integration
Fully integrated with ROS/ROS2 Real-time motor/sensor control and device synchronization Delivering deployable on-device AI for edge and robotic environments
AirPath® provides a fully integrated workflow that connects AI model development, optimization, hardware mapping, firmware integration, and on-device execution into a single unified pipeline.
AirPath® is engineered to meet the new era of robotics,
where low-latency, high-performance AI must run directly at the edge.
Built on Qualcomm’s Dragonwing AI SoC, AirPath® offers a deeply integrated on-device AI platform with tailored BSPs and seamless support for multi-OS environments including ROS2, Android, and Linux.
With full-stack acceleration for LLM-based real-time inference, multimodal VLA (Vision-Language-Action) models, and robot behavior generation, AirPath® empowers developers to build next-generation intelligent robots—faster, lighter, and more cost-effectively.
Whether for humanoids, mobile platforms, or industrial agents, AirPath® provides the foundation for scalable, production-ready AI robotics that operate independently of the cloud.
Real-Time, Secure, Autonomous Systems
Integrit’s on-device AI architecture minimizes network dependency, ensuring ultra-low latency, robust security, and full autonomy for intelligent robots and embedded systems.
Cost-Efficient Architecture for Sustainable ROI
With optimized battery life, reduced maintenance costs, and enhanced system reliability, AirPath enables practical automation and measurable ROI — supporting scalable and sustainable robotics deployment.
Modular & Scalable Architecture for OEMs
AirPath’s software-defined modular architecture integrates sensing, control, and AI functions — offering OEMs flexible configuration and seamless scalability across robotics and smart devices.
API-Centric Design for Seamless Integration
Designed for openness, AirPath offers standardized APIs that enable effortless integration with external AI models, business platforms, and third-party services.
Next-Gen Qualcomm Dragonwing™ AI SoC with Powerful Edge Compute
Powered by the 8th Gen Qualcomm® Dragonwing™ SoC with Hexagon™ Tensor Accelerator, AirPath delivers high-speed multimedia processing, machine learning acceleration, and vision computing with integrated ISP, EVA, and NPU cores.
Seamless 5G Connectivity for Cloud-Edge Integration
With built-in Bluetooth 5.1, Wi-Fi 6, and LTE/5G (including 3.5GHz–3.7GHz and 4.7GHz private 5G bands), AirPath enables wide-coverage, low-latency, always-on connectivity between robots and cloud intelligence platforms.
Remote Control, Data Streaming & Digital Twin Support
AirPath supports real-time sensing, motion data streaming, and edge-to-cloud synchronization—powering digital twins, remote monitoring, and supervisory control via standardized APIs and robot middleware.
Security, Provisioning, and Data Trust Framework for Robots
AirPath provides secure device registration, encryption, and trust management across the AI data lifecycle—ensuring safe, high-quality datasets and protected system operations in robotics environments.
AirPath® supports a wide range of global open AI middleware
integrating essential intelligence for advanced robotics and accelerating your development.
AirPath®의 AI Stack
Engineering AI & Enabling Edge
AirPath® delivers a fully integrated AI stack combining LLMs, vision AI, multimodal intelligence, RFM (Robot Foundation Model), and physical AI control engines — all on a single on-device platform.
From perception to reasoning and real-time response, the entire intelligence cycle is processed locally. Beyond open-source compatibility (ROS2, Android, ONNX, TensorRT), it supports enterprise-specific behavior generation and AI integration.
인티그리트는 엣지 기반의 로컬 AI 추론과 멀티모달 인식, RFM(로봇 파운데이션 모델), VLA(비전·언어·행동) 을 통합하는 온 디바이스 아키텍처로, 차세대 지능형 로봇과 휴머노이드, 스마트 홈, AI 어시스턴트 등 ‘차세대 로봇을 위한 브레인’, AI 제어와 확장을 위한 ‘Key Solution’을 제공합니다.
Integrit’s proprietary lightweight framework and middleware provide:
Pipeline for AI Model Development & Optimization
VLA-Based Framework (SynaAI VLA)
AI Agent Framework & Behavior Controller
VLA-Based Autonomous Navigation Engine
Matter-Compatible Smart Home Connectivity Manager
ROS / ROS2 Wrapping & Porting Environment
Firmware Development
& Optimization for On-Device AI
Engineering AI & Enabling Edge
On-device AI and autonomous robotics require a fully integrated system stack. True on-device AI is not achieved by simply compressing a model and deploying it to a device; it demands the seamless integration of the AI SoC, firmware, operating system, sensor pipelines, multimedia subsystems, networking, and real-time control.
Integrit possesses deep-tech capabilities to build software-defined hardware, developing and integrating the most fundamental and complex layers—including the kernel, system components, firmware, and hardware interfaces—to deliver a fully optimized on-device AI platform.
Building a truly optimized on-device AI system requires an integrated and finely tuned stack that spans the AI SoC–based hardware, firmware, operating system, and robotics layers. On-device AI is not achieved through high-level models or applications alone; it depends on a deeply engineered foundation—kernel optimizations, device drivers, sensor control, multimedia pipelines, connectivity frameworks, security, and the full boot chain—all tailored to the underlying hardware.
Integrit has established the full-stack capabilities required for this foundation, including:
Linux/Ubuntu-based Rootfs development
Kernel and device-driver optimization (I2C, SPI, USB, PCIe, UFS, Ethernet, HDMI, and more)
Multimedia and compute pipelines for multi-sensor and camera processing (V4L2, OpenCL, OpenGLES, GStreamer, etc.)
Connectivity stacks for WLAN, Bluetooth, and 5G
AI/CV acceleration using Qualcomm® Hexagon DSP, Neural Processing SDK, DirectNN, and Computer Vision SDK
Robotics frameworks including MoveIt, SLAM, and ROS/ROS2
System integrity and update mechanisms such as bootloaders, Fastboot, A/B OTA, and DM-verity
These layers form the essential infrastructure enabling real, deterministic on-device AI performance.
Integrit provides firmware development and optimization across all foundational layers—OS, kernel, hardware interfaces, AI accelerators, and robotics modules—delivering the core system technologies required to power advanced AI systems.
From industrial sites and mobility systems to home devices and humanoid robots,
the AirPath® platform delivers tailored core intelligence optimized for advanced robotics
through seamless integration of high-performance AI SoCs and robust AI frameworks.
Edge AI Brain, AirPath® Platform Line-up
Engineering AI & Enabling Edge
From industrial sites and mobility systems to home devices and humanoid robotics, AirPath® delivers custom-tailored core intelligence for next-generation robotics through the effective integration of high-performance AI SoCs and on-device AI frameworks.
With the latest Qualcomm® Dragonwing APs and powerful NPUs optimized for low-power environments, the AirPath® platform offers a scalable and hybrid-ready alternative for developing advanced robotic systems requiring vision AI, multi-LLM, and multimodal intelligence.
Low-power AI Robotics Dev Kit with Qualcomm SoC
Based on Qualcomm® QCS5165N, this robotics development kit supports fusion of camera and LiDAR (2D/3D) data and accommodates MIPI, FHD USB 3D, and up to 9 camera modules.
First-in-industry 5G Private Network AI Robotics Kit
Also built on Qualcomm® QCS5165N, this model is the first AI robotics development kit designed to support private 5G networks, enabling ultra-reliable, low-latency AI robotics in industrial and enterprise environments.
On-Device AI Kit with High-Performance NPU (FuriosaAI Companion)
Integrates a companion NPU chip from FuriosaAI with Qualcomm SoC to provide 48 TOPS computing, ideal for high-throughput AI workloads in edge robotics.
On-Device AI Dev Kit with Hailo NPU Integration
Combines Qualcomm SoC and Hailo NPU in a compact platform optimized for on-device multimodal AI, supporting low-power real-time inference and edge deployment.
On-Device AI Dev Kit with DeepX NPU Integration
Embedded with DeepX NPU, providing scalable AI compute for compact robotics. Optimized for ultra-low-latency edge deployment in industrial or mobile applications.
AI brain for robotics & mobility ,Edge platform based on QCS8550
Designed for real-time autonomous navigation using vision-based AI inference, based on Qualcomm® 8 Gen 2 SoC (QCS8550).
Flagship on-device AI platform for multimodal robotics systems
Based on Qualcomm® Dragonwing QCS8550. Supports integrated camera & sensor control, LLM+VLM+STT inference, and ROS2 middleware, enabling seamless behavior generation from vision inputs.
Vision kit based on V3 Chrono with camera Module
Vision kit based on V3 Chrono with GMSL camera integration Provides out-of-the-box camera vision support for SLAM, perception, and navigation.
Ultra-compact AI Edge for Humanoid & Quadruped Robots
Tailored for lightweight robotics, McQueen integrates VLA models, Korean-optimized LLMs, and robotic foundation models (RFM) into a single on-device stack. It enables real-time reasoning and execution for humanoids and agile 4-legged platforms.
Industrial Edge AI Platform Pushing Robotics Boundaries
ased on Qualcomm Dragonwing IQ9, Stratos brings extreme compute capability with enhanced GPU/NPU performance, supporting on-device LLM + VLA + RFM + low-latency control. Ideal for social humanoids and advanced RFM applications.
High-Compute Hybrid AI Brain with Dual-Chip Architecture
An industry-first heterogeneous dual-SoC platform, combining Qualcomm QCS9075 + NVIDIA Jetson Orin into a single board. Hyperion performs on-device inference + control + learning simultaneously, enabling next-gen AI cognition and autonomy.
5G External Antenna Module
High-gain antenna for N79 band and public/private 5G Designed for service robots, mobility, and humanoids requiring reliable long-range connectivity.
AirPath® empowers you to bring intelligent robotics to life
faster, smarter, and ready for the real world. Discover your next edge
Integrit designs the future of robotics and physical AI through decades of global collaboration and deep expertise in telecommunications and embedded systems.
AirPath® is more than a hardware platform — it is a scalable intelligence core that accelerates the next generation of autonomous machines.
From compact edge modules to GPU-SoC hybrid systems, AirPath® offers a full range of AI computing tailored for diverse robotics use cases.
Whether it’s STT, LLMs, or VLA models, every AirPath® unit — from the V2 Kit to the V9 Hyperion — supports real-time multimodal inference and physical AI execution entirely on-device.
By integrating ultra-efficient Qualcomm® Dragonwing SoCs
with high-performance NVIDIA® Orin platforms,
AirPath® enables reliable and adaptive AI deployment — from R&D to mission-critical industrial robotics.
| Category | V3 Chrono | V4 McQueen | V5 Stratos | V9 Hyperion |
|---|---|---|---|---|
| Core Platform | Qualcomm QCS8550 | Qualcomm QCS8550 + External AI Module | Qualcomm QCS9075 + Cloud AI100 Ultra | Qualcomm QCS9075 + NVIDIA Jetson Orin NX |
| Chip Architecture | Kryo Gen6 + Adreno 740 + Hexagon DSP | Kryo Gen6 + Hexagon DSP + AI Hub Expansion | Kryo Gen6 + Hexagon Tensor + Cloud AI100 (Up to 400 TOPS) | Kryo Gen6 + Hexagon Tensor + NVIDIA Ampere GPU (Jetson Orin NX) |
| AI Performance | Approx. 40 TOPS (INT8) | Approx. 40 TOPS + External Module | 100 TOPS + 400 TOPS (Cloud AI) | 100 TOPS + 100–275 TOPS (Orin NX) |
| Memory / Storage | 16GB LPDDR5 / 128GB UFS | 16GB LPDDR5 / 128GB UFS | 36GB LPDDR5 / Dual 128GB UFS | 64GB LPDDR5 + NVMe / eMMC + SSD Expansion |
| Graphics | Adreno 740 GPU | Adreno 740 GPU | Adreno 663 GPU | NVIDIA Ampere GPU (1024 CUDA Cores) |
| Operating System | Ubuntu 22.04 / ROS2 Humble | Ubuntu 22.04 / ROS2 / Docker | Ubuntu 22.04 / ROS2 / Docker | Ubuntu 22.04 / ROS2 / CUDA / JetPack / Isaac Sim |
| Middleware Compatibility | ROS2, Docker | ROS2, Docker, AI Hub | ROS2, ONNX, Edge Deployment Suite | ROS2, Isaac GROOT, ONNX, NVIDIA NGC |
| On-device AI Capabilities | STT, LLM (3B), Vision AI | Multimodal STT, LLM, Action Control | Vision + Language + Action (VLA), RFM | Dual-AI: VLA, VLM, Simulation AI |
| External Sensor I/F | USB3 / CSI / Ethernet | USB3 / PCIe / Ethernet / MIPI | PCIe Gen4 x2 / x4, 2.5G LAN ×2, GPS | PCIe, USB4, GMSL, 10GbE, LiDAR, IMU, Multi-Cam |
| 5G / Connectivity | Built-in 5G (Sub-6) | Built-in 5G + Wi-Fi | Built-in 5G + GPS, Wi-Fi 6E | Qualcomm 5G + Multi-Gigabit Ethernet |
| Main Applications | AI Service Robot, Infrastructure Bot, Home Agent | Smart Service Robot, Retail / Concierge | Industrial / Defense Robot, Drone, Mobile Platform | High-end Humanoid, Human-like Robot, Simulation AI |
| Key Features | Compact + On-device LLM | Optimized for Vision AI with High-Speed Inference | Cloud AI100 Max 500 TOPS supported | Qualcomm + NVIDIA Hybrid Integration |
| Operating Temp / Durability | -20°C ~ 85°C / Industrial Grade | -20°C ~ 85°C / Industrial Grade | -40°C ~ 115°C / Extreme Industrial | -40°C ~ 115°C / MIL Grade Ready |
| Form Factor | 80 × 60 mm board | 90 × 70 mm board + Module | 65 × 65 mm board (IQ9075M) | NVIDIA Carrier + Qualcomm Board |
Accelerate your robotics journey
with powerful AI cores, modular flexibility,
and real-time autonomy. Build with AirPath®
AirPath® Reference & Customers
Integrit collaborates with global top-tier companies—organizations that define industry standards, shape technology roadmaps, and lead innovation worldwide.
By delivering flagship on-device AI and robotics projects with these industry leaders, Integrit helps open the next generation of on-device AI markets and sets new benchmarks for real-world deployment.
Experience AirPath® V4 McQueen, an on-device RFM platform that empowers the next generation of quadruped and humanoid robots — enabling real-time intelligence and breakthrough performance in motion, control, and autonomy.
Get into More