Industrial Physical AI
for Smart Factory
Unified Intelligence for Smart Factories
AI Inspection Software
withAI™ is a state-of-the-art AI inspection platform designed to seamlessly collect
and integrate inspection data, analytics, and equipment information.
→ It enables effortless upgrades from existing PC-based environments,
facilitating rapid deployment on a proven and validated system architecture.
Core Values
Real-Time Defect Detection Hardware Edge Device
A powerful AI-driven inspection system delivered through a One-Device, All-in-One architecture.
Supports GPU acceleration, D2D processing, and OCR capabilities?enabling high-speed, real-time inspection directly at the edge.
Application Areas
-
Medical & Pharmaceutical
- Ultrasound image analysis and diagnosis
- Cell inspection (classification of Live, Dead, and Debris)
- Drug type and grade classification
- Foreign material and print inspection in syringes
- Classification of inspection targets by type
- Detection of defects in ampoules (cracks, scratches, color, shape)
-
Material & Components
- Metal surface defect and scratch inspection
- Cross-section inspection of wires and cable color verification
- Capsule content inspection (bubbles, foreign particles)
- Wafer defect inspection (scratches, contamination)
- Soldering inspection (insufficient, excess, small, and soft soldering)
- Rubber ring inspection (foreign materials, deformation, edge defects)
-
Display & Secondary battery
- Foreign material inspection inside panels
- Edge crack and chipping inspection
- Lead burr, open, and solder defect inspection
- Surface defect inspection (spots, lines, dents, inclusions)
- Scratch detection on mobile devices, camera lenses, and fingerprint sensors, including external crack inspection
-
Logistics & Distribution
- Sorting moving boxes
- Sorting and alignment of screws for product assembly
- Automated classification of agricultural products
- Inspection of ordered items, shipping, and return comparison
- OCR reading for packaging and containers
-
Foods & Packaging
- Inspection of cracks, scratches, and foreign materials in cosmetic containers
- Expiration date printing inspection inside packaging
- Inspection of printing and damage on cans, bottles, and packaged goods
- Detection of contamination and foreign substances in food
-
Robotics Autonomous Driving
- AMR (Autonomous Mobile Robots)
- Localization and simultaneous map generation (SLAM)
- Path planning, obstacle avoidance, and collision prevention through object recognition
- Prediction of drivable areas using segmentation
- Multi-robot control and obstacle avoidance based on reinforcement learning, anomaly detection
- Real-time collaborative operation recognition using camera sensing and DDS integration
withAI 2.0 VLM goes beyond object detection-based AI by analyzing the relationships and movement of operators, vehicles, equipment, and obstacles, delivering industrial Vision Intelligence that enables comprehensive situational awareness in manufacturing environments.
-
Decision-Making Based on Scene Understanding in Industrial Environments
- Object-Level Understanding : Identifies and classifies on-site entities?such as people, vehicles, equipment, pallets, loads, and obstacles?into meaningful semantic units
- Context & Relationship Analysis : Interprets interactions between objects (e.g., Is a person within a robot’s movement path?, Is a vehicle approaching?)
- State & Behavior Estimation : Analyzes behavioral patterns such as stop, movement, approach, departure, and intrusion into hazardous zones
- Risk Level Classification : Categorizes situations into normal, caution, or risk based on factors such as unsafe distances, collision likelihood, and worker proximity
-
Connecting “Perception and Action” through 3D + VLM + Physical AI Integration
- 3D-Based Spatial Awareness : Accurately captures physical space information such as distance, direction, height, and collision margins
- VLM-Based Semantic Decision-Making : Enables context-aware decisions (e.g., human approach, vehicle path intrusion, obstacle type and risk level)
- Physical AI for Execution Logic : Enhances on-site control policies including collision avoidance, speed regulation, rerouting, and emergency stop tuning
- Self-Supervised Calibration : Automatically compensates for variations caused by sensor, camera, or equipment changes, reducing maintenance overhead
-
Real-Time Edge Deployment (Immediate On-Site Operation)
- Real-Time Processing : Enables instant decision-making on-site without network latency
- Stable On-Site Operation : Ensures production line stability by minimizing dependence on servers and cloud systems
- Easy Monitoring & Configuration : Supports USB/LAN/Tablet-based configuration changes with immediate result verification
- Data Integration & Transmission : Transmits factory status data to Agency and Operational Twin systems in structured formats
- Scalable Operations : Deploys consistent policies and models across multiple edge devices, ensuring standardized criteria across lines and processes
Decision-Making Based on Scene Understanding in Industrial Environments
- Object-Level Understanding : Identifies and classifies on-site entities?such as people, vehicles, equipment, pallets, loads, and obstacles?into meaningful semantic units
- Context & Relationship Analysis : Interprets interactions between objects (e.g., Is a person within a robot’s movement path?, Is a vehicle approaching?)
- State & Behavior Estimation : Analyzes behavioral patterns such as stop, movement, approach, departure, and intrusion into hazardous zones
- Risk Level Classification : Categorizes situations into normal, caution, or risk based on factors such as unsafe distances, collision likelihood, and worker proximity
Connecting “Perception and Action” through 3D + VLM + Physical AI Integration
- 3D-Based Spatial Awareness : Accurately captures physical space information such as distance, direction, height, and collision margins
- VLM-Based Semantic Decision-Making : Enables context-aware decisions (e.g., human approach, vehicle path intrusion, obstacle type and risk level)
- Physical AI for Execution Logic : Enhances on-site control policies including collision avoidance, speed regulation, rerouting, and emergency stop tuning
- Self-Supervised Calibration : Automatically compensates for variations caused by sensor, camera, or equipment changes, reducing maintenance overhead
Real-Time Edge Deployment (Immediate On-Site Operation)
- Real-Time Processing : Enables instant decision-making on-site without network latency
- Stable On-Site Operation : Ensures production line stability by minimizing dependence on servers and cloud systems
- Easy Monitoring & Configuration : Supports USB/LAN/Tablet-based configuration changes with immediate result verification
- Data Integration & Transmission : Transmits factory status data to Agency and Operational Twin systems in structured formats
- Scalable Operations : Deploys consistent policies and models across multiple edge devices, ensuring standardized criteria across lines and processes
Specification
| 구분 | 권장 사양 | ||
|---|---|---|---|
| Train PC | CPU | Intel i7 이상 | |
| RAM | 32GB | ||
| CUDA Compute Capability | 3.5 | ||
| GPU | GeForce RTX 4090 (24GB 이상) | ||
| OS | Windows 10 x64, Windows 11 x64 | ||
| Inference PC | CPU | Intel i7 이상 | |
| RAM | 16GB 이상 | 32GB 이상 | |
| GPU | Geforce RTX 4070 (12GB 이상) | Geforce RTX 4080 (16GB 이상) | |
| OS | Windows 10 x64, Windows 11 x64 | ||
| 개발 환경 | Visual Studio 2017 이상 | ||
| 구분 | Target Image | 추론 시간 (RTX 4080) | 최적화 추론 시간 (RTX 4080) | |
|---|---|---|---|---|
| Detection | Lite | 2048 x 1536 | 79ms | 10ms |
| LiteX | 90ms | 11ms | ||
| Pro | 150ms | 28ms | ||
| Segmentation | Lite | 2048 x 1536 | 98ms | 13ms |
| LiteX | 105ms | 15ms | ||
| Pro | 165ms | 33ms | ||
| Classification | Lite | 200 x 200 | 6.3ms | 1.8ms |
| LiteX | 7ms | 1.9ms | ||
| Pro | 8.2ms | 2.1ms | ||
-
Inference Time Segmentation (Pro) 38.02ms Segmentation (Lite) 37.10ms Semantic Segmentation 36.30ms Detection (Pro) 34.56ms Detection (Lite) 26.50ms Classification (Lite) 2.44ms Specification CPU 12-core Arm® Cortex®-A78AE v8.2 64-bit CPU 3MB L2 + 6MB L3 GPU 2048-core NVIDIA Ampere architecture GPU with 64 Tensor Cores, 275TOPs Memory 64GB 256-bit LPDDR5 204.8GB/s Storage 1 x M.2. key M 2280 for SSD / 64GB eMMC Networking 1 x GbE RJ-45 USB 4x USB 3.2 Gen 2 Type-A / 2x USB 3.2 Gen2 Type-C IO 40-pin header (UART, SPI, I2S, I2C, CAN, PWM, GPIO) Power 15W - 60W Dimension 110mm x 110mm x 71.65mm Camera 4x 2D cameras, 2x 3D cameras AI Model fully compatible with withAI 2.0 3D supports Physical AI models, semantic segmentation, point-cloud/3D processing, pre-processing. -
Inference Time Segmentation (Pro) 78.01ms Segmentation (Lite) 73.24ms Semantic Segmentation 57.71ms Detection (Pro) 69.27ms Detection (Lite) 54.15ms Classification (Lite) 2.99ms Specification CPU 6-core Arm® Cortex®-A78AE v8.2 64-bit CPU 1.5MB L2 + 4MB L3 GPU 1024-core NVIDIA Ampere architecture GPU with 32 Tensor Cores, 67TOPs Memory 8GB 128-bit LPDDR5 102.4GB/s Storage 1 x M.2. key M 2280 for SSD Networking 1 x GbE RJ-45 USB 4x USB 3.2 Gen 1 Type-A IO 40-pin header (UART, SPI, I2S, I2C, CAN, PWM, GPIO) Power 15W - 25W Dimension 151mm x 98.5mm x 73mm Camera 2x 2D cameras, 2x 3D cameras AI Model fully compatible with withAI 2.0 3D supports Physical AI models, semantic segmentation, point-cloud/3D processing, pre-processing. -
Inference Time Segmentation (Pro) 62.81ms Segmentation (Lite) 60.32ms Semantic Segmentation 49.12ms Detection (Pro) 58.46ms Detection (Lite) 46.87ms Classification (Lite) 2.85ms Specification CPU 8-core Arm® Cortex®-A78AE v8.2 64-bit CPU 2MB L2 + 4MB L3 GPU 1024-core NVIDIA Ampere architecture GPU with 32 Tensor Cores, 157TOPs Memory 16GB 128-bit LPDDR5 102.4GB/s Storage 1 x M.2. key M 2280 for SSD Networking 1 x GbE RJ-45 USB 4x USB 3.2 Gen 1 Type-A IO 40-pin header (UART, SPI, I2S, I2C, CAN, PWM, GPIO) Power 10W - 40W Dimension 151mm x 98.5mm x 73mm Camera 2x 2D cameras, 2x 3D cameras AI Model fully compatible with withAI 2.0 3D supports Physical AI models, semantic segmentation, point-cloud/3D processing, pre-processing.
Operational Twin
The Brain of the Factory
-
1. What Is the “Brain” We Build?
- Operational Twin is an intelligent operations platform that understands the real-time status of a factory and recommends the next best actions.
-
2. Technology Architecure
- Real-time Line Monitoring
- Event-based Setup Log Engine
- Guide LLM : Recommendation Engine
- Structured Situation Modeling
- Data-driven Continuous Learning
-
3. Why V-ONE Tech Brain?
- Integration of Inspection, Robotics, and Line Data
- Built on 20 Years of Accumulated Field Setup and Operational Data
- Continuous Data Accumulation through Setup Agency-Based Operations
- Improved Reusability and Performance across Similar Production Lines
- Guide LLM Architecture that Continuously Learns During Operation
