Physical AI in Action: BoothPilot Brings Real Time Edge AI Conversations to Automate 2026
At Automate 2026, NODKA is showcasing BoothPilot, a Physical AI demo built on the NODKA AP-6121 Embodied AI Robot Controller. The demo highlights how five local AI models can run together on one Intel Core Ultra Series 3 platform, using CPU, iGPU, and NPU acceleration to support real time perception, speech, language, OCR, and response generation.
BoothPilot is part of the NODKA and Intel Core Ultra Series 3 Embodied AI showcase. Powered by Panther Lake, Intel Robotics AI Suite, OpenVINO, and ROS 2 Jazzy, the demo shows how a single industrial edge AI controller can coordinate multiple AI workloads locally without cloud connectivity.
Five local AI models on one edge AI controller
The BoothPilot demo runs five AI models on the NODKA AP-6121 Embodied AI Robot Controller. Each model supports a different part of the Physical AI interaction loop, from detecting a visitor and processing speech to generating responses and reading badge text.
| Model | Function | Compute Resource |
|---|---|---|
| YOLOv8 | Visitor detection | Intel® Arc™ GPU |
| distil-whisper-large-v3 | Speech-to-text (ASR) | Intel® NPU |
| Qwen2.5 3B Instruct | Local LLM response generation | Intel® Arc™ GPU |
| SpeechT5 | Text-to-speech synthesis | CPU |
| PaddleOCR PP-OCRv4 | Conference badge OCR | CPU |
This model mix demonstrates how vision, ASR, LLM, TTS, and OCR workloads can be distributed across the Intel compute blocks inside one industrial PC. Instead of relying on a cloud server or external GPU system, the full loop runs locally on the AP-6121.
Hardware platform
| Component | Role |
|---|---|
| NODKA AP-6121 Embodied AI Robot Controller | Industrial embedded PC |
| Intel® Core™ Ultra X7 358H | Integrated CPU, GPU, and NPU on a single processor |
| 32 GB LPDDR5X | Unified memory shared across compute blocks |
| Intel® RealSense™ D435IF | Depth camera and badge image capture |
| Jabra PHS002W | Voice input and speech output device |
Software stack
| Software | Purpose |
|---|---|
| Ubuntu 24.04 LTS | Operating system |
| ROS 2 Jazzy Jalisco | Robot framework and orchestration |
| Intel® Robotics AI Suite | AI deployment framework |
| OpenVINO™ 2025.3 | Inference runtime across CPU, GPU, and NPU |
| PaddleOCR & GenAI Components | OCR, LLM, ASR, and TTS model support |
System architecture
Inside the NODKA AP-6121 Embodied AI Robot Controller, orchestration coordinates five AI models across the NPU, iGPU, and CPU during a live Physical AI interaction.
The architecture centers on orchestration and HMI gateway logic. ROS 2 Jazzy coordinates the interaction flow, Intel Robotics AI Suite supports edge AI deployment, and OpenVINO 2025.3 serves as the inference runtime across the available compute blocks.
The NPU runs speech recognition, the iGPU handles visitor detection and LLM response generation, and the CPU supports text to speech and badge OCR. The RealSense camera, Jabra speakerphone, and visitor screen complete the local interaction system.
Vision and Inside the demo flow
Each step in BoothPilot engages a specific input device, AI model, or compute block. The full loop runs on the same industrial PC, with no cloud round trip.
Why this matters for Physical AI
BoothPilot demonstrates what Physical AI can look like when multiple AI models run locally on a single industrial platform. Vision, speech recognition, language generation, speech synthesis, and OCR are coordinated together on one Intel Core Ultra Series 3 Panther Lake processor.
This matters because Physical AI systems need to process sensor input, reason locally, and respond quickly. BoothPilot shows this type of architecture in a compact edge AI controller, using CPU, iGPU, and NPU compute together instead of depending on cloud connectivity or a GPU server.
The same Panther Lake platform was introduced by NODKA and Intel as a Core Ultra Series 3 embodied AI robot controller architecture. BoothPilot provides a practical look at how this platform can support humanoid robots, collaborative robots, AMRs, industrial arms, and intelligent edge systems where responsive, private, on device AI is required.
Why this matters for Physical AI
BoothPilot demonstrates what Physical AI can look like when multiple AI models run locally on a single industrial platform. Vision, speech recognition, language generation, speech synthesis, and OCR are coordinated together on one Intel Core Ultra Series 3 Panther Lake processor.
This matters because Physical AI systems need to process sensor input, reason locally, and respond quickly. BoothPilot shows this type of architecture in a compact edge AI controller, using CPU, iGPU, and NPU compute together instead of depending on cloud connectivity or a GPU server.
The same Panther Lake platform was introduced by NODKA and Intel as a Core Ultra Series 3 embodied AI robot controller architecture. BoothPilot provides a practical look at how this platform can support humanoid robots, collaborative robots, AMRs, industrial arms, and intelligent edge systems where responsive, private, on device AI is required.
See how BoothPilot runs five local AI models, including large language models, speech recognition, text to speech, speaker recognition, and computer vision, on a single Intel® Core™ Ultra Series 3 edge AI platform, Download the BoothPilot Technical Overview.