Next-gen Touchscreens
With Edge AI

No, ARM64 apps cannot run directly on the ESP32-P4. The ESP32-P4 uses a RISC-V 32-bit architecture and runs FreeRTOS (a real-time operating system), not Linux. ARM64 binaries are incompatible — it's a completely different instruction set and operating environment.

However, the VLM Box runs on ARM64 Linux (i.MX8M Plus with Cortex-A53), so existing ARM64 Linux applications can potentially run there with minimal modifications. This architecture gives you the best of both worlds: a responsive RTOS-based touchscreen for HMI, and a Linux-based edge computer for complex processing tasks.

The ESP32-P4 can perform basic face detection using Espressif's ESP-WHO framework. However, for secure face recognition with liveness detection (anti-spoofing), the processing requirements exceed what the ESP32-P4 can handle in real-time.

For advanced biometric applications, we use a two-step approach: the ESP32-P4 handles initial face detection, then streams the camera frames along with Time-of-Flight (ToF) sensor data to the i.MX8M Plus edge computer. The i.MX8M Plus runs the liveness detection algorithm using ToF depth data to distinguish real faces from photos, videos, or masks — all processed locally without cloud dependency.

The ESP32-P4 is purpose-built for Human-Machine Interface (HMI) applications. Its RTOS foundation delivers consistent, predictable response times — critical for touch interfaces. It handles display rendering, touch input, camera preview, audio I/O, and network communication with sub-millisecond latency.

More powerful Linux-based processors introduce unpredictable latency, higher power consumption, longer boot times, and greater complexity. By separating the UI (ESP32-P4) from heavy compute (VLM Box), we optimize each component for its specific role.

No, all core functionality runs offline. The touchscreen and VLM Box communicate over your local network. AI processing, face recognition, voice commands, and VLM inference all happen on-premises — no cloud accounts, no subscriptions, no data leaving your network.

Internet connectivity is optional and only needed if you want features like remote access, OTA updates, or integration with external services.

The touchscreen (ESP32-P4) is the user interface — it handles display, touch, camera, microphone, speaker, and real-time interactions. It runs an RTOS for instant responsiveness and can perform basic on-device AI like face detection.

The VLM Box is an edge AI computer with a 40 TOPS NPU. It handles compute-intensive tasks: running 7B Vision Language Models, advanced face recognition, natural language understanding, and complex reasoning — all locally. Think of it as a local AI server that your touchscreen can query.

Yes. The VLM can be fine-tuned with your own RAG (Retrieval-Augmented Generation) data to provide domain-specific responses. You can customize the knowledge base, personality, and response style for your specific application — whether that's a smart home assistant, industrial support system, or customer service kiosk.

The touchscreen UI is also fully customizable with your branding, color schemes, and interface layouts.