VOOZH about

URL: https://deepwiki.com/AidenAI-IO/aiden-hardware-demo

⇱ AidenAI-IO/aiden-hardware-demo | DeepWiki


Loading...
Last indexed: 2 June 2026 (f15919)
Menu

Overview

Aiden Hardware Demo is a specialized automation and AI demonstration platform built for the Luckfox Pico Zero (RV1106). It integrates high-speed HDMI frame capture, full-duplex audio, and USB HID peripheral emulation to create an autonomous system capable of observing and interacting with external devices (such as smartphones or PCs) via a Large Language Model (LLM) agent.

The system bridges the gap between low-level hardware control and high-level AI reasoning by exposing hardware capabilities through a layered C++ service architecture and a Go-based agent runtime.

System Architecture

The project is organized into a hierarchical stack, ensuring that hardware resources are managed by dedicated owners while providing flexible access to high-level applications.

Overall System Diagram

The following diagram illustrates how the major software components interact and their corresponding locations in the codebase.

System Block Diagram


Sources: docs/02-architecture/overview.md5-34 src/aiden_sdk.cpp1-20 src/agent/pkg/runtime/runtime.go1-50

Major Subsystems

1. Hardware Platform

The project targets the Luckfox Pico Zero, utilizing its RV1106 SoC. Key hardware integrations include:

  • HDMI Capture: Handled via the TC358743XBG bridge connected to the CSI interface.
  • USB HID: Emulates a keyboard (/dev/hidg0) and a touchscreen/mouse (/dev/hidg1) using the Linux Gadget framework.
  • GPIO: Used for hardware wakeup triggers.

For details on wiring and device nodes, see Hardware Platform and Connections.

2. C++ Hardware Services

To prevent resource contention, dedicated C++ daemons manage the hardware:

For details on the IPC protocol and service implementations, see C++ Hardware Services.

3. Go Agent

The Aiden Go Agent is the "brain" of the system. It runs a ReAct-style loop to:

  • Observe: Take screenshots via frame_service.
  • Reason: Consult an LLM (OpenAI, OpenRouter, or Ollama).
  • Act: Execute HID commands (clicks, swipes, typing) or system commands.
  • Interact: Process voice via a pipeline involving VAD (Voice Activity Detection), STT (Speech-to-Text), and TTS (Text-to-Speech).

For details on the agent runtime and tools, see System Architecture and Data Flows.

Data Flow: Observation to Action

The following diagram maps the flow of a single "turn" from the agent perceiving the screen to performing a touch gesture.

Visual Control Data Flow


Sources: docs/02-architecture/overview.md55-73 src/agent/pkg/tools/tools_hid.go1-100 src/frame_ipc.cpp1-40

Child Pages

Sources: README.md1-70 docs/README.md1-69 docs/02-architecture/overview.md36-53