Skip to main content

RDK S600 Application Cases Manual

This documentation collects typical application cases for the RDK S600 platform, organized by difficulty from basic peripheral integration to on-device AI inference, multimodal interaction, and embodied intelligence—helping you get started quickly and dive deeper step by step.

Documentation Structure​

1. Getting Started​

Introduces how to connect and use common peripherals, helping you set up the hardware environment and verify basic functionality.

  • USB Peripherals: Python examples for USB serial ports, USB cameras, and USB audio devices, including OpenCV image capture and sounddevice recording/playback.
  • UART: UART communication principles and pyserial usage, with an RS485 control demo using STS3215 bus servos.

2. Basic Cases​

Deploy entry-level AI models on the S600 device, covering vision, audio, and LLM scenarios with complete environment setup, case launch, and result demonstration workflows.

3. Intermediate Cases​

Combines multiple modality capabilities to build interaction experiences closer to real products.

  • Multimodal Interactive Assistant: Combines Whisper-medium and Qwen3VL-8B with a USB microphone and camera for "voice + vision" multimodal dialogue—for example, asking "What do you see?"

4. Advanced Cases​

For embodied intelligence scenarios, demonstrating PC simulation co-debugging and on-device deployment of the VLA (Vision-Language-Action) model, plus real-robot cases such as HoloBrain general object grasping and Pi0 & Pi05 towel folding.

  • Pi0 Hammer the Block (Simulation): Deploy the RoboTwin simulation environment on PC and the Pi0 model on S600 for inference, covering PC simulation deployment, S600 on-device deployment, and run results.
  • HoloBrain General Object Grasping (Real Robot): Configure the HoloBrain environment on RDK S600, deploy the model, and run a general object-grasping task, covering hardware setup, environment setup, and start/stop. See FAQ for common issues.
  • Pi0 & Pi05 Fold the Towel (Real Robot): Train Pi0 and Pi05 for the towel-folding task with LeRobot, quantize and compile with OELLM 2.0, and deploy and run on-device; the training and deployment flow is covered in the Pi0 and Pi05 pages.

5. More Resources​

Provides SDK documentation, quantization toolchain, S600 user manual, and extended reading on Pi0 quantization and real-robot deployment; see References.

6. FAQ​

Common issues encountered in practice, such as BPU memory allocation failures and HBM model format errors, with troubleshooting steps and solutions; see FAQ.