As robotics move toward smaller, smarter, and more energy-efficient designs, choosing the right embedded computing platform becomes critical.

Microduck, a compact desktop biped robot, provides an interesting example of how Rockchip RK3566 can be used beyond traditional applications such as industrial HMI, POS terminals, and smart displays. The platform combines Linux computing, AI inference, sensor processing, wireless connectivity, and real-time actuator control in a compact robotic system.

According to the project information, Microduck is approximately 25 cm tall and 800 g, using RK3566 as its main controller to manage 15 servos with a 50 Hz control loop.

Why RK3566 for Robotics?

When AI and robotics are mentioned, high-performance platforms such as RK3588 or NVIDIA Jetson are often considered first.

However, not every robotic application requires large-scale AI computing.

Microduck demonstrates another approach: using a compact embedded SoC for lightweight AI inference and deterministic real-time control.

The RK3566 platform provides a combination of:

• Quad-core ARM Cortex-A55 CPU
• AI acceleration capabilities
• MIPI CSI camera interface
• ISP and video processing
• UART, I2C, SPI and GPIO
• Wi-Fi and Bluetooth connectivity
• Linux support

This combination makes RK3566 suitable for compact robotics, edge AI devices, and embedded control applications. The Microduck project uses RK3566 with Linux and an open-source runtime.

50Hz Real-Time Robot Control

One of the most interesting aspects of the Microduck design is its 50 Hz control loop.

A 50 Hz control frequency means that the robot has approximately 20 ms for each control cycle:

50 Hz → 20 ms per cycle

A simplified control workflow is:

Sensors → Robot State → AI Policy → Motion Commands → 15 Servos → Feedback

The RK3566 therefore needs to continuously process sensor information, calculate the robot’s state, execute the AI policy, and generate actuator commands within the control cycle.

This highlights an important point for robotics hardware design:

AI performance is not the only metric. Stable and predictable control-loop execution is equally important.

The project documentation identifies the combination of RK3566 + 50 Hz control + 15 servos as a key part of the system architecture.

Edge AI: Training on GPU, Inference on RK3566

Microduck follows a typical Sim-to-Real AI workflow.

The reinforcement learning model is trained using MuJoCo and PPO, then exported in ONNX format for deployment on the robot.

The architecture can be summarized as:

MuJoCo Simulation → PPO Training → AI Policy → ONNX → RK3566 → Robot Control

This is an important edge AI architecture because the computationally intensive training process does not need to run on the robot itself.

Instead, the embedded RK3566 platform focuses on real-time inference and control.

The policy network takes robot-state information such as IMU data, joint position, joint velocity and target motion as inputs, and generates motion commands for the robot’s actuators.

Integrated Sensor and Servo Architecture

Another interesting design feature is the integration of the IMU with the Dynamixel servo bus.

Instead of treating the IMU and servos as completely independent interfaces, the architecture allows sensor and actuator information to be managed through the same bus structure.

This can simplify data synchronization and is particularly valuable for a robotic system running a 50 Hz motion-control loop.

The analyzed project identifies the IMU as an LSM6DSV16X device integrated as a Dynamixel bus device.

Linux + Modular Runtime + OTA

The Microduck architecture is not simply a single robot application.

Its Linux runtime is organized around multiple services, including:

• Control daemon
• Servo daemon
• Camera daemon
• Wireless services
• Update daemon

This modular architecture makes it easier to separate hardware control, AI inference, communication, and system management.

OTA capability is another important part of the system. The project includes update mechanisms and deployment processes, creating a complete software lifecycle from development and release to deployment and robot updates.

For commercial robotics products, this architecture can provide a useful reference for building Linux-based embedded robot controllers with remote firmware updates.

What This Case Means for RK3566 ODM Development

From an ODM perspective, Microduck demonstrates that RK3566 can be considered for more than conventional embedded applications.

A customized RK3566 robotics mainboard can potentially integrate:

RK3566

↓

AI Inference

↓

Camera / IMU

↓

Servo / Motor Control

↓

Wi-Fi / Bluetooth

↓

Linux Runtime + OTA

The platform capabilities listed in the project include Cortex-A55 processing, AI acceleration, camera interfaces, UART-based Dynamixel communication, USB, wireless connectivity, GPIO and PCIe expansion.

For an ODM project, the hardware can be further customized according to the robot’s requirements, including 2GB/4GB RAM, eMMC storage, MIPI CSI, I2C/SPI, UART, RS485/CAN, wireless connectivity, and battery power management.

Key Takeaways

The Microduck project highlights three important ideas for embedded robotics:

1. RK3566 Can Go Beyond Traditional Embedded Applications

RK3566 can support Linux-based robotics applications involving AI inference, sensor processing and real-time motion control.

2. Edge AI Does Not Always Require High-End Computing

A lightweight reinforcement learning policy can be deployed directly on an embedded SoC, while model training remains on a more powerful PC or GPU platform.

3. Hardware Design Should Focus on the Complete Control Loop

For robotics, the key challenge is not simply CPU or NPU performance. The complete loop of:

Sensor → Data Processing → AI Inference → Decision → Actuator → Feedback

must operate reliably within the required control period.

KICKPI ODM Solution

The Microduck case demonstrates a practical combination of RK3566 + Edge AI + Linux + Robotics Control.

For robotics manufacturers developing a customized controller, KICKPI can provide RK3566/RK3568-based embedded board design and ODM development, including hardware customization, peripheral integration, Linux adaptation, AI deployment and product-level optimization.

Looking for a customized RK3566 robotics mainboard or Edge AI control board? Contact KICKPI to discuss your hardware requirements and ODM project.