- Integrated Computation and Control Robot Development Kit
- CPU+BPU+MCU Super Heterogeneous
- Support hardware-accelerated ROS2 feature package
- 200+ Open-source algorithms and application examples
- Provides a variety of offline and online mainstream Al large model development materials
Tutorial Link：https://www.yahboom.net/study/RDK_S100
RDK S100/S100P is a developer kit designed specifically for embodied intelligent robots. It employs humanoid-Inspired dual-brain architecture, integrating a high-performance CPU, an 80TOPS/128TOPS BPU(Brain Processing Unit), and a real-time MCU to achieve integrated computing and control. Onboard multiple interface for easy startup, it significantly enhances the robot perception, decision-making, and control capabilities in complex scenarios through heterogeneous computing and dynamic task scheduling, helping developers quickly build intelligent robot applications.
◆Integrates three core units: a CPU (6x Cortex-A78AE, 100K DMIPS), a BPU(Brain Processing Unit) (new Nash architecture, 80 TOPS/128 TOPS), and an MCU (4x Cortex-R52, 6K+ DMIPS).
CPU handles real-time scheduling and low-latency tasks;
BPU(Brain Processing Unit) is optimized for CNN/Transformer, supporting 160+ ONNX operators to improve AI inference energy efficiency;
MCU enables high-frame-rate joint control, forming a "brain-cerebellum" collaborative architecture that balances intelligence and real-time performance.
◆Efficient model collaboration is achieved through a three-tiered hardware division of labor:
CPU, BPU(Brain Processing Unit), MCU.Large models (vision/point cloud/LLM/VLM) are handled by the CPU and BPU(Brain Processing Unit) for perception and decision-making.
Small models (motion control) are executed in conjunction with the MCU and BPU(Brain Processing Unit), freeing up CPU resources.
Dynamic model switching across the perception-execution chain is supported, ensuring on-demand resource allocation in complex tasks. The MCU specializes in high real-time control, while the BPU(Brain Processing Unit) assists in computation and reduces its workload.
◆End-to-end millisecond-level response closed loop
Visually acquired data undergoes semantic detection (LLM/VLM) and state analysis via BPU(Brain Processing Unit)+CPU; the motion control model generates instructions through BPU(Brain Processing Unit) inference, and the MCU executes high-frame-rate joint closed-loop control.
Threechip collaboration achieves millisecond-level response from human commands to mechanical execution, adapting to dynamic scenario requirements.
◆Full-interface design for flexible storage expansion
Provides a classic 40pin interface + dual M.2 Key expansion interfaces, supporting JTAG debugging, camera expansion, and MCU interfaces.
Equipped with 64GB eMMC storage, 4×USB 3.0 ports and dual Gigabit Ethernet ensure high-speed data transmission, and supports Type-C quick-connect. The layered interface design caters to both engineering debugging and modular expansion needs.
◆Open-source algorithms cover multiple robot forms
Through 200+ open source algorithms and application examples, it quickly adapts to scenarios such as semi-humanoid/quadruped/humanoid/point-legged/bipedal robots and robotic arms (LeRobot), lowering the development threshold for multi-form robots and accelerating deployment from hardware to specific scenarios.
