D-Robotics RDK S100 S100P Developer Kit (80TOPS/128TOPS)

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.