Compute-Control Integration

Putting AI inference compute and real-time motion control on the same chip or the same board.

Traditional robots often use two separate pieces of hardware: one compute platform runs perception and decision-making models, while a separate MCU (microcontroller) or motion-control board handles millisecond-level joint control, with the two communicating over Ethernet or a bus. Compute-control integration means combining both onto the same SoC (system on chip) or the same main board: a high-performance CPU/NPU handles the “brain's” model inference, while a real-time core on the same chip handles the “cerebellum's” motion control. The benefits are less board-to-board communication latency, less wiring, and lower cost and power draw; the challenge is keeping the control task's real-time performance and functional safety free from interference by the AI workload. Products like D-Robotics' RDK S100 are marketed around this feature.

ExampleD-Robotics' RDK S100 integrates a 6-core Cortex-A78AE CPU, an 80-TOPS BPU, and a 4-core Cortex-R52+ MCU on a single SoC, according to the company's own announcement.

Also called

Integrated Compute-Control, Combined Brain-Cerebellum Chip

Related

Brain–Cerebellum Architecture · D-Robotics RDK S100 · System on Chip (SoC) · Microcontroller Unit (MCU) · Real-Time Control · Onboard Compute Platform

Sources

地瓜机器人发布首款单SoC算控一体化机器人开发套件（36氪） (Chinese)
RDK S100 开发套件（地瓜机器人开发者社区） (Chinese)

As of

2025-06