Robot Controllers (Brain & Cerebellum) Research Report, 2025

Robot Controller Research: Brain-Cerebellum Integration Becomes a Trend, and Automotive-Grade Chips Migrate to Robots

The publisher has released the Robot Controllers (Brain & Cerebellum) Research Report, 2025. The report conducts research, analysis, and summary on: ① the definition, functions, structure, and classification of robot controllers (brain & cerebellum); ② the development history, integration solutions, and challenges of robot controller (brain & cerebellum) integration; ③ the development trends of robot controllers (brain & cerebellum); ④ the application of robot controller chips; ⑤ the controller deployment solutions of robot integrators; and ⑥ robot controller suppliers and their products.
 
By analogy with the structure and division of labor of the human brain, the control functions of robots are divided into the 'brain' and the 'cerebellum'. The brain is responsible for functions such as perception, planning, decision, interaction, and learning. Specifically, it processes massive amounts of multi-modal sensory information (e.g., vision, force, language), understands task objectives, and conducts complex logical reasoning and long-term behavior planning. This process is accompanied by natural interaction with the environment and continuous learning, which mainly relies on powerful AI algorithms and computing capabilities.

The cerebellum is responsible for motion control, coordination, feedback regulation, and stability. That is, it receives decision instructions from the brain, converts them into specific action commands, and controls motor joints at an extremely high frequency and precision to ensure smooth, stable, and safe movements. This part has extremely high requirements for determinism and low latency.

The integration of a robot's 'brain & cerebellum' refers to a collaborative working architecture between the robot's perception and decision system (brain) and its motion control system (cerebellum). It aims to simulate the division of labor and collaboration mechanism between high-level cognition (brain) and motion coordination (cerebellum) in the human nervous system, and realize a 'perception-decision-execution' closed loop through software-hardware collaboration. The integrated system can enhance the capability of long-range operation tasks.

Leading Robot Integrators Have Begun to Adopt 'Brain & Cerebellum' Integration Solutions

Unitree Robotics initially adopted a separate deployment solution for the 'brain & cerebellum' in its early launched robot products, such as the first-generation consumer-grade quadruped robot Unitree A1, the first-generation industrial-grade quadruped robot Unitree B1, and the first general-purpose humanoid robot Unitree H1. However, all products launched in the past year have adopted the 'brain & cerebellum' integration solution. For example, the industrial-grade quadruped robot Unitree A2, launched in August 2025, is equipped with an 8-core high-performance CPU as standard to handle all basic platform functions; it is also equipped with an Intel Core i7 specifically to support users in secondary development; additionally, a computing power expansion dock is available as an option to meet high-level needs such as AI algorithm deployment and real-time data processing.

AgiBot released its commercial humanoid robot Lingxi X2 in March 2025, which also adopts the 'brain & cerebellum' integration solution. Its basic computing power board is equipped with two Rockchip RK3588 chips, with a computing power of 6 TOPS per chip and support for INT4/INT8/INT16/FP16 mixed operations. The Rockchip RK3588 has an 8-core CPU (4Cortex-A76 + 4Cortex-A55), a GPU with a built-in high-performance 2D image acceleration module, and an NPU with 6 TOPS of computing power. In addition, the flagship version (Ultra) of Lingxi X2 is equipped with an additional high-computing-power board, the NVIDIA Jetson Orin NX, with a computing power of 157 TOPS.

FOURIER Robotics released its open-source humanoid robot N1 in April 2025, which also adopts the 'brain & cerebellum' integration solution. Its computing chip is the Intel Core™ i7-13700H, which has 14 cores (including 6 performance cores and 8 efficiency cores) and a total of 20 threads, with an integer operation capability of 96 GOPS and a floating-point operation capability of 69 GFLOPS.

Multiple Suppliers Have Jointly Launched Integrated 'Brain & Cerebellum' Products for Robots

On June 11, 2025, D-Robotics, a subsidiary of Horizon Robotics, released the industry's first single-SoC computing-control integrated robot development kit RDK S100 to meet the high-level needs of embodied AI application scenarios (such as humanoid robots and bionic robots) for perception accuracy and generalization ability, with a market price of 2,799 yuan. The RDK S100 adopts a human-like 'brain & cerebellum' architecture of 'CPU+BPU+MCU'. The 'brain' consists of a CPU and a BPU, mainly responsible for high-level cognitive computing such as image recognition and target behavior planning. Among them, the CPU is 6xArm z with a computing power of 100K DMIPS, and the BPU adopts Horizon's Nash architecture, providing an equivalent computing power of 80 TOPS (INT8 quantization). The 'cerebellum' consists of an MCU (4xArm Cortex-R52+@1.2GHz) with a computing power of 6K+ DMIPS, focusing on processing tasks with extremely high requirements for low latency and real-time performance, such as motion control and motor drive.

In July 2025, D-Robotics announced a collaboration with iMotion, Luxshare Precision, SIM Technology, Huaqin Technology, G-Pulse, and other companies to develop robot controllers based on the RDK S100 series intelligent computing platform. Among them, iMotion's controller product 'iMotion iRC100' mainly targets intelligent robots in scenarios such as industry, services, and logistics; Luxshare Precision's controller product 'CORTEX100' mainly targets scientific research and industrial robots; SIM Technology's 'AI Computing Terminal CD01' mainly targets scenarios such as industrial quality inspection, park monitoring, and educational research; Huaqin Technology's 'Humanoid Intelligent Robot Main Controller' mainly targets scenarios such as smart factories, medical rehabilitation, and special operations; G-Pulse's 'GRC1.0' mainly targets industrial automation scenarios.

In April 2025, Intel released an embodied AI 'brain & cerebellum' integration solution, which includes the Core Ultra series processors, an embodied AI software development kit, and an AI acceleration framework. This solution enables efficient integration of perception, interaction, task planning, and motion control in a unified system. The Intel Core Ultra series processors integrate CPU, GPU, and NPU. The CPU supports complex motion control, the GPU is used for embodied AI to process complex tasks such as environmental perception and task recognition, and the NPU is used to undertake AI tasks that require long-term operation, such as speech recognition, real-time visual processing, and sensor data analysis, thereby realizing the integration of computing and control for the 'brain & cerebellum'.

Automotive-Grade Intelligent Driving Chips Are Migrating to Robots

Automotive-grade intelligent driving chips and robot chips share technical homology. Manufacturers of automotive-grade chips can leverage their existing production processes and supply chain systems to reduce production costs. D-Robotics, a subsidiary of Horizon Robotics mentioned above, is a good example of this.

In addition, in May 2025, SemiDrive launched the D9-Max based on technologies accumulated in the automotive-grade field, such as the ISO26262 ASIL-D functional safety certification process, hardware ECC verification, and independent safety island design. This product provides high-reliability underlying computing support for scenarios such as industrial control and service robots. The D9-Max adopts a multi-core heterogeneous computing architecture of 'CPU+GPU+NPU+DSP+MCU', integrating a 12-core Cortex-A55 (with a main frequency of 2.0GHz), a 3-core dual-core lockstep Cortex-R5F (800MHz), an NPU with 8 TOPS of INT8 computing power, and a GPU with 115 GFLOPS of FP32 computing power. It can run multiple operating systems such as Linux, Android, and RTOS simultaneously.

In 2024, Black Sesame Technologies successively released the Huashan A2000 'brain' chip and the Wudang C1236 'cerebellum' chip for humanoid robots. Currently, these two chips have started installation and testing on platforms such as Tianwen Robots. On August 28, 2025, Black Sesame Technologies reached a strategic cooperation with the robot company DEEP Robotics. Black Sesame Technologies will take its self-developed high-performance automotive-grade computing chips as the core to provide DEEP Robotics with embodied AI industry solutions, high-performance chips, and edge computing capabilities. They will jointly explore embodied AI solutions for scenarios such as ship inspection, smart construction, and scientific research and teaching, and actively promote overseas market expansion, with plans to complete the implementation of the first batch of overseas demonstration projects within 2025.