As the wave of embodied intelligence and humanoid robots continues to surge, enabling robots to "move steadily," "grasp accurately," and "operate for extended periods" has become a core challenge that the industry urgently needs to overcome. In traditional perception, chip capabilities are often divided into the "brain" responsible for logical reasoning and the "cerebellum" responsible for motion control. During a recent technical demonstration and exchange at the 4th China (Shanghai) Embodied Intelligence Robot Industry Conference and Exhibition in 2026, Renesas Electronics provided a detailed showcase of its system-level technology puzzle at the robot "cerebellum" level—from an edge AI development board with 80 TOPS of sparse computing power, to single-chip integrated 9-axis motor control, and further to high-precision eddy current sensors and hard real-time operating systems. This series of solutions not only demonstrates Renesas's profound accumulation in the industrial and automotive chip fields but also reflects the current robot industry's urgent demand for highly integrated and highly reliable solutions.
Vision and Force Fusion: How Edge AI Empowers Robotic Hands with "Autonomous Grasping" Capabilities
The dexterous manipulation of robots is inseparable from the coordination of "eyes" and "hands." The vision-guided autonomous grasping solution showcased by Renesas is primarily based on its RZ/V2H RDK development board. In terms of AI (Artificial Intelligence) computing power, this chip features 80 TOPS of sparse computing power and approximately 8 TOPS of dense inference capability. Thanks to an exceptionally high energy efficiency ratio of 10 TOPS/W, the development board can stably run complex AI vision tasks without the need for active cooling fans.
Its working logic constructs a clear edge-side closed-loop chain: the depth camera at the end of the robotic arm is responsible for executing AI vision detection and object classification, such as distinguishing objects of different materials like cola, carrots, and sponges. After completing target recognition, the chip immediately runs the Inverse Kinematics algorithm to convert the position in the camera coordinate system into the real-world coordinate system of the robotic arm, and sends commands to the joint actuators via the CAN bus.
Notably, this system is not a single visual closed loop but integrates force feedback. For different objects, the system has built-in differentiated grasping force thresholds (e.g., approximately 1000 grams for cola, 500 grams for carrots, and 150 grams for sponges). The moment the robotic hand touches an object, the independent force sensors at the finger joints transmit data in real time, achieving adaptive grasping with "vision classifying first, followed by force control closing the loop."
Renesas pointed out that the RZ/V2H series chips have taken substantial steps in commercialization and are currently in the market introduction and customer validation phase. However, the clear market positioning of this chip is to serve as the robot's "cerebellum," responsible for motion control, real-time balancing, and joint-level execution. To fill the computing power gap required by the "brain" for general large models, Renesas has planned the next-generation R-Car series high-computing-power chips.
Microscopic Perception of Tactile and Position: Electronic Skin and High-Precision Encoders
Beyond vision and force, the robot's delicate perception of the external environment also relies on tactile and position sensing. In the dexterous hand solution, Renesas introduced an "electronic skin" (impedance/capacitive tactile array) driven by the RAA2S4704 chip. This solution communicates with the RZ/V2H main controller via 6 SPI interfaces and can simultaneously decompose both resistance and capacitance variation signals, making it more accurate than single-contact methods.
There have been extensive discussions in the industry regarding the application scenarios for electronic skin. In this demonstration, Renesas clarified the engineering boundaries of its current stage: the electronic skin is currently positioned as an HMI (Human-Machine Interface) and safety warning interface, primarily used to identify contact between humans and machines (such as long-press reset, obstacle avoidance, etc.), rather than being directly used for closed-loop control of grasping force. The true closed-loop grasping force still relies on independent force sensors distributed at the finger joints. This clear division of labor—"tactile for interaction, force for closed-loop"—reflects the pragmatic logic of engineering implementation.
In terms of position sensing for robot joints, Renesas has extended its mature technologies from the automotive field to robotic application scenarios. Its high-precision position sensor is based on eddy current induction technology. Unlike traditional Hall sensors, this solution completely eliminates the need for magnets, which not only reduces costs but also eliminates interference from stray magnetic fields. In terms of performance, this sensor features a 19-bit effective resolution and 15-bit absolute accuracy, supporting a wide operating temperature range from -40°C to 160°C, and can achieve high-precision positioning of 0.02 degrees in the demonstration. It is reported that in a single robot joint, two sensors are typically required: one located at the motor end (high-speed, incremental) to detect rotational speed, and the other at the end of the reducer (low-speed, absolute) to detect absolute position. The RAA2P series sensors have launched customized products for various scenarios such as industrial robots, automotive EPS steering, and two/three-wheelers.
Control and Communication Integration: Single Chip Overcomes 9-Axis Motor and Industrial Network Challenges
For multi-axis industrial robotic arms or collaborative robots, highly integrated motor control solutions mean smaller form factors, lower costs, and better reliability. The Renesas RZ/T2H series MPU is designed precisely to address this pain point.
In terms of architectural design, the RZ/T2H adopts a highly targeted heterogeneous multi-core design. The chip integrates four Cortex-A55 (1.2GHz) cores dedicated to running the Linux operating system, host computer interfaces, and a full suite of industrial network protocol stacks (covering EtherCAT, PROFINET RT/IRT, EtherNet/IP, CC-Link IE Basic, TSN, OPC UA over TSN, etc.); the other two Cortex-R52 (1GHz) cores are exclusively dedicated to running the RTOS (Real-Time Operating System), handling all real-time current loop, speed loop, and position loop control.
Traditional solutions often require complex combinations of "CPU+FPGA" or multiple MCUs (Microcontroller Units) to achieve multi-axis control and network communication, whereas the RZ/T2H can achieve high-speed control of 9-axis motors with a single chip. Renesas emphasized in the exchange that integrating EtherCAT slave or master functions within a single chip is an extremely difficult engineering challenge in the industry. This is because EtherCAT has extremely stringent requirements for the synchronization of the internal bus, requiring solutions for anti-interference, bus conflicts, and high-precision synchronization. The realization of the single-chip solution allows the entire controller to significantly reduce its size, and it can even be stuffed into the robot's thoracic cavity, providing a hardware foundation for integrated joint design.
Software Digital Foundation: The Indispensability of Hard Real-Time Operating Systems
The efficient operation of the "cerebellum" is inseparable from the strong support of the underlying operating system. In this solution demonstration, iSoft, Renesas's ecosystem partner, showcased the AUTOSAR OS and EtherCAT master station solution specifically adapted for the RZ/T2H.
In typical robot control, although general Linux RT solutions are widely used, they have obvious shortcomings in extremely stringent real-time control scenarios. Data presented by iSoft shows that its dedicated real-time operating system can stably achieve a 7-node 125-microsecond synchronization cycle when executing multi-node cascaded EtherCAT communication, with the maximum interrupt jitter controlled within 4 microseconds. By comparison, the jitter of conventional Linux RT solutions in similar scenarios is about 700 microseconds, meaning the real-time performance of the new solution has improved by two orders of magnitude.
The realization of this extremely high real-time performance stems from the operating system's deep optimization at the kernel level for interrupt task switching, task affinity, and kernel streamlining. Meanwhile, to accommodate the extensive demands of the current developer community, the system is compatible with the Linux ecosystem at the underlying level, supports ROS 2 and the Tengine AI inference library, and comes with preset capabilities for functional safety and cybersecurity. The cooperation model between Renesas and iSoft not only provides chip hardware but also offers a "hardcore real-time digital foundation" and comprehensive "nanny-style" turnkey engineering services. This hardware-software collaborative approach is critical for shortening the robot R&D cycle.
Conclusion
From the visual AI computing power of the RZ/V2H to the single-chip multi-axis control of the RZ/T2H; from high-precision eddy current sensors to the tactile interaction of electronic skin; and further to the escort of iSoft's hard real-time OS, this "cerebellum" robot solution from Renesas outlines a very clear technological path: compared to merely emphasizing the computing power parameters of the "brain," this solution focuses more on the hardcore capabilities of underlying robot motion control, sensor fusion, and system integration.
Currently, the development of the humanoid robot industry is moving from "proof of concept" into the deep-water zone of "engineering validation." The industry has put forward nearly extreme requirements for high integration, high reliability, low power consumption, and real-time performance. Renesas's technical demonstration this time is not only a reuse and expansion of its own automotive and industrial chip technology assets but also reveals that the true implementation of robots requires chip manufacturers, sensor suppliers, and software ecosystem partners to jointly weave a solid underlying technology network. When everything returns to its essence, the "IQ" of a robot may depend on the brain, but its "EQ" and "survival ability" are firmly controlled by an excellent cerebellum and sensing system.