China Just Built A Two Brain AI Robot: One Body, Two Minds

Key Concepts

• Embodied Intelligence: The integration of AI into physical robotic bodies to enable interaction with the real world.
• Dual-Brain Architecture: A system design separating high-level AI reasoning (cerebrum) from low-level real-time motion control (cerebellum).
• Degrees of Freedom (DoF): The number of independent movements a robot can perform; higher DoF allows for more complex, human-like dexterity.
• Heterogeneous Computing: Using different types of processors (e.g., CPUs, GPUs) to handle specialized tasks like AI logic versus motor control.
• Vertical Integration: A manufacturing strategy where a company develops all key components (mechanical, electrical, software) in-house.
• Teleoperation: Controlling a robot remotely, often using motion capture or VR, to perform tasks in hazardous or complex environments.

---

1. Jaka Robotics (China): The Pi Humanoid

Jaka Robotics has shifted from industrial "cobots" to humanoid research with the Pi, a compact 4-ft (1.22 m) robot weighing 92 lb (42 kg).

• Hardware: Features 27 degrees of freedom with integrated joint modules 15–27% smaller than previous generations. Knee joints provide 120 Nm of torque, and arms support a 3 kg payload.
• Fusion Brain Architecture:
  • Cerebrum: Handles AI reasoning, Large Language Models (LLMs), and vision perception.
  • Cerebellum: Manages real-time movement via an EtherCAT-based network with millisecond-level latency.
• Strategic Context: Jaka is leveraging its history in industrial automation (Zoo and Pro series) to transition into embodied AI, aiming to bridge the gap between industrial reliability and human-like interaction.

2. Vin Group (Vietnam): Dino and VRH3

Vietnam has entered the humanoid race with two distinct platforms developed by the Vin Group ecosystem.

• Dino (Vin Big Dynamics):
  • Focus: Security, surveillance, and household assistance.
  • Real-world Application: Successfully piloted at Vinpearl Safari Phu Quoc as an autonomous robotic guide, demonstrating the ability to navigate noisy, unpredictable outdoor environments.
  • Components: Features the VDM 80 actuator (1 kg, 48V, 235 RPM) and a 6-DoF robotic hand with integrated force sensors.
• VRH3 (Vin Robotics):
  • Focus: Industrial automation, material handling, and assembly.
  • Technical Specs: 31 actuators, 6–8 kg payload capacity, and dual onboard edge computers for local data processing.
  • Vertical Integration: Vin Robotics emphasizes that all mechanical, electrical, and AI control frameworks were developed in-house.
  • Teleoperation: Demonstrated at ICRA 2026 using VR-based motion capture, allowing operators to control the robot naturally from a distance.

3. NVIDIA: Isaac GR00T Reference Platform

NVIDIA is providing a standardized "full-stack" platform to accelerate global humanoid research, moving away from fragmented hardware/software ecosystems.

• The Platform: Built on the Unitree H2 (6-ft tall, 150 lb), the system features 75 total degrees of freedom (31 in the body, 44 in the dual five-finger tactile hands).
• Compute Power: Powered by the Jetson AGX Thor T5000, utilizing a Blackwell-based GPU capable of 2,070 FP4 teraflops of AI performance, 14-core ARM CPU, and 128 GB of unified memory.
• Software Ecosystem:
  • Isaac Sim/Lab: For virtual training and testing.
  • Isaac ROS: Middleware for deploying AI policies.
  • Isaac Teleop: For capturing human demonstration data.
• Notable Quote: Steve Cousins (Stanford Robotics Center) stated: "Robotics moves fastest when researchers can build on open platforms, share code, and test ideas on real machines."

---

Synthesis and Conclusion

The humanoid robotics sector is undergoing a rapid evolution characterized by three distinct trends:

1. Miniaturization and Specialization: Companies like Jaka are proving that compact, research-focused humanoids can handle complex tasks through advanced joint integration.
2. Vertical Integration: Firms like Vin Robotics are prioritizing in-house development of hardware and software to ensure reliability and control over the entire robotic stack.
3. Standardization: NVIDIA’s reference design aims to solve the "fragmentation problem" by providing a high-performance, unified platform (Isaac GR00T) that allows researchers to focus on AI reasoning rather than basic hardware assembly.

The transition from lab-based demos to real-world deployment—such as the outdoor testing of Dino or the industrial use of VRH3—indicates that the industry is moving toward robots that can handle the unpredictability of human environments.