Tech • IA • Crypto
China Just Built a Two-Brain AI Robot: One Body, Two Minds
8/10TL;DR
China, Vietnam, and NVIDIA unveiled new humanoid robotics platforms, signaling intensifying global competition across compact AI-driven robots, service humanoids, and high-performance research systems.
KEY POINTS
China’s compact humanoid “Pi”
Jaka Robotics introduced Pi, a 4-ft (1.22 m), 42 kg humanoid designed for research and real-world testing. With 27 degrees of freedom and compact joint modules reduced by up to 27%, the robot emphasizes mobility, efficiency, and modular experimentation. Its arms can handle 3 kg payloads, while joints deliver up to 120 Nm torque, enabling stable walking and manipulation tasks.
Dual-brain architecture for embodied AI
Pi’s standout feature is its “fusion brain” architecture, splitting cognition and motion. A high-level cerebrum system handles reasoning, vision, and language models, while a low-level cerebellum system manages real-time motor control via EtherCAT with millisecond latency. This separation reflects a broader push to unify AI reasoning with precise physical execution in humanoids.
Jaka’s shift beyond industrial cobots
Founded in Shanghai in 2015, Jaka built its reputation on industrial collaborative robots like the Zu and Pro series. With Pi and earlier K-series humanoids, the company is expanding into embodied AI, integrating machine vision, force sensing, and language models into flexible humanoid platforms suited for logistics and service environments.
Vietnam enters with service-focused humanoid “Dino”
Vin Big Dynamics debuted Dino, a humanoid aimed at security, surveillance, and household assistance. While specifications remain undisclosed, the robot integrates autonomous navigation, environmental awareness, and manipulation, targeting both public and domestic use—two challenging domains requiring different safety and interaction capabilities.
Real-world pilot in outdoor environments
Dino was tested at Vinpearl Safari Phu Quoc as an autonomous guide. It operated outdoors using multilingual interaction and real-time perception, navigating crowds and changing conditions. The deployment demonstrated sustained interaction in unpredictable environments, a key hurdle for service robotics beyond controlled lab settings.
Vietnam’s in-house robotics stack
Supporting Dino, Vin developed components like the VDM 80 actuator (under 1 kg, 48V, 10,000-hour lifespan) and a robotic hand with 11 joints and force sensing. These components form a vertically integrated platform combining hardware, AI training data, and control systems for future humanoids.
Industrial humanoid VRH3 targets factories
Vin Robotics introduced VRH3, a third-generation humanoid for industrial automation. With 31+ actuators, dual onboard edge computers, and payload capacity of 6–8 kg, it is designed for assembly, transport, and operational support in dynamic environments.
Teleoperation via VR control
VRH3 includes a motion-capture teleoperation system embedded in a VR headset, allowing real-time control without external tracking. This enables applications in hazardous or remote environments, where human operators can guide robots safely and intuitively.
NVIDIA unveils open humanoid platform
NVIDIA introduced an open reference humanoid built on Isaac GR00T, using the Unitree H2 robot. Standing nearly 6 ft tall with 75 total degrees of freedom (including dexterous hands), the platform is designed as a full-stack solution for robotics research rather than a single product.
High-performance onboard AI computing
The system runs on Jetson AGX Thor T5000, delivering 2,070 FP4 teraflops with a Blackwell GPU, 14-core ARM CPU, and 128 GB memory. This enables real-time processing for perception, control, and learning directly onboard, reducing reliance on external compute.
End-to-end robotics ecosystem
NVIDIA’s platform integrates Isaac Sim, Isaac Lab, Isaac ROS, and teleoperation tools, covering data collection, simulation, training, and deployment. It also includes foundation models for humanoid reasoning, allowing researchers to build multi-task robotic behaviors more efficiently.
Adoption by leading institutions
Organizations such as ETH Zurich, Stanford Robotics Center, UC San Diego, and AI2 are adopting the platform. The approach emphasizes open collaboration, aiming to accelerate progress through shared tools and standardized hardware.
CONCLUSION
The latest humanoid launches show a clear shift toward integrating AI reasoning, physical control, and real-world deployment, with competition spanning compact systems, service robots, and high-performance research platforms.