Chinese startup Matrix Robotics has officially introduced MATRIX-3, its third-generation humanoid robot, representing a significant advance in physical AI.
The platform is a complete from-scratch overhaul of algorithms, hardware, and applications, moving humanoid robots beyond rigid task performance toward flexible, real-world interaction.
Built to be safe, autonomous, and highly adaptable, MATRIX-3 integrates biomimetic perception, precise manipulation, natural human-like motion, and a new cognitive core that supports zero-shot learning.
The company says the robot is designed to operate beyond factories, extending into commercial, healthcare, and household environments.
Advancing Adaptive, Human-Like AI
MATRIX-3 is framed as a significant step forward in physical artificial intelligence. Designed as a safe, autonomous, and adaptable platform, it handles complex, human-like tasks in real-world conditions.
“Our vision with MATRIX-3 is to bring machine intelligence into human environments in the most natural and secure way possible,” said Allen Zhang, CEO of Matrix Robotics, in a statement.
MATRIX-3 features a biomimetic interface with “skin” and touch, covered in flexible 3D fabric that houses an underlying sensor network. This design absorbs physical contact and monitors impact forces in real time, enhancing safety during close interactions with people.
Advanced Visual–Tactile Perception for Precise and Safe Manipulation
A multimodal perception system combines high-sensitivity fingertip sensors with advanced vision, creating a visual–tactile loop that lets MATRIX-3 safely handle fragile and flexible objects.
MATRIX-3 also marks a major advance in mobility and manipulation. Its dexterous 27-DOF hand mimics human anatomy with lightweight, cable-driven actuation for fast, precise motion. This allows the robot to handle everyday tools, operate delicate equipment, and manipulate soft materials like fabrics.
Whole-body movement uses a natural gait generated by a motion control model trained on human motion-capture data. Built-in linear actuators deliver high power density with minimal noise, allowing for stable, agile, and well-coordinated full-body motion.
Matrix’s intelligence division built a new cognitive core that underpins these abilities. Its neural network enables zero-shot learning, letting MATRIX-3 perform new tasks from natural-language instructions without task-specific training.
Using universal intelligent manipulation, the robot can autonomously plan its grasps, modulate force in real time, and navigate around obstacles through smooth hand–eye coordination.
Real-World Performance Yet to Be Verified
So far, videos show MATRIX-3’s capabilities, but researchers have not yet verified its real-world performance; consistently replicating its hand dexterity would mark a major robotics breakthrough.
Matrix Robotics has launched an early access program for select industry partners, with pilot deployments of MATRIX-3 expected to begin in mid-2026.
Read the original article on: Interestingengineering
Read more:Rainbow Around Nearby Dead Star Puzzles Scientists