The Kungfu robot defeats people in the couple

Chinese robot company Unitree Robotics shares videos showing the agility and balance of the compact G1 -shaped robot. In the 15 -second clip, a person holding a stick continuously backed when the robot swung his leg to fly weapons while performing a 720 -degree rotation, Interesting Engineering On March 5 reported. A week ago, Unitree’s G1 robot learned how to master the martial arts Kungfu, now it can confront humans and win.

Unitree constantly updates the G1’s algorithm, allows it to learn and handle many movements. The latest update enhances the ability to balance and expand its movement range. In February, G1 also showed good agility and balance in a video with smooth dance and accurate steps even when being disturbed. In January, Unitree revealed the G1’s walking and running system, highlighting its stability and good control on the slope and less flat surface.

With freelance rotation (DOF) 23 degrees and improved stability, G1 reveals the level of movement and special flexibility. However, Unitree also recommends not adjusting robots in a dangerous way or teaching it really fighting skills. Designed for difficult tasks or repeated at home, factory and hospital, G1 is a testament to the view of a human -shaped robot as a companion in the work and life of Unitree.

1.32 m high and weighs 35 kg, G1 can be folded with dimensions of 69 x 45 x 30 cm for easy storage. The robot model is equipped with many advanced technologies, including 3D Lidar, RealSense depth and noise reducing microphone to respond to voice commands. G1 works thanks to the 9,000 mAh battery that provides two hours of operation with fast battery change for longer use. The robot also owns a central processor (CPU) of 8 high -performance core, joints in the arms, legs and stems, allowing 23 degrees to rotate. The design of G1 helps it move quickly and reach a walking speed of 2 m/s.

The training process begins in virtual environment, using NVIDIA’s Isaac simulation tool, in which robots learn complicated behavior before manufacturing. Through enhancement learning, G1 perfect skills in simulation. The process of Sim2real moves the action that has learned to real robots, allowing practical applications. Unitree also shared the open body data set to enhance the natural movement of human-shaped robots such as H1, H1-2, and G1. The data kit will improve coordination and flexibility.

With Lafan1 motion photography technology, the data set is fully compatible with Unitree’s main robot. It includes a changing algorithm, integrating interactive grid and reversing motion, optimizing motion through limited consideration in posture, joint position and speed. Thanks to the open source data set, Unitree wants to encourage researchers and those who are interested in exploring new applications of human -shaped robots in the actual context.