The latest video of Unitree Company shows that their small human -shaped robots are proficient in the Kungfu martial arts, including punching and kicking the bridge but still keeps balance perfectly.
Chinese company Unitree is in the process of upgrading the G1’s algorithm, allowing it to learn and perform any movement. The most recent update seems to expand the ability to balance and the portfolio of human -shaped robots, Interesting Engineering On February 27, reported.
In the video, G1 shows agility and balance, performing a smooth movement, accurate foot movement despite being interrupted. In January, Unitree introduces the G1 robotic and running system, showing that it can maintain stability and accurately control movement on the steep distance and rough terrain. Now, upgrading new algorithms to improve motion, allowing robots to improve their ability to learn and conduct complicated tasks.
G1 reveals the ability to coordinate movement and special flexibility with freely rotating joints 23 degrees. With the martial arts level, the company warns that users should not adjust the robot in any way that can threaten others or teach it true fighting skills. According to Unitree, G1 is built to solve difficult work, dirt and repetition in many contexts, including family, factory and hospital. It is embodied for Unitree’s concept that human -shaped robots can become a useful companion in work and life.
The training process starts with the virtual context, using NVIDIA’s Isaac Simulator simulation application, helping robots learn complicated behavior. This method includes building a digital twin version of a human -shaped robot, using motion photos and video data to observe human actions. These behaviors are trained in virtual environment through strengthening learning. Thanks to the process called Sim2real converts the simulation action into a real world application, the capabilities are then transmitted to robots.
In order to improve the natural movement of human-shaped robots such as H1, H1-2, and G1, Unitree Robotics models announce a set of open-body data. This information enhances the flexibility and coordination of robots, allowing them to imitate human motion like dancing. Using Lafan1 motion technology, the data kit is fully compatible with the key robot model of Unitree, providing the same human motion. The data kit includes a changing algorithm combining mesh -style interaction technology and reverse movement. Through consideration of limiting posture, joint position and speed, new methods help maximize the motion of the robot.
Unitree hopes that researchers and interested people can learn how to use human -shaped robots in the real world through allowing the community to access the data kit. H1 and H1-2 high-performance samples are produced to own advanced capabilities. In contrast, the G1 model offers flexible options with more reasonable prices for research and development.