According to AIbase, leading domestic robot company Unitree has made a key breakthrough in enhancing the "brain" decision-making capabilities of robots. According to Qicha Cha information, the patent "A Robot Interaction Control Method and Server Based on Diffusion Strategy" applied for by Unitree has been officially authorized. This patent aims to solve the problem of action understanding deviation in existing robot models due to the lack of accurate awareness of future states.
The core of this technology lies in building a three-model system consisting of scene understanding, scene interaction prediction, and diffusion strategy decision-making. Through this system, robots can predict the changes their actions will bring to the surrounding environment and generate detailed interaction scenario change information. After processing this information, the system can directly generate control instructions, achieving end-to-end precise control from perception to action.
This invention not only gives robots the ability to perceive future states but also greatly enhances the accuracy and reliability of control. By simplifying the control process and shortening the response time, Unitree's robots will demonstrate stronger scenario adaptability, laying the technical foundation for dexterous operations in complex environments.
