Mingguo Zhao：Kicking Like Humans: Legged Embodied-intelligence of Humanoid скачать в хорошем качестве

Mingguo Zhao：Kicking Like Humans: Legged Embodied-intelligence of Humanoid

The RoboCup initiative aims to develop a humanoid robot soccer team that will try to beat the human World Cup champions around 2050. The process of achieving this ambitious goal will inevitably lead to many technological advances and lay the foundation for solving some of the major social problems, especially when dealing with challenges such as an aging population, which humanoid robots may provide. In line with the goals of RoboCup, this report views soccer playing as the core problem of embodied intelligence in humanoid robots. It defines three levels of ability to measure this embodied intelligence: 1. Flexible motion based on Proprioception: Humanoid robots use proprioception to perform agile actions like humans, such as running and static ball kicking. 2. Ball handling skills based on environment perception: The robot completes ball handling through multimodal perception, such as dribbling and dynamic kicking. 3. Multi-agent-based teamwork: Robots can collaborate and confront each other by forming themselves into games, such as passing and running coordination. This report elaborates on the definition of three levels of capability in the development route of embodied intelligence in humanoid robot legs. By analyzing the actual situation of humanoid teams at RoboCup 2024 and 2025 German Open, we summarize the current state of the art and look forward to advancing this challenge through imitation learning, reinforcement learning, world models, and multi-agent cooperation. Mingguo Zhao, professor in the Department of Automation at Tsinghua University, where he serves as the director of the Robot Control Laboratory and the Neuromorphic Robotics Center. He has published over a hundred research papers and holds over ten patents. In humanoid robotics, Zhao has introduced innovative methods such as virtual slope walking, generalized model predictive control, and whole-body control with task priority transitions. His research has been applied in robot soccer competitions, where he has achieved runner-up awards in the RoboCup humanoid league. In the realm of neuromorphic computing, he utilized neuromorphic technology to develop a riderless bicycle, with the results featured on the cover of Nature. This work was recognized as one of China's top ten scientific advancements in 2019 and received major project funding from the STI2030 program.