After the "big model", embodied intelligence is believed by the industry to be the next wave in the field of artificial intelligence.



Recently, at the "Internet 3.0: Future Internet Industry Development Forum" sponsored by the Beijing Association for Science and Technology, Wang Tianmiao, honorary director of the Robotics Research Institute of Beijing University of Aeronautics and Astronautics and president of Zhongguancun Zhiyou Research Institute, proposed that embodied intelligence is moving towards a "singularity" moment, and he predicted that by 2035, embodied intelligence will reach a market size of 38 billion dollars. However, before that, a series of technical challenges still need to be addressed.

Embodied intelligence is a further extension of artificial intelligence in the physical world. It is a machine or system that integrates the organic intelligence of thinking, perception, and action. It not only has the ability of human-computer interaction and natural language understanding, but also can calibrate and interact with time-varying environments and objects in real-time through perception, cognition, decision-making, and assist people in completing corresponding decision-making and action tasks.

Wang Tianmiao introduced that in the past five years, some important technological advancements have provided favorable conditions for the development of embodied intelligence, such as the implementation of semantic logic reasoning in large models, making it possible for robots to perform complex tasks; The structure, motor drive, and visual perception of humanoid robots have made breakthroughs, making robots expected to play a role in manufacturing, commerce, hazardous operations, home care, and other fields; Brain computer interface technology can assist people with disabilities in controlling external devices by encoding and decoding EEG signals, and these advances have pushed embodied intelligence to the door of technological explosion. But he also bluntly pointed out that currently, robots are still relatively "intellectually disabled" in many environments, unable to understand scene requirements, and unable to eliminate interference.

So, people are trying to improve the intelligence level of robots in reality using the "perceptual big model" and trying to equip various types of robots with a universal "brain". This can be seen as some progress in consulting conversational embodied intelligence, and many people have already experienced it in hotel, exhibition and other settings. However, such "embodied intelligence" cannot yet act and work autonomously. To achieve substantial breakthroughs, in Wang Tianmiao's view, there are still some challenges that need to be overcome.



"We have made progress in language and visual models, but we have not yet found similar models in the field of robotics. The breakthrough in robot models is the foundation for the development of universal artificial intelligence robots," said Wang Tianmiao.

He said that from an evolutionary perspective, there is still a need for breakthroughs in how to coordinate the three computational spaces of language cognitive intelligence, behavioral intelligence, and visual intelligence, and efficiently utilize real, high-quality, and massive data.



If these technological bottlenecks are overcome, which scenarios may embodied intelligence first land in?

According to Wang Tianmiao's understanding, people do call for embodied intelligence to be applied in different scenarios. But the current consensus in academia and industry is that the purpose of embodied intelligence is to improve productivity, not to replace human work. In the future, its main application scenarios may focus more on tasks that humans are unwilling to do or unable to perform.

He believes that the academic community focuses on solving complex needs, while the industrial community emphasizes production efficiency. Therefore, the application of embodied intelligence may start with business scenarios, including logistics, low altitude economy, etc. In addition, agricultural labor and hazardous chemical and nuclear industries in industrial scenarios may also be implemented earlier. He said, "Only in the end can we possibly move towards the family, because moving towards the family involves safety issues, cost issues, ethical issues, and so on."