The Zhitong Finance App learned that the research department under Toyota Motor (TM.US), the world's largest automaker, and Boston Dynamics (Boston Dynamics), a humanoid robot research company under Hyundai Motor Company, are teaming up to accelerate the development of humanoid robots based on artificial intelligence technology. This move marks that Toyota, the well-deserved hegemon of the global automotive industry, is beginning to follow the steps of Tesla and new Chinese car builders to join the cutting-edge technology field of humanoid robots.
On Wednesday, they said in a statement that the partnership fully combines Toyota Research's expertise in learning large-scale machine-based behavioral models with Boston Dynamics' humanoid robot Atlas. They said that TRI and Boston Dynamics teams from Boston will study actual use cases of artificial intelligence training humanoid robots in fields such as human-computer interaction.
Various “AI humanoid robots” created based on AI models such as GPT under OpenAI, are also known as “embodied AI” in the technology industry — meaning that artificial intelligence (AI) is fully integrated with entities in the physical world, so that AI not only exists in a virtual environment, but also interacts with the real world through some physical form (such as humanoid robots, sensors, etc.). The focus of embodied AI is to sense, understand, and deeply interact with the environment and humans through physical devices, which can completely mimic human behavior in the real world.
Toyota said that the company has made a major breakthrough in artificial intelligence teaching robot learning, and Boston Dynamics was acquired by Hyundai as early as 2020, and its robot guard dog and human-like mobile robot arm for warehouse replenishment have achieved great commercial success. The major collaboration they announced can be described as a potential challenge to other smart robot participants, such as the Optimus robot introduced last week by Tesla, Toyota's strongest competitor in the electric vehicle field. Additionally, many humanoid robotics startups, including Figure AI, have attracted investors' attention and attracted billions of dollars in investment.
According to information, the collaboration between Toyota and Boston Dynamics will focus on basic research and focus on ultimate commercial use, but executives from the two companies declined to disclose the specific commercialization schedule or the specific budget for the project.
“This technology has huge prospects for the future of the company,” Toyota Chief Scientist Jill Pratt said in an interview. “Our work on generative artificial intelligence is a huge complement to the work done by Boston Dynamics.”
Pratt said the ultimate goal is to bring humanoid robots based on large AI models to factory assembly lines and homes, especially for geriatric care.
Automakers have been at the forefront of automating more assembly line processes to cut labor costs and improve worker safety systems. According to the latest data from the Frankfurt-based International Federation of Robotics (International Federation of Robotics), globally, the automotive industry is second only to electronics manufacturers deploying robot assistants in factories, and the automobile manufacturing industry accounts for about a quarter of new robotic devices in 2022.
Boston Dynamics Chief Technology Officer Aaron Sanders said that the dexterity of Boston Dynamics robots combined with Toyota's human-like behavior technology is a key difference between the company and other companies making intelligent humanoid robots.
“This will provide an iconic stage for the promise that many tech companies are now bringing humanoid robots to the world on a large scale.” he said.
Sanders said Boston Dynamics plans to deploy human-like robots focused on specific tasks in modern car plants “in the next few years,” while collaborating with Toyota to research long-term goals of humanoid robots based on artificial intelligence for multi-tasking.
The major introduction of generative artificial intelligence such as ChatGPT is expected to enable humanoid robots to master new skills more quickly based on learned behavior. Just as virtual auxiliary programs such as ChatGPT were developed from the GPT big language model, large-scale behavioral models will also be used for humanoid robot programming, and the level of perception and interaction of humanoid robots with the support of big language models such as GPT will increase exponentially.
“The Toyota Research Institute is testing large-scale robot behavior models to enable humanoid robots to perform multiple tasks.” Russ Tedlake, a professor at the Massachusetts Institute of Technology and head of robotics research at TRI, said that the ultimate goal is to have a robot programmed to beat eggs that can also handle any number of other things, such as folding shirts, with the support of large behavioral models expected by the big language model.
“We are making good progress on this,” Tedrake said. “We're working on this very quickly and very actively to demonstrate this new capability.”
“Godfather of AI” Hwang In-hoon: The next big wave of artificial intelligence will be embodied AI
With the advent of Sora, the big OpenAI Wensheng video AI model that can understand and simulate the physical world in motion, the AI big model will have more powerful physical world simulation capabilities and more comprehensive reasoning capabilities based on multi-modal standards, which may comprehensively promote the prosperity and development of the humanoid robot industry.
Nvidia's founder and CEO Hwang In-hoon, who has the title of “Godfather of AI,” recently stated: “The next big wave of artificial intelligence will focus on the field of robotics, and one of the most exciting developments is physical artificial intelligence.” “We're advancing the entire NVIDIA robotics stack, opening up technology platform access for humanoid robot developers and companies around the world to use the platforms, accelerators, and big models of artificial intelligence that best suit their needs.”
Nvidia can be described as spending its R&D priorities and R&D expenses in the field of humanoid robots in recent years, and is convinced that humanoid robots will be the core application scenario for AI technology. At the end of July, Nvidia announced that it will provide a set of services, models, and robot computing power platforms to the world's leading humanoid robot developers, AI big model developers, and software makers to develop, train, and build the next generation of humanoid robots.
Nvidia's new range of service products include a new NVIDIA NIM microservice and framework for humanoid robot simulation and deep learning, NVIDIA OSMO orchestration services to run massive workloads for multi-stage humanoid robots, and remote operation workflows that support new AI technology and humanoid robot simulation, allowing developers to train humanoid robots using very small amounts of real human demonstration data.
According to information, Nvidia's newly launched NIM microservice provides pre-built containers supported by NVIDIA's powerful inference software, enabling humanoid robot developers to reduce the deployment time of robot simulation solutions from weeks to just a few minutes. Two newly launched Nvidia AI microservices allow robot development experts to enhance the simulation workflow of generative physical AI in NVIDIA Isaac Sim, a robot simulation engineering reference application built on the NVIDIA Omniverse platform.
According to the forecast report recently released by Markets And Markets, the global humanoid robot market is expected to be only about 2 billion US dollars in 2024, and it is expected to rapidly expand to 13.25 billion US dollars by 2029, and the compound annual growth rate over 5 years is expected to exceed 45%. Musk made an amazing statement at Tesla's shareholders' meeting in June. He said that the “Optimus Prime” humanoid robot may become the core catalyst for Tesla's market value to rise to 25 trillion US dollars.
In August of this year, Figure AI, an AI humanoid robot startup with huge financial support from the tech community, such as OpenAI, Microsoft, and Nvidia, as well as Amazon founder Bezos, released Figure 02, a second-generation humanoid robot. As its name suggests, this currently the most advanced humanoid robot is the successor to Figure 01 launched in 2023.
Figure AI's new humanoid robot is fully integrated into the multi-modal AI model launched by OpenAI. The tech community generally regards this humanoid robot as a “walking ChatGPT,” and has many powerful abilities such as imitating human behavior, deep learning and thinking, and natural and efficient communication with humans.
According to Figure AI, this is the most powerful and versatile humanoid robot launched by Figure AI to the global market, combining the flexibility of human shape with OpenAI's most advanced artificial intelligence model. This humanoid robot's powerful imitation and deep learning capabilities mean that it can accurately perform various complex and dangerous tasks in enterprise production and manufacturing activities, and play a more intelligent role in human production activities to help humans increase productivity. Also, in the near future, it is expected to fully penetrate the home and become a consumer electronics product with a penetration rate comparable to that of iPhones and iPads.