Earlier this month, over 12,000 runners participated in a half marathon in Beijing, alongside more than 20 humanoid robots. The first robot to cross the finish line was awarded a prize of CNY 5,000 (approximately $691). This event symbolizes the start of a significant technological and military competition.
The concept of human-like robots has long been a theme in science fiction, dating back to Karel Čapek’s play, R.U.R. (Rossum’s Universal Robots), and extending to contemporary narratives like the Terminator films. Recent advances in robotics are beginning to turn this fiction into reality. Duke University roboticist Boyuan Chen emphasizes that “intelligence can’t be born without having the perspective of physical embodiments,” suggesting that physical interaction with the world is crucial for the development of AI.
Recent years have seen remarkable advancements in robotics, fueled by breakthroughs in artificial intelligence. Key players, including academic researchers, startups, and large tech companies, are utilizing large language models, speech synthesis, and advanced image recognition. Covariant CEO Peter Chen highlighted that “The algorithms can transfer, which is powering this renaissance of robotics.”
Additionally, the development of multimodal models enables robots to process diverse inputs, including text, images, and audio. Vision-language-action models allow robots to learn from their environments, giving them a form of common sense, or “embodied AI.” These capabilities allow robots to perform tasks like identifying and manipulating objects without specific training, providing engineers with the ability to monitor and adjust their actions in real time.
A novel method called “diffusion policy” has also emerged, enabling rapid training for robots. Originally used in AI image generation, this approach can generate and refine new movement patterns quickly, enhancing the adaptability of robots. Researchers are forming large behavior models akin to language models for physical applications. The concept of “fleet learning” allows for the wireless transmission of new skills between robots, further accelerating their training.
Industry leaders believe we are on the brink of a major robotics breakthrough. Nvidia CEO Jensen Huang remarked, “The ChatGPT moment for general robotics is just around the corner.” For robots to be effective, they must seamlessly integrate into human-designed environments and tools, highlighting the significance of humanoid robots. EX Robot CEO Li Boyang stated at the 2024 Summer Davos Forum that humanoid robots “are the jewel in the crown of the robotics industry” because they incorporate advanced AI, manufacturing, and new materials.
In China, there is a strong push for the development of humanoid robotics, propelled by President Xi Jinping’s vision of “new quality productive forces.” In 2023, the Ministry of Industry and Information Technology listed humanoid robots as a key innovative product. They also released “Guiding Opinions on the Innovative Development of Humanoid Robots,” which serves as a comprehensive roadmap for the industry. The inaugural Chinese Humanoid Robot Industry Conference in Beijing last year attracted more than 1,200 stakeholders, emphasizing China’s significant presence in this field, with Morgan Stanley reporting that Chinese companies hold 63 percent of the global humanoid robot supply chain.
The Beijing Humanoid Robot Innovation Center, led by UBTECH, showcases advancements such as the “Tiangong” robot. It features 3D vision, high-precision inertial sensors, and powerful CPUs to solve complex motion-control issues, capable of running up to 12 km/h and navigating various terrains. CTO Xiong Youjun aims for Tiangong to be an open-source platform for embodied intelligence.
From a military perspective, the People’s Liberation Army (PLA) identifies three main advantages of deploying humanoid robots: versatility, scalability, and survivability. Humanoid robots are closer in structure to humans and can better interact with complex environments, making them ideal for various tasks. The PLA has noted the need for a significant increase in capabilities, learning from drone swarm tactics observed in the Ukraine conflict. They believe that employing large clusters of humanoid robots could greatly enhance attack capabilities.
Moreover, humanoid robots have the potential to operate alongside human soldiers, serving as decoys to absorb enemy fire, thus minimizing human casualties. There is recognition within the PLA that the future military applications of humanoid robots could mirror the unforeseen utility of airplanes in warfare.
Despite these advantages, the PLA acknowledges several challenges, including trust in AI’s reliability, concerns over hacking, and the limitations posed by Moravec’s Paradox, which highlights the disparity in robots’ abilities in complex tasks versus intuitive actions. There are also hardware limitations where innovation and reliability in components must catch up with international standards.
For the United States to remain competitive in humanoid robotics, a rapid acceleration in research, collaboration with allies for ethical guidelines in autonomous weapons, and investment in secure communication networks and hardware developments for military applications will be crucial.
The next decade will be pivotal in determining the role and integration of humanoid robots in military contexts, as they transition from running alongside humans to potentially fighting alongside them.
