At the NVIDIA GTC 2025 conference on March 19, Blue, a robot jointly developed by NVIDIA, Google DeepMind and Disney, made its public debut. It is equipped with NVIDIA's latest GR00T N1 robot universal base model, which walks naturally and smoothly, and has a cute and moving voice, which attracts everyone's attention as soon as it appears.
It is understood that the GR00T N1 adopts a dual-system architecture inspired by the principles of human cognition, which can be easily generalized to a variety of common tasks, and can also perform multi-step tasks that require long context and general skill sets. Nvidia founder and CEO Jensen Huang said, "The era of general-purpose robots has arrived. With NVIDIA Isaac GR00T N1 and a new data generation and robotics learning framework, together we will open new frontiers in the era of AI.”
Recently, there have been a lot of big moves in the humanoid robot industry. U.S.-based company Figure has announced the start of construction of its BotQ factory, the world's first mass-produced facility designed specifically for humanoid robots. The first generation of the production line has an annual production capacity of 12,000 units, and the company plans to increase the annual production capacity to 100,000 units in the next four years. Figure uses a vertically integrated manufacturing model, from raw material sourcing to assembly, which allows for rapid response to design changes and shorter product iteration cycles. In terms of production technology, the BotQ factory introduces software platforms such as Manufacturing Execution System (MES) and Product Lifecycle Management (PLM), and uses the self-developed AI software Helix to directly control the production line robots, reducing the dependence on traditional conveying systems, improving the level of automation, and laying the foundation for the closed-loop model of "robot manufacturing robots" in the future. In terms of products, Figure is transitioning from the prototype Figure02 to the mass-produced Figure03, which reduces the number of parts by simplifying the design, and uses injection molding, die-casting and other processes to shorten the manufacturing time of parts from weeks to seconds, greatly reducing production costs and preparing for large-scale commercialization.
Domestic companies are not idle, Midea Group for the first time disclosed a humanoid robot prototype, which can shake hands, compare hearts, and execute voice commands. This is developed by the newly established humanoid robot innovation center, and Wei Chang, vice president of Midea Group, said that the company adopts a "two-legged walk" strategy, on the one hand, to promote the short-term landing of home appliance robotization, and on the other hand, to lay out the medium and long-term technology of humanoid robots. Midea's previous acquisition of KUKA has accumulated experience in industrial robots, and intends to apply humanoid robot technology to industrial manufacturing scenarios. Changan Automobile accelerates the research and development of humanoid robots through open sourcing, and plans to release products before 2027, focusing on breakthroughs in core technologies such as drive and dynamic control, so that the robot has lower cost and higher reliability.
Pictured: Nvidia, Google's DeepMind, and Blue, a robot jointly created, were unveiled
The humanoid robot industry has recently made new progress in many aspects. In terms of policy, Hangzhou, Shanghai and other places vigorously support the artificial intelligence and robotics industry. Hangzhou should strengthen the integrated application of computing power in intelligent manufacturing, embodied intelligence and other fields, and build an ecosystem of the whole industrial chain of artificial intelligence; Chen Jining, Secretary of the Shanghai Municipal Party Committee, emphasized deepening the layout of the whole chain of the robot industry when investigating Agibot. At the technical level, Black Sesame Intelligence cooperated with Wuhan University to improve the perception and decision-making ability of the humanoid robot "Tianwen" with A2000 and C1236 chips. UBTECH and Beijing Humanoid Robot Innovation Center released the scientific research robot "Tiangong Walker", which is planned to be listed in the second quarter, and the Beijing Innovation Center also launched the "Huisi Kaiwu" general embodied intelligence platform to promote the upgrading of robots to autonomous decision-making in complex environments; Dobot Corp released the world's first "dexterous operation + straight knee walking" DobotAtom robot, which is suitable for industrial and service industry scenarios; Agibot launched the bipedal humanoid robot "Lingxi X2", equipped with an affective computing engine to enhance interaction capabilities. The field of investment and financing is also very active, UBTECH has set up a 10 million yuan humanoid robot research fund, and Shanghai State Investment Corporation has cooperated with Zhiyuan Robot to promote the large-scale application of embodied intelligent technology; Amio Robotics completed a seed round of financing of 100 million yuan, and Digital Huaxia received tens of millions of yuan in angel + round financing; Haosen Intelligence and Agibot explore the application of humanoid robots in the field of automobile manufacturing, Google DeepMind and Apptronik cooperate to develop the next generation of humanoid robots based on the Gemini2.0 model, and Lianke Technology aims at the humanoid robot skin market for material technology innovation.
At the beginning of 2025, the humanoid robot industry will usher in both policy and technology. From January to February, the humanoid robot index rose by 39%, led by core links such as dexterous hands, reducers, and sensors. Beijing, Shenzhen and other places have introduced a 100-billion-level industry cultivation plan for embodied intelligence, and the government work report for the first time included "embodied intelligence" in the focus of future industrial development. Tesla, UBTECH, Unitree and other head machine factories have shipped more than 1,000 units, robots have been upgraded from a single execution tool to an autonomous agent under the empowerment of large models, Apple and other companies have redefined the value of humanoid interaction through smart home scenarios, and the industry is undergoing an all-round transformation from hardware innovation to application landing. In terms of software, large-scale model technology promotes the development of humanoid robots to multi-modal integration and embodied intelligence, Shanghai has built the first heterogeneous humanoid robot training ground in China, and NVIDIA's Cosmos and other physical AI virtual training fields have broken through physical limitations, and the technological synergy advantages of car companies have emerged, and its "perception-decision-execution" architecture is the same as that of automotive intelligent driving technology, which can accelerate the iteration of robot technology. The integration of hardware-side technology is also accelerating, Tesla's third-generation dexterous hand adopts 17 actuators and 22 degrees of freedom design, through the "coreless motor + micro lead screw + tendon rope" scheme to achieve high flexibility, the motor front wrist optimizes joint accuracy, tactile sensors, tendon ropes, etc. have become the focus of research and development, the PC-based controller route has become the mainstream, solid-state batteries are regarded as a long-term solution to improve battery life, and fast charging technology is the focus of short- and medium-term iteration. The modular design of Tesla's dexterous hands improves the response speed, enables it to have fine operation capabilities, provides an engineering model for the industry, and drives the development of upstream supply chain enterprises. Apple explores the value of emotional interaction of humanoid robots through smart home scenarios, and research shows that home robots can transmit intentions and emotions through movement postures, and this non-voice interaction mode provides a new direction for embodied intelligence to land in home scenarios.
