Daily AI & Robotics Wrap: Humanoids Take Center Stage with Major AI and Global Initiatives
Nvidia Unveils Comprehensive “Physical AI” Stack, Driving Humanoid Advancement
Nvidia has made a significant announcement at the Conference on Robot Learning (CoRL) 2025 in Seoul, staking its claim on what it terms the next trillion-dollar market: physical AI. The tech giant unveiled a new suite of models and hardware designed to empower robots to perceive, reason, and act within the physical world. Nvidia asserts that humanoid robots represent the “superset” of all robotic forms, believing that breakthroughs in this area will naturally extend to industrial arms, mobile bots, and logistics systems.
Central to this initiative are three foundation models: Newton, GR00T (v1.6), and Cosmos. Newton, co-developed with Disney Research and DeepMind, focuses on manipulation, teaching robots how to grasp and handle objects. Early adopters of Newton include Zurich’s Robotic Systems Lab, the Technical University of Munich, and Peking University.
GR00T 1.6 is specifically tailored for humanoid robots, designed to coordinate arms, joints, and mobility for fluid movement and action. The third model, Cosmos, is an open-source reasoning model described as a “brain” for robots, providing vision-language-action capabilities that enable machines to generalize to unfamiliar environments, similar to how large language models function with text.
These models operate on Jetson Thor, Nvidia’s new robotics compute platform, which is built on Blackwell chips and provides the necessary power for real-time foundation model execution. Leading robot manufacturers such as Figure AI, Galbot, Google DeepMind, Mentee Robotics, Meta, Skild AI, and Unitree are reportedly evaluating Isaac GR00T N models for developing general-purpose robots.
- Nvidia’s new “physical AI” stack aims to enable robots to perceive, reason, and act in the physical world.
- Key models include Newton for manipulation, GR00T 1.6 for humanoid coordination, and Cosmos for open-source reasoning.
- The Jetson Thor platform, powered by Blackwell chips, supports real-time AI processing for these models.
Google DeepMind Introduces New AI Models to Enhance Robotic Reasoning and Vision
Google DeepMind has unveiled two new artificial intelligence models, Gemini Robotics-ER 1.5 and Gemini Robotics 1.5, which are designed to significantly enhance the capabilities of general-purpose robots. These models work in tandem to improve a robot’s reasoning, vision, and ability to execute actions in real-world settings.
Gemini Robotics-ER 1.5 is a planning and reasoning model that can generate multi-step plans for complex tasks and demonstrates strong performance in spatial understanding. It can also access external resources, such as Google Search, to gather information and inform its decision-making in physical environments.
Following the creation of a plan by ER 1.5, Gemini Robotics 1.5, a vision-language-action (VLA) model, takes over. This model translates instructions and visual data into motor commands, allowing the robot to execute tasks efficiently. It can also provide natural language explanations for its decisions, enabling robots to handle intricate, multi-step commands smoothly.
DeepMind highlights that these AI models are adaptable to various robot forms and sizes due to their flexible design and spatial awareness, marking a step towards integrating generative AI into robotics. The ER 1.5 planner is currently available to developers via the Gemini API in Google AI Studio, while the VLA model is restricted to select partners.
Japanese Government Sets Ambitious Goal for Multipurpose Humanoid Robots by 2030
The Japanese Cabinet Office plans to establish a new objective within its Moonshot Research and Development Program: the creation of multipurpose humanoid robots by 2030. This initiative aims to develop robots capable of operating in diverse environments, including households, manufacturing plants, and disaster-stricken areas.
The primary goal is to produce a prototype by 2030 that can handle tasks with human assistance, thereby laying a foundational framework for the technology. This prototype is expected to stimulate private-sector investment and encourage further AI research.
The Moonshot program, initiated in 2020 with approximately ¥400 billion in funding, supports projects designed to bring about positive societal changes. While the program has previously backed non-humanoid robot development, this new goal specifically targets humanoids. The government envisions robots that possess human-level or superior physical capabilities and can make autonomous decisions through learning.
The Cabinet Office will publicly solicit leaders for research teams by the end of the current fiscal year to pursue this ambitious goal.
ABS and Persona AI Partner to Integrate Humanoid Robotics in Shipyards
In a move to enhance safety and productivity in shipyards, ABS (American Bureau of Shipping) and Persona AI have formalized a collaboration to develop inspection technologies for Persona AI’s humanoid robot platform. This partnership aims to adapt Persona AI’s humanoid robots, which incorporate NASA’s robotic hand technology, for a variety of shipyard tasks.
Unlike conventional industrial robots, humanoid robots are particularly well-suited for shipyard environments, which are typically designed for human workers. Their flexibility and mobility allow them to operate effectively in complex, confined, or ergonomically challenging spaces.
Nic Radford, CEO and Co-Founder of Persona AI, emphasized that this partnership with ABS signifies that humanoid robotics are transitioning from a theoretical concept to a certified reality, poised to revolutionize how the shipbuilding industry innovates and competes. The project will also leverage ABS’s substantial investments in AI evaluation of data and data quality.
China’s Robotics Boom: Affordable Humanoids and Manufacturing Dominance
China is rapidly asserting its dominance in the global robotics landscape, with a new report indicating that the country’s robotic workforce now exceeds that of the rest of the world combined. This surge is fueled by significant government support and a booming startup ecosystem focused on advanced robotics, including humanoid machines.
The International Federation of Robotics (IFR) report highlights that Chinese factories installed nearly 300,000 new industrial robots last year alone, significantly outpacing other nations. This strategic push, outlined in the “Made in China 2025” blueprint, prioritizes robotics, AI, and semiconductors to transform China into a leading technological powerhouse, addressing rising labor costs and an aging population.
A key aspect of this development is the emergence of affordable humanoid robots. Companies like Unitree Robotics are already selling basic humanoid models in China for approximately $6,000, a fraction of the cost of similar robots from Western pioneers like Boston Dynamics. This accessibility is accelerating the integration of human-like machines into various sectors.
KEENON Robotics and Aibotics Eye Collaboration on Humanoid XMAN Robots
Aibotics, Inc., a developer and manager of AI- and robotics-enhanced technologies, is in discussions with KEENON Robotics, a global leader in service robotics, regarding a potential collaboration on humanoid XMAN robots. This partnership is part of Aibotics’ accelerated international expansion, particularly across the Middle East and the CARICOM region.
While the immediate focus of their broader partnership involves the distribution of KEENON’s service robots (such as DinerBot, ButlerBot, and KleenBot) for hospitality, retail, and healthcare sectors, Aibotics has signaled its intent to collaborate on humanoid XMAN robots.
The global humanoid robotics sector is projected to surpass $38 billion by 2035, indicating a significant growth opportunity that this potential collaboration aims to tap into.