Humanoid
Assembly Line
NVIDIA’s Humanoid Robot Partners Unveiled at CES 2025, Featuring China’s Unitree, XPeng, and BYD
As the humanoid robotics field has become one of the hottest topics lately, NVIDIA introduced its Cosmos family of foundational AI models, designed to train humanoid and industrial robots, at CES 2025.
According to mydrivers, Unitree Robotics recently launched its first general-purpose humanoid robot, H1, which is the first full-size humanoid robot capable of running in China.
Samsung Electronics To Become Largest Shareholder in Rainbow Robotics Accelerating Future Robot Development
Samsung Electronics today announced that it will become the largest shareholder in Rainbow Robotics to accelerate future robot development such as humanoid robots. Samsung first acquired a 14.7% stake in the Korean firm in 2023 with investment of KRW 86.8 billion, and exercises a call option to increase its stake to 35%. Rainbow Robotics will also be incorporated as a subsidiary under Samsung Electronics’ consolidated financial statements.
Through collaboration with Rainbow Robotics, Samsung will further strengthen its foundation in the development of advanced robot technology. Rainbow Robotics was founded in 2011 by researchers from the Korea Advanced Institute of Science & Technology’s (KAIST) Humanoid Robot Research Center who developed the first two-legged walking robot “Hubo” in Korea.
By combining Samsung Electronics’ AI and software technology with Rainbow Robotics’ robotics technology, the collaboration plans to accelerate the development of intelligent advanced humanoids.
🤖 EngineAI releases PM01 humanoid robot for commercial, educational use
EngineAI Robotics unveiled the PM01, a “lightweight, highly dynamic, and fully open universal embodied intelligent robot.” The PM01 is 1.38 m (4.5 ft.) tall, weighs about 40 kg (88 lb.), and features 24 degrees of freedom (DoF), achieving a movement speed of 2 m/sec. (4.4 mph). Its 320-degree waist rotation enables complex motions, complemented by mechanical and natural humanoid gait walking modes. The robot’s computing suite is based on X86 architecture and NVIDIA Jetson Orin high-performance modules, and it supports cross-platform algorithm deployment and application development.
Nvidia eyes Taiwan for humanoid robot assembly; supply chain gaps remain
Taiwanese manufacturers have advanced significantly in component production, though complete system development remains underdeveloped. Hiwin and Main Drive Corp. stand out as two of the few Taiwanese firms capable of mass-producing harmonic reducers. Hiwin has supplied components for Boston Dynamics’ BigDog and is partnering with US firms to develop logistics robots. Meanwhile, Main Drive is transitioning from reducers to joint modules, targeting the collaborative robot (cobot) market with subsystem innovations. These advancements are crucial in bolstering Taiwan’s robotics ecosystem.
Analysts cited by Money.udn.com suggest that Nvidia’s potential Taiwanese supply chain for robotics could involve Main Drive (a subsidiary of Mirle Automation) and Hiwin for harmonic reducers, Apex Dynamics and Turvo International for planetary reducers, and Chieftek Precision and Hiwin Mikrosystem for linear motors.
Reportedly, key targets also include Hota Industrial Mfg., Ace Pillar, and Solomon Technology, focusing on humanoid and specialized robots. Insiders predict that plans could materialize by 2025, followed by certifications, paving the way for a Taiwan-US robotics supply chain.
🤖 Pudu Robotics Unveils PUDU D9: A Full-Sized Humanoid Robot Driving Commercially Viable Embodied Intelligence
Pudu Robotics has launched its first Full-sized Bipedal Humanoid Robot, the PUDU D9. The PUDU D9 features 42 degrees of freedom across its joints, with a maximum joint torque of 352 Nm. It is equipped with the previously released PUDU DH11 dexterous hand, enabling human-like bipedal walking and dual-hand operational capabilities. The PUDU D9 is designed with a human-centric philosophy, embodying the principle of “Born to Serve.” As a fully anthropomorphic robot, its design closely mirrors human capabilities while striving to provide practical assistance across a diverse range of applications.
🤖 China's BYD, Nio do more with AI-powered humanoid factory robots
China’s BYD, Nio do more with AI-powered humanoid factory robots. Electric vehicle makers BYD and Nio are among the latest manufacturers in China to deploy robots that can walk, talk and carry objects on factory floors, as the nation faces a shortage of production workers. BYD introduced the humanoid robot Walker S1 from Hong Kong-listed Chinese startup UBTech Robotics at its factory in Hunan province this year.
ExBody2: Advanced Expressive Humanoid Whole-Body Control
This paper enables real-world humanoid robots to maintain stability while performing expressive motions like humans do. We propose ExBody2, a generalized whole-body tracking framework that can take any reference motion inputs and control the humanoid to mimic the motion. The model is trained in simulation with Reinforcement Learning and then transferred to the real world. It decouples keypoint tracking with velocity control, and effectively leverages a privileged teacher policy to distill precise mimic skills into the target student policy, which enables high-fidelity replication of dynamic movements such as running, crouching, dancing, and other challenging motions. We present a comprehensive qualitative and quantitative analysis of crucial design factors in the paper. We conduct our experiments on two humanoid platforms and demonstrate the superiority of our approach against state-of-the-arts, providing practical guidelines to pursue the extreme of whole-body control for humanoid robots.
Pioneering the Future of Production 🦾
Fourier Trains Humanoid Robots for Real-World Roles Using NVIDIA Isaac Gym
Fourier, a Shanghai-based robotics company, is doing the heavy lifting by developing advanced humanoid robots that can be integrated into real-world applications where precision and agility are critical. The company announced the expansion of its GRx humanoid robot series with the launch of GR-2 in late September. Building on the previous-generation GR-1, the world’s first mass-produced humanoid robot, GR-2 features an upgraded hardware design, greater adaptability, advanced dexterity and a humanlike range of motion.
To develop and test GR-2, the Fourier team turned to NVIDIA Isaac Gym (now deprecated) for reinforcement learning. They are currently porting their workflows to the recently launched NVIDIA Isaac Lab, an open-source modular framework for robot learning designed to simplify how robots adapt to new skills.
While training GR-2 for the floor-to-stand maneuver, Fourier simulated the physical demands required for completing tasks at different levels of elevation. By replicating the GR-2 model, they tested how it performs under various settings and completed 3,000 iterations in around 15 hours, a notable reduction compared to traditional training methods. When transferred directly to GR-2’s physical controls, the model’s action tensors achieved an 89% success rate.
Crossing the Sim2Real Gap With NVIDIA Isaac Lab
We’ve been able to demonstrate this in areas like step-recovery, where physics are particularly hard to model. In situations where Digit loses its footing, it’s often a result of an environment where we don’t have a good model of what’s going on – there might be something pushing on or caught on Digit, or its feet might be slipping on the ground in an unexpected way. Digit might not even be able to tell which issue it’s having. But we can train a controller to be robust to many of these disturbances with reinforcement-learning, training it on many possible ways that the robot might fall until it comes up with a controller that works well in many situations. In the following chart, you can see how big of a difference that training can make.
Early last year, we started using NVIDIA Isaac Lab to train these types of models. Working with NVIDIA, we were able to make some basic policies that allowed Digit to walk around.
The net result has been a huge step forward in our RL software stack. Instead of a pile of stacked reward functions over everything from “stop wiggling your foot” to “stand up straighter”, we have a handful of rewards around things like energy consumption and symmetry that are not only simpler, but follow our basic intuitions about how Digit should move. Investing the time to understand why the simulation differed has taught us a lot more about why we want Digit to move a certain way in the first place. And most importantly, coupled with fast NVIDIA Isaac Sim, a reference application built on NVIDIA Omniverse for simulating an testing AI-driven robots, it’s enabled us to explore the impact of different physical characteristics that we might want in future generations of Digit.
Atlas Goes Hands On
HOVER: Versatile Neural Whole-Body Controller for Humanoid Robots
Humanoid whole-body control requires adapting to diverse tasks such as navigation, loco-manipulation, and tabletop manipulation, each demanding a different mode of control. For example, navigation relies on root velocity tracking, while tabletop manipulation prioritizes upper-body joint angle tracking. Existing approaches typically train individual policies tailored to a specific command space, limiting their transferability across modes. We present the key insight that full-body kinematic motion imitation can serve as a common abstraction for all these tasks and provide general-purpose motor skills for learning multiple modes of whole-body control. Building on this, we propose HOVER (Humanoid Versatile Controller), a multi-mode policy distillation framework that consolidates diverse control modes into a unified policy. HOVER enables seamless transitions between control modes while preserving the distinct advantages of each, offering a robust and scalable solution for humanoid control across a wide range of modes. By eliminating the need for policy retraining for each control mode, our approach improves efficiency and flexibility for future humanoid applications.
Inside the world's first humanoid factory, where robots could eventually build themselves
1X’s Generative World Model for Robot Interactions
World models solve a very practical and yet often overlooked challenge when building general-purpose robots: evaluation. If you train a robot to perform 1000 unique tasks, it is very hard to know whether a new model has made the robot better at all 1000 tasks, compared to a prior model. Even the same model weights can experience a rapid degradation in performance in a matter of days due to subtle changes in the environment background or ambient lighting.
Physics-based simulation (Bullet, Mujoco, Isaac Sim, Drake) are a reasonable way to quickly test robot policies. They are resettable and reproducible, allowing researchers to carefully compare different control algorithms. However, these simulators are mostly designed for rigid body dynamics and require a lot of manual asset authoring.
We’re taking a radically new approach to evaluation of general-purpose robots: learning a simulator directly from raw sensor data and using it to evaluate our policies across millions of scenarios. By learning a simulator directly from real data, you can absorb the full complexity of the real world without manual asset creation.
To help accelerate progress towards solving world models for robotics, we are releasing over 100 hours of vector-quantized video (Apache 2.0), pretrained baseline models, and the 1X World Model Challenge, a three-stage challenge with cash prizes.
Why Nvidia, Tesla, Amazon And More Are Betting Big On AI-Powered Humanoid Robots
Figure Status Update - BMW Full Use Case
Everything changes when Bots make Bots
Accenture Invests in Sanctuary AI to Bring AI-Powered, Humanoid Robotics to Work Alongside Humans
Accenture has made a strategic investment, through Accenture Ventures, in Sanctuary AI, a developer of humanoid general-purpose robots that are powered by AI and can perform a wide variety of work tasks quickly, safely and effectively.
Sanctuary AI’s general-purpose robot PhoenixTM, recently recognized as one of TIME magazine’s “Best Inventions of 2023,” can perform a multitude of work tasks. For instance, at a Mark’s retail store in Langley, BC, Canada, Phoenix has performed more than 100 tasks, including choosing and packing merchandise, and correctly cleaning, tagging, labeling and folding items, with robotic hands that rival human hand dexterity and fine manipulation. Phoenix is powered by the company’s AI control system, CarbonTM, which mimics subsystems found in the human brain, such as memory, sight, sound and touch, and translates natural language into action in the real world.
🤖 Apptronik Collaborates with NVIDIA to Advance AI for General-Purpose Humanoid Robots
Apptronik, a leader in next-generation general-purpose humanoid robots designed to change the way we live and work, today announced its collaboration with NVIDIA to enhance humanoid robot learning. Apptronik’s Apollo humanoid robot will integrate with NVIDIA’s new general-purpose foundation model for robot learning, Project GR00T, to make possible a humanoid robot-augmented future that will better serve humanity.
Apollo’s main computing system, including onboard NVIDIA Jetson AGX Orin and Jetson Orin NX modules, enables the AI-powered robot to efficiently leverage cutting-edge models such as GR00T to perform a wide variety of tasks, while its humanoid form allows it to learn from human demonstrations. This results in faster skill development, advancing the field of general-purpose robots as Apptronik continues to develop its next-generation robotics computing platform. Apptronik will leverage NVIDIA’s portfolio of AI offerings for humanoid robots, including the new GR00T foundation models, Isaac Lab, and the OSMO compute orchestration service, announced at GTC.
NVIDIA Announces Project GR00T Foundation Model for Humanoid Robots and Major Isaac Robotics Platform Update
NVIDIA announced Project GR00T, a general-purpose foundation model for humanoid robots, designed to further its work driving breakthroughs in robotics and embodied AI.
As part of the initiative, the company also unveiled a new computer, Jetson Thor, for humanoid robots based on the NVIDIA Thor system-on-a-chip (SoC), as well as significant upgrades to the NVIDIA Isaac™ robotics platform, including generative AI foundation models and tools for simulation and AI workflow infrastructure.
The SoC includes a next-generation GPU based on the NVIDIA Blackwell architecture with a transformer engine delivering 800 teraflops of 8-bit floating point AI performance to run multimodal generative AI models like GR00T. With an integrated functional safety processor, a high-performance CPU cluster and 100GB of ethernet bandwidth, it significantly simplifies design and integration efforts.
Robots powered by GR00T, which stands for Generalist Robot 00 Technology, will be designed to understand natural language and emulate movements by observing human actions — quickly learning coordination, dexterity and other skills in order to navigate, adapt and interact with the real world. In his GTC keynote, Huang demonstrated several such robots completing a variety of tasks.
Agility Robotics Brings Operational Visibility to Deployment of Digit Fleets with the Launch of Agility Arc™
In its first iteration, Agility Arc will provide customers with operational visibility into critical KPIs like uptime, throughput, Mean Time Between Incidents (MTBI), and robot status, allowing customers to understand what’s happening in the workcell and how Digit is performing. Additionally, Agility Arc will provide industry standard APIs to simplify integration with existing Warehouse Management Systems (WMS), Warehouse Execution Systems (WES), and Manufacturing Execution Systems (MES) among others.
The global market for humanoid robots could reach $38 billion by 2035
The total addressable market for humanoid robots is projected to reach $38 billion by 2035, up more than sixfold from a previous projection of $6 billion, Goldman Sachs Research analyst Jacqueline Du, head of China Industrial Technology research, writes in the report. Their estimate for robot shipments increased fourfold, to 1.4 million units, over the same time frame, with a much faster path to profitability on a 40% reduction in the cost of materials.
There are signs that robot components, from high-precision gears to actuators, could also cost less than previously expected, leading to faster commercialization. The manufacturing cost of humanoid robots has dropped — from a range that ran between an estimated $50,000 (for lower-end models) and $250,000 (for state-of-the art versions) per unit last year, to a range of between $30,000 and $150,000 now. Where our analysts had expected a decline of 15-20% per annum, the cost declined 40%.