Graphics processing powerhouse Nvidia is set to further its foray into the robotics sector with the anticipated launch of its Jetson Thor compact computing platform in the first half of 2025. Specifically designed for humanoid robots, this move, initially reported by the Financial Times on January 1st, underscores Nvidia’s growing commitment to what many industry experts predict will be a multi-billion dollar market.
Nvidia’s announcement comes amidst a flurry of activity from tech giants exploring the potential of humanoid robots. Notably, Google recently partnered with robotics firm Apptronik to integrate its Gemini 2.0 multimodal model’s spatial reasoning capabilities into robots designed to operate in complex environments.
The diverse strategies adopted by these major players reflect the immense potential of the robotics industry. A McKinsey report cited in the original article projects the sector’s revenue to surge to between $190 billion and $910 billion by 2040, boasting an impressive average annual growth rate of 13% to 23%.
Nvidia has positioned itself as a comprehensive technology platform provider for the “physical AI” domain. The company offers an end-to-end solution encompassing both the software layers required for robot training and the crucial robotic chips.
A key aspect of Nvidia’s approach is its robust software ecosystem, including NVIDIA Isaac, NVIDIA Metropolis, Nvidia Isaac Sim, and Nvidia OSMO. These tools provide comprehensive support for the development, training, simulation, deployment, operation, and optimization of AI-powered robots. On the hardware front, Nvidia’s Jetson Orin series chips have been widely adopted by domestic and international humanoid robot manufacturers as dedicated edge AI computing platforms, offering up to 275 trillion operations per second (TOPS) of processing power. The upcoming Jetson Thor aims to further enhance this capability for humanoid applications.
In the realm of robot training, Nvidia provides NVIDIA NeMo on its DGX platform for training and fine-tuning robot models. The company has also launched “GR00T,” a foundational model project for general-purpose humanoid robots. Furthermore, Nvidia established the GEAR lab last February, dedicated to advancing general-purpose embodied intelligence capable of generalizing various skills and adapting to real-world scenarios.
Nvidia’s efforts extend to robot simulation with tools like Omniverse, a 3D simulation engine for generating vast amounts of synthetic data, and Isaac Sim, a simulation and testing platform designed to realistically replicate diverse tasks and create high-fidelity 3D environments, accurately mimicking real-world applications for embodied intelligence. The company also continues to innovate at the software algorithm level, with recent releases including the text-to-3D generator LATTE3D and the physics-based unified character controller MaskedMimic in 2024.
China’s Domestic Market Gears Up to Challenge Nvidia
While Nvidia establishes a strong foothold, the Chinese domestic market is also witnessing significant developments in humanoid robotics. The integration of software ecosystems with hardware is identified as a critical challenge in enhancing the embodied intelligence of robots.
On the hardware side, China boasts a relatively complete and mature supply chain for core components such as servo motors, reducers, controllers, sensors, and batteries, which are fundamental for humanoid robot movement. However, the software infrastructure in China, particularly around the core modules of perception, decision-making and planning, and action generation, is considered weaker, lacking a unified operating system and standardized software development toolchains.
Fueled by breakthroughs in AI large models and robot simulation training, the increasing intelligence of humanoid robots is driving changes across the industrial chain, necessitating the development of new algorithms, hardware, and software ecosystems. This presents new opportunities for China’s humanoid robot industry.
Currently, many emerging Chinese humanoid robot manufacturers are adopting layered end-to-end solutions, utilizing a collaborative “brain large model” and “cerebellum small model” approach to enhance robot intelligence and improve their ability to autonomously complete complex tasks. The design, manufacturing, and integration of the robot body remain central to the industrial chain. Domestic manufacturers are typically involved in robot body R&D, control systems, core control algorithms, and the development of key components.
The embodied intelligence trend has spurred the emergence of numerous humanoid robot companies in China, with over 50 such startups appearing between 2023 and August 2024, according to NetEase data. This includes established players like UBTECH, recently founded ventures like ZHIYUAN Robotics, and entries from internet and automotive giants such as Xiaomi and XPENG Motors.
Key Chinese Players in the Humanoid Robot Race:
- Xiaomi: Established its robot company in April 2024, building upon its earlier CyberOne full-size humanoid robot (launched in 2022) and the CyberDog quadruped robot. Xiaomi’s focus is on applications in 3C electronics and automotive manufacturing. The CyberOne, standing at 1.77 meters and weighing 52kg, features 21 degrees of freedom with millisecond-level real-time response, emotion sensing capabilities, advanced visual spatial systems, and self-developed whole-body control algorithms.
- Unitree Robotics: Initially focused on quadruped robots, Unitree launched its Unitree G1 humanoid robot in May 2024, garnering attention for its competitive starting price. The G1 boasts 23 to 45 joint motors for high flexibility and can perform complex actions. It features AI technology for self-learning and upgrades and utilizes Unitree’s unified large model, UnifoLM, to continuously optimize behavior and movement strategies. Unitree independently develops key components like motors, reducers, controllers, and lidar.
- AgiBot: Founded in recent years, AgiBot focuses on deep integration of AI and robotics, developing foundational “brain” and “cerebellum” models for general-purpose robots. The company has launched the “Expedition” and “Lingxi” series of humanoid robots and recently unveiled five commercial humanoid robots in August. AgiBot is building a large-scale robot data collection factory in Shanghai and has released AgiBot World, an open-source dataset of real-world robot skills. The company categorizes robot systems into power, perception, communication, and control domains, developing key technologies in each area, including the PowerFlow joint module and the open-source AmRT communication framework.
- UBTECH: A long-standing player in AI and humanoid robotics, UBTECH’s Walker series robots feature high-performance servo joints and comprehensive sensing systems. The company recently showcased its Walker S robot working in a NIO automotive factory. Walker S, standing at 1.7 meters with 41 high-performance servo joints, possesses multi-dimensional force feedback, multi-eye stereo vision, omnidirectional hearing, and inertial measurement units for comprehensive environmental perception. UBTECH also holds a significant number of key technology patents.
- Fourier Intelligence: A general-purpose robotics company with roots in intelligent rehabilitation solutions, Fourier Intelligence launched its GR-1 humanoid robot in July 2023, followed by the GR-2 in September 2024. The GR-2, with 53 degrees of freedom and a 3kg single-arm payload capacity, features an optimized bionic design and improved physical specifications. Fourier also offers a full range of dedicated FSA actuators for humanoid robots.
Beyond robot manufacturers, China’s robotics ecosystem includes multimodal embodied large model developers like Baidu, ByteDance, and Alibaba, as well as cloud service providers such as Baidu AI Cloud, Huawei Cloud, Alibaba Cloud, and SenseTime, which offer algorithm and software support.
This dynamic landscape highlights the intense global competition in the burgeoning field of humanoid robotics, with Nvidia’s strategic hardware and software offerings facing a rapidly evolving and ambitious domestic market in China.
Author: Lindsay Tse | Editor: Chain Zhang
The post first appeared on .
