{
  "id": "unitree-g1",
  "title": "Unitree G1",
  "lang": "en",
  "source_lang": "en",
  "summary": "The Unitree G1 is a compact, electrically actuated humanoid robot introduced in 2024, with standard and developer-oriented configurations for motion demonstrations and robotics research.",
  "category": "technology",
  "subcategory": "engineering",
  "type": "product",
  "aliases": [
    "G1 humanoid robot",
    "Unitree G1 Humanoid Agent"
  ],
  "generated_at": "2026-10-15",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/unitree-g1/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/unitree-g1/",
      "json": "https://aiwiki.page/zh/unitree-g1.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "The Unitree G1 is a humanoid robot developed by Unitree Robotics, a company based in Hangzhou, China. Introduced on May 13, 2024, it combines a two-legged body with articulated arms and optional dexterous hands. The product family includes a standard G1 and a developer-oriented G1 EDU, with capabilities that depend on hardware configuration and installed software. Its development ecosystem supports locomotion research, human-motion tracking, and remotely controlled manipulation. (prnewswire.com)"
    },
    {
      "heading": "Development and release",
      "text": "Unitree announced the G1 in May 2024, describing it as a platform using reinforcement learning and simulation-based training. Early demonstrations included standing up, folding into a compact seated posture, and coordinated whole-body movements. The company announced a mass-production version on August 19, 2024, following changes to performance and exterior design. (prnewswire.com)\n\nEarly announcements described a height of approximately 1.27 metres, whereas the subsequent production specification lists 1.32 metres. These figures refer to different published descriptions of the developing product rather than a single unchanging specification. (prnewswire.com)"
    },
    {
      "heading": "Hardware and configurations",
      "text": "The production G1 has an approximately 35-kilogram mass including its battery. Its standard configuration has 23 independently actuated joints: six in each leg, five in each arm, and one at the waist. G1 EDU configurations are advertised with 23–43 joint degrees of freedom, depending on additional waist, wrist, and hand mechanisms. The optional three-finger Dex3-1 hand has seven active degrees of freedom. (unitree.com)\n\nThe robot uses permanent-magnet synchronous electric motors. Listed sensing equipment includes a depth camera and three-dimensional lidar, while onboard computing includes an eight-core processor. A removable battery provides a manufacturer-rated operating time of approximately two hours. The body folds for transport and storage. These are manufacturer specifications, with actual performance dependent on configuration and operating conditions. (unitree.com)\n\nThe standard and EDU editions differ in development access, not simply in joint count. Unitree’s store explicitly states that the standard G1 does not support secondary development and directs customization requirements to the EDU edition. Consequently, the availability of public programming examples does not imply that every retail G1 supports the same developer functions. (shop.unitree.com)"
    },
    {
      "heading": "Motion control and learning",
      "text": "Unitree publishes a G1-compatible workflow for learning motion-control policies in simulation and transferring them to physical hardware. In this context, a policy is the learned controller that selects actions from observations. The documented workflow proceeds through training, inspection of the resulting behavior, testing in another simulator, and deployment on the robot. (github.com)\n\nThe intermediate simulation test is intended to identify controllers that depend too strongly on the characteristics of one simulated environment. Unitree’s implementation supports training with Isaac Gym and subsequent testing in MuJoCo before physical deployment. This separates the process of learning a movement from the engineering work needed to reproduce it on hardware. (github.com)\n\nA separate Unitree MuJoCo package provides a simulated environment for low-level controller testing. Its interfaces include motor commands, motor-state information, and inertial measurements. The project describes its principal role as simulation-to-reality verification of controllers, rather than as a complete autonomous task-solving system. (github.com)"
    },
    {
      "heading": "Programming and teleoperation",
      "text": "Unitree’s SDK2 includes G1 examples for locomotion, arm control, audio, and dexterous hands. The examples distinguish high-level functions from lower-level control interfaces, allowing different kinds of software experiments where the robot’s configuration and development permissions support them. (github.com)\n\nThe company also publishes an extended-reality teleoperation system. It supports both 23- and 29-degree-of-freedom G1 body configurations, with options for tracked hands or handheld controllers and different robot end effectors. In this arrangement, human input supplies commands to the robot; it is therefore distinct from independently learned or autonomous behavior. (github.com)\n\nThese software resources make the platform relevant to robotics research involving learned control, human–robot motion correspondence, and manipulation. They also make it important to distinguish the physical robot from the particular controller running on it: the same hardware can be operated through substantially different software systems. (github.com)"
    },
    {
      "heading": "Research use",
      "text": "The G1 has been used as a physical evaluation platform in published research. The 2025 paper UniTracker: Learning Universal Whole-Body Motion Tracker for Humanoid Robots, for example, reports simulation and real-world experiments using a G1. Its method combines a teacher controller, a student controller based on a conditional variational autoencoder, and an adaptation stage for difficult motion sequences. (arxiv.org)\n\nSuch results concern a specified research system operating on G1 hardware. They should not be interpreted as evidence that the corresponding behavior is included in the standard commercial software. The distinction is particularly relevant where a paper uses custom training procedures or controllers rather than the manufacturer’s supplied functions. (arxiv.org)"
    },
    {
      "heading": "Limitations and safety",
      "text": "Demonstrated capabilities are not necessarily available as delivered features. Unitree states that some functions shown on its product page remain under development or testing. It also warns that the robot’s powerful mechanisms require separation from nearby people. These qualifications limit what can be inferred from promotional demonstrations alone. (unitree.com)"
    }
  ],
  "links": [
    "humanoid-robot",
    "unitree-robotics",
    "hangzhou",
    "china",
    "reinforcement-learning",
    "battery",
    "electric-motor",
    "policy-reinforcement-learning",
    "robotics"
  ],
  "sources": [
    {
      "title": "Unitree Robotics introduces G1 Humanoid agent AI avatar",
      "url": "https://www.prnewswire.com/news-releases/unitree-robotics-introduces-g1-humanoid-agent-ai-avatar-302146198.html"
    },
    {
      "title": "Unitree Robotics Biorobots at World Robotics Congress 2024",
      "url": "https://www.unitree.com/news1/"
    },
    {
      "title": "Humanoid robot G1_Humanoid Robot Functions_Humanoid Robot Price | Unitree Robotics",
      "url": "https://www.unitree.com/g1/"
    },
    {
      "title": "Unitree G1 – UnitreeRobotics",
      "url": "https://shop.unitree.com/products/unitree-g1"
    },
    {
      "title": "GitHub - unitreerobotics/unitree_rl_gym",
      "url": "https://github.com/unitreerobotics/unitree_rl_gym"
    },
    {
      "title": "GitHub - unitreerobotics/unitree_mujoco",
      "url": "https://github.com/unitreerobotics/unitree_mujoco"
    },
    {
      "title": "unitree_sdk2/example/g1/CMakeLists.txt at main",
      "url": "https://github.com/unitreerobotics/unitree_sdk2/blob/main/example/g1/CMakeLists.txt"
    },
    {
      "title": "GitHub - unitreerobotics/xr_teleoperate",
      "url": "https://github.com/unitreerobotics/xr_teleoperate"
    },
    {
      "title": "UniTracker: Learning Universal Whole-Body Motion Tracker for Humanoid Robots",
      "url": "https://arxiv.org/abs/2507.07356"
    }
  ],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}