CtrlK
The Galaxea R1 is a series of wheeled dual-arm [[humanoid robot]]s developed by Galaxea AI (also known as Galaxea Dynamics), a Chinese [[embodied AI]] and [[robotics]] startup founded in 2023. The R1 series combines an upper-body humanoid design with a wheeled mobile base, giving it the manipulation dexterity of a humanoid and the practical mobility of a wheeled platform. The lineup includes three variants: the standard R1, the upgraded R1 Pro, and the compact R1 Lite. Each model is built around the [[NVIDIA]] Jetson AGX Orin computing platform and is designed for industrial manipulation, research, and service applications.
Since its initial release in 2024, the R1 has become one of Galaxea AI's flagship products. The company has shipped units to more than 40 clients worldwide, including [[Huawei]] Cloud, [[Volkswagen]], Haier, [[Samsung]], [[ByteDance]], Physical Intelligence, [[Stanford University]], and MIT.
Galaxea AI was founded in September 2023 by a team of scientists and engineers with roots at [[Tsinghua University]], Stanford, and other leading institutions. The company is headquartered in Beijing, China, with operations in Suzhou, Jiangsu province.
The company was co-founded by four individuals:
| Name | Role | Background |
|---|---|---|
| Gao Jiyang (高继扬) | CEO and Founder | Tsinghua University undergraduate; Ph.D. in computer vision from USC (completed in three years); former engineer at [[Waymo]] and Momenta; intern at Google and [[SenseTime]] |
| Xu Huazhe | Co-Founder and Co-Chief Science Officer | Bachelor's from Tsinghua University; Ph.D. from UC Berkeley (BAIR Lab under Trevor Darrell); postdoctoral researcher at Stanford Vision and Learning Lab; currently Tenure-Track Assistant Professor at Tsinghua IIIS, leading the Tsinghua Embodied AI Lab (TEA Lab); winner of best system paper at CoRL 2023 |
| Hang Zhao | Co-Founder | Ph.D. from MIT; former colleague of Gao Jiyang at Waymo |
| Tianwei Li | Co-Founder | Master's from UCL; former Senior Director at Momenta |
Gao Jiyang, born in 1992, was admitted to Tsinghua through a physics competition during high school. After completing his doctorate at USC under Professor Ram Nevatia, he joined Waymo's perception team, where he co-authored the VectorNet paper, a neural network-based prediction and planning algorithm that was widely adopted across the [[autonomous driving]] industry. He later moved to Momenta in China, where he improved parking success rates from 60% to 95% and delivered Navigation on Autopilot (NOA) systems to SAIC Motor before departing to found Galaxea AI.
Galaxea AI has raised substantial capital in a short period, reflecting strong investor confidence in the Chinese humanoid robotics sector.
| Round | Date | Amount | Valuation | Key Investors |
|---|---|---|---|---|
| Early rounds (A-series) | 2024 | Not publicly disclosed | Not disclosed | Baidu Ventures, GSR Ventures, IDG Capital |
| A4 and A5 rounds | 2025 | Combined over $100 million | $700 million | Capital Today, Meituan Long-Z Investments, Ant Group, IDG Capital, GL Ventures, FunPlus |
| Series B | February 2026 | ~$144 million (CNY 1 billion) | ~$1.4 billion (CNY 10 billion) | Multiple institutional investors |
| Series B+ | April 2026 | ~$291 million (CNY 2 billion) | ~$2.9 billion (CNY 20 billion) | Nearly 20 institutional investors including Lens Technology (hardware and mass production partner), state-backed capital, industrial investors |
By April 2026, Galaxea AI's cumulative fundraising exceeded $500 million, placing it among the most well-funded humanoid robotics startups in China alongside competitors such as Galbot and [[Unitree Robotics]].
Galaxea AI's stated mission is to build "embodied intelligence at a global scale: 10 billion robots for 10 billion people." The company's long-term development roadmap proceeds in three stages:
CEO Gao Jiyang has stated that humanoid robots could enter homes for cooking and cleaning tasks within less than a decade. The company plans to release a two-legged bipedal humanoid robot as part of its 2026 product roadmap.
The Galaxea R1 series consists of three models, each targeting different use cases while sharing a common design philosophy: a human-like upper body with dual arms mounted on a wheeled chassis. This approach prioritizes practical deployability over the bipedal locomotion used by robots like [[Tesla Optimus]] or [[Boston Dynamics]]' [[Atlas (robot)|Atlas]], since wheels offer more reliable mobility in structured indoor environments such as factories, warehouses, and laboratories.
All R1 variants share several core features:
| Specification | R1 | R1 Pro | R1 Lite |
|---|---|---|---|
| Total Degrees of Freedom | 24 | 26 | 23 |
| Arm DOF (per arm) | 6 | 7 | 6 |
| Arm Model | Galaxea A1 | Galaxea A2 | Galaxea A1 |
| Arm Working Radius | 70 cm (86 cm with gripper) | 62 cm (arm only) | 70 cm |
| Rated Single-Arm Payload | 3.5 kg at 0.6 m | 3.5 kg at 0.6 m | N/A |
| Maximum Single-Arm Payload | 5 kg at 0.5 m | 5 kg at 0.6 m | 5 kg |
| Maximum Dual-Arm Payload | 10 kg | 10 kg | N/A |
| Operating Height Range | 0 to 200 cm | 0 to 200 cm | Up to 170 cm |
| Standing Height | 170 cm | 170 cm | 128 cm |
| Chassis Width | 675 mm | 675 mm | N/A |
| Weight (with battery) | 96 kg | 96 kg | 96 kg |
| Maximum Speed | 5.4 km/h | 5.4 km/h | 5.4 km/h |
| Battery | 48 V / 35 Ah Li-ion (~1,680 Wh) | 48 V / 35 Ah Li-ion (~1,680 Wh) | 48 V Li-ion (~1,680 Wh) |
| Operating Time | ~2 hours | ~2 hours | N/A |
| Max End-Effector Velocity | 7.5 m/s | 7.5 m/s | N/A |
| Repeat Positioning Accuracy | +/-0.5 mm | +/-0.5 mm | N/A |
| Deep Learning Compute | 550 TOPS | 200 TOPS | N/A |
| Head Cameras | 1 stereo | 2 monocular (1920x1080 at 30fps) | 2 (central platform) |
| Wrist Cameras | 2 | 2 (optional depth, 1280x720 at 30fps) | N/A |
| Chassis Cameras | 5 | 5 monocular (1920x1080 at 30fps) | N/A |
| LiDAR | 1x 360-degree (optional 2nd) | 1 to 2x 360-degree (0.1 m min range) | 360-degree LiDAR |
| IMU | Yes | Yes | N/A |
| Computing Platform | NVIDIA Jetson AGX Orin 32GB | NVIDIA Jetson AGX Orin 32GB | Intel Core i9-12900HK |
| Connectivity | Ethernet, USB, Wi-Fi | 4x Gigabit Ethernet, 8x GMSL, USB, Wi-Fi | Ethernet, USB, Wi-Fi |
| Teleoperation | Isomorphic + VR | Isomorphic + VR | Isomorphic |
| Force Sensors | No | Integrated | No |
| Speech Interaction | Yes | Yes | N/A |
| Approximate Price | ~$28,000 USD (199,000 RMB) | ~$44,500 to $64,000 USD | ~$40,000 USD |
The standard R1 is the original model in the series. It features a 24-DOF configuration with dual 6-DOF Galaxea A1 robot arms, a 4-DOF torso, and a 6-DOF three-wheel steering vector chassis. The A1 arms are lightweight, force-controlled units connected by dual rods, with each joint equipped with a planetary gear motor that enables independent variable-speed operation with high precision and torque. The arms are paired with Galaxea G1 parallel grippers.
The R1 stands 170 cm tall and weighs 96 kg with its battery. It can operate at heights ranging from ground level up to 200 cm, giving it a versatile workspace for tasks like bin picking, shelf manipulation, and object transport. The robot's binocular stereo head camera provides depth perception, while five chassis-mounted cameras and a 360-degree LiDAR enable autonomous navigation and obstacle avoidance.
With a starting price around 199,000 RMB (approximately $28,000 USD), the standard R1 is the most affordable entry point in the series.
The R1 Pro is the most capable model in the lineup. Its primary upgrade over the standard R1 is the use of dual 7-DOF Galaxea A2 robot arms, which provide an additional degree of freedom per arm for more dexterous manipulation. The 7-DOF configuration enables the arms to reach around obstacles and adopt more flexible postures during complex tasks. The R1 Pro also features integrated force sensors, allowing force-controlled interaction with the environment, which is critical for tasks involving delicate objects or human collaboration.
The R1 Pro's torso provides a yaw range of plus or minus 175 degrees and a pitch range from negative 105 degrees to 90 degrees, giving the robot considerable upper-body flexibility. Its 1,680 Wh lithium-ion battery provides approximately two hours of operation. The cooling system uses low-noise local air cooling with two air ducts operating at 82 cubic meters per minute.
The computing unit features 32 GB LPDDR5 memory and 1 TB SSD storage, with eight GMSL camera interfaces and four Gigabit Ethernet ports for sensor connectivity. The R1 Pro supports both isomorphic and VR teleoperation, with optional depth cameras at the wrists for close-range manipulation tasks.
Priced between approximately $44,500 and $64,000 USD, the R1 Pro targets industrial collaboration, manufacturing, and complex manipulation scenarios.
The R1 Lite is the most compact variant, designed primarily as a data collection platform for training [[machine learning]] and embodied AI models. It stands 128 cm tall, significantly shorter than the R1 and R1 Pro, with a maximum operating height of 170 cm. Despite its smaller stature, it retains dual 6-DOF arms with a 70 cm reach and a 5 kg per arm payload capacity.
A notable difference is the R1 Lite's computing platform: instead of the NVIDIA Jetson AGX Orin used in the other models, the R1 Lite uses an Intel Core i9-12900HK processor with 1 TB SSD storage. It supports isomorphic teleoperation for remote data collection but does not include VR control capability.
The R1 Lite is priced at approximately $40,000 USD and supports compatibility with Galaxea's VLA-based control systems.
A key feature of the R1 series is its teleoperation capability, which serves both as a direct control mechanism and as a means of collecting training data for AI models. Galaxea offers the R1-T Isomorphic Remote Operation Platform, a scaled-down physical replica of the R1 that enables operators to control the robot through direct body movement mapping.
The isomorphic teleoperation system provides full-joint mapping between the operator's movements and the robot's kinematics. This approach has several advantages over VR-based teleoperation:
The system achieves millimeter-level precision with millisecond-level response times. Default chassis velocities are 0.2 m/s for linear motion and 0.6 rad/s for angular motion. Both wired and wireless connections are supported, with wired connections recommended for stability in multi-machine setups.
The R1 and R1 Pro also support VR-based teleoperation, offering an alternative control paradigm where operators use a VR headset and controllers. This mode is useful for remote operation scenarios where physical proximity to an isomorphic control platform is not practical.
Teleoperation plays a central role in Galaxea AI's approach to embodied AI development. Rather than relying primarily on simulated data or [[reinforcement learning]], the company emphasizes real-world data collection through teleoperation combined with [[imitation learning]]. CEO Gao Jiyang has stated that "at least for the next three years, reinforcement learning with simulators will not lead to the ultimate goal" for upper-limb manipulation tasks. The teleoperated data collected by R1 and R1 Lite units feeds directly into the training pipeline for Galaxea's AI models.
Galaxea AI has developed a proprietary family of AI models called G0, which serve as the intelligence layer for the R1 series. The G0 system is built on Vision-Language-Action (VLA) architecture, a class of models that combine visual perception, [[natural language processing|natural language understanding]], and physical action generation into a unified framework.
The G0 model uses a dual-system framework inspired by cognitive science:
The two systems run asynchronously at different frequencies, with System 2 operating at a slower planning cadence and System 1 executing actions at higher frequency for real-time control.
G0 is trained using the Galaxea Open-World Dataset, which comprises over 500 hours of high-fidelity data collected in real-world environments. The dataset covers more than 150 distinct tasks across 50 different scenes where people live and work. Training follows a three-stage curriculum:
In January 2026, Galaxea released G0Plus, an updated VLA model for multi-task robot manipulation. The G0Plus weights were further updated in February 2026, trained on larger-scale teleoperation and web data. The company also released G0Tiny, a 250-million-parameter model built on a SmolVLM2 backbone, optimized specifically for edge deployment on the R1 Pro's Jetson Orin computing platform.
The G0 models and associated tools are available through Galaxea's open-source GalaxeaVLA repository on GitHub and model weights are hosted on [[Hugging Face]].
The R1 series targets several application domains:
The primary commercial use cases include bin picking, line feeding, kitting, and parcel handling in factory and warehouse environments. The R1 Pro's 10 kg dual-arm payload and force-controlled interaction make it suitable for tasks requiring safe human proximity and compliant manipulation. In the second half of 2025, the R1 platform entered large-scale deployment, with multiple reports indicating that Galaxea secured orders or letters of intent for thousands of units from leading companies in the automotive and logistics sectors.
Universities and research laboratories use the R1 and A-series arms as testbeds for learning-based control, dexterous manipulation, and human-robot interaction research. The platform's ROS/ROS2 compatibility, open-source VLA models, and comprehensive documentation make it accessible for academic use. Clients in this sector include Stanford University and MIT.
The R1 can perform repetitive tasks such as restocking shelves or delivering items in hospitality and retail environments. Its autonomous navigation capabilities, combined with speech interaction features, enable basic customer-facing interactions.
Galaxea AI produces two robotic arm models used in the R1 series:
Both arms are also sold separately as standalone research and development platforms.
Galaxea AI operates in an increasingly crowded Chinese humanoid robotics market. According to the Beijing Academy of Artificial Intelligence, the competitive landscape in 2026 is organized into tiers:
| Tier | Companies |
|---|---|
| First Tier | [[Unitree Robotics]], Zhiyuan Robotics, [[UBTECH Robotics]], Galbot |
| Second Tier | Galaxea AI, LimX Dynamics, [[Xiaomi]], Meituan |
| Third Tier | Casbot, [[SenseTime]], [[XPeng]] |
Compared to bipedal humanoid competitors like Unitree's H1 and G1 or UBTECH's Walker, the R1's wheeled design gives it an advantage in stability and immediate deployability for indoor tasks, at the cost of reduced terrain versatility. The R1's starting price of approximately $28,000 USD is competitive within the Chinese market, where companies such as Unitree and [[Agibot]] have also pushed prices below $30,000 for entry-level humanoid platforms.
Galaxea AI distinguishes itself through its emphasis on real-world data collection and VLA-based AI models. While some competitors rely more heavily on simulation-based training, Galaxea's strategy of coupling teleoperation-based data collection with imitation learning reflects CEO Gao Jiyang's view that simulation alone is insufficient for achieving robust manipulation capabilities.
Galaxea AI has outlined several directions for future development:
With its Series B+ funding exceeding $291 million and a valuation surpassing $2.9 billion as of April 2026, Galaxea AI is positioned as one of the most heavily capitalized players in China's humanoid robotics sector.