CtrlK
| Galbot G1 Pro | |
|---|---|
| General information | |
| Manufacturer | Beijing Galaxy General Robot (Galbot) |
| Country of origin | China |
| Year introduced | 2025 |
| Status | In production (limited availability) |
| Type | Wheeled semi-humanoid robot / mobile manipulator |
| Also known as | Galbot G1 Premium |
| Price | ~$55,000 USD (estimated) |
| Website | galbot.com |
The Galbot G1 Pro (also referred to as the Galbot G1 Premium) is an upgraded variant of the Galbot G1 wheeled semi-humanoid robot, developed by Beijing Galaxy General Robot Co., Ltd. (commonly known as Galbot, Chinese: 银河通用机器人). Unveiled at the 2025 World Robotics Conference (WRC) in Beijing, the G1 Pro represents a significant hardware and software advancement over the base G1, most notably through the integration of the NVIDIA Jetson AGX Thor computing platform. Galbot claims to be the first company in the world to integrate the Jetson Thor processor into a commercially deployed humanoid robot.[1]
While the standard Galbot G1 established itself as a practical mobile manipulator for retail and pharmacy automation, the G1 Pro is designed for higher-demand industrial and commercial environments. It features an increased payload capacity of 15 kg (up from 5 kg on the base model), enhanced dual-arm dexterity for sorting and assembly tasks, and substantially more AI compute power for real-time reasoning, perception, and decision-making.[2][3]
The base Galbot G1 was introduced in 2024 as Galbot's first commercial product. It combined a human-like dual-arm upper body with a 360-degree omnidirectional wheeled chassis, prioritizing stability, long battery life, and practical deployment over bipedal locomotion. Powered by an 8-core CPU and NVIDIA Jetson Orin AI processors, the standard G1 quickly found commercial traction in Meituan-partnered pharmacies and convenience stores across Beijing.
The G1 Pro builds on this foundation while addressing two key limitations identified during early commercial deployments. First, the 5 kg per-arm payload of the base model restricted the range of items the robot could handle, particularly in industrial and logistics settings where heavier objects are common. Second, the Jetson Orin platform, while capable for basic autonomous retail operations, lacked the compute headroom needed for more complex real-time reasoning tasks such as multi-step manipulation planning, dense environment navigation, and handling intentional disruptions from human workers on factory floors.[4]
Development of the G1 Pro was driven by feedback from Galbot's early industrial partners, including Bosch, CATL, and Toyota, who required a platform that could operate in more demanding manufacturing environments while retaining the G1's proven wheeled-base design and long operational runtime.
The G1 Pro sits between the standard G1 and the Galbot S1 in Galbot's product hierarchy. The standard G1 targets retail, pharmacy, and household applications at a price point of approximately $87,000 to $97,000 USD. The G1 Pro, priced at roughly $55,000 USD according to early listings, is optimized for light-to-medium industrial use cases requiring greater payload and AI capability. The S1, a fully bipedal humanoid unveiled in January 2026, addresses heavy-duty manufacturing with a 50 kg continuous dual-arm payload and stair-climbing ability.[5]
This three-tier product strategy allows Galbot to serve a broad range of deployment scenarios, from unmanned convenience stores to automotive production lines, with purpose-built hardware at each level.
The G1 Pro retains the same basic form factor as the standard G1: a humanoid torso with dual articulated arms mounted on a 360-degree omnidirectional wheeled chassis. The robot stands 173 cm tall with a 190 cm arm span. The torso lift mechanism provides 65 cm of vertical travel, enabling an upward reach of approximately 2.4 meters, and the foldable body design allows the robot to bend down to access ground-level objects.[2]
The most significant hardware upgrades in the G1 Pro are the tripled payload capacity (15 kg total, up from 5 kg per arm), the integration of the NVIDIA Jetson AGX Thor processor, and refined multi-modal sensor arrays for enhanced perception.
| Category | Specification | Galbot G1 Pro |
|---|---|---|
| Physical | Height | 173 cm (5 ft 8 in) |
| Physical | Weight | 88 kg (194 lbs) |
| Physical | Arm span | 190 cm (6 ft 3 in) |
| Physical | Torso lift | 65 cm (reach up to 240 cm) |
| Mobility | Locomotion type | 360-degree omnidirectional wheeled chassis |
| Mobility | Degrees of freedom | 25 |
| Mobility | Max speed | 5 km/h (3.1 mph) |
| Mobility | Walking speed | 3 km/h (1.9 mph) |
| Manipulation | Total payload capacity | 15 kg |
| Manipulation | Grasp success rate | 95-97% (diverse materials) |
| Power | Battery life | 2 hours (high-performance mode) / up to 10 hours (standard tasks) |
| Computing | Processor | NVIDIA Jetson AGX Thor (Blackwell architecture) |
| Computing | AI compute | Up to 2,070 FP4 TFLOPS |
| Computing | Previous generation | NVIDIA Jetson Orin (base G1) |
| Sensors | Vision | RGB-D cameras (pure vision-based navigation) |
| Sensors | Tactile | Tactile sensors in hands |
| Sensors | Audio | 4-microphone array |
| Connectivity | Wireless | Wi-Fi (2.4 GHz and 5 GHz) |
| Connectivity | Wired | USB, Ethernet |
| Connectivity | Cloud | Cloud integration with NVIDIA Isaac Sim support |
| Software | AI models | GraspVLA, GroceryVLA, NavFoM, TrackVLA |
| Software | OS | EmbOS (proprietary), ROS2 compatible |
The following table highlights the key differences between the standard Galbot G1 and the G1 Pro.
| Feature | Galbot G1 (Standard) | Galbot G1 Pro / Premium |
|---|---|---|
| Processor | NVIDIA Jetson Orin + 8-core CPU | NVIDIA Jetson AGX Thor (Blackwell) |
| AI compute | ~275 TOPS (INT8) | Up to 2,070 FP4 TFLOPS (7.5x increase) |
| Energy efficiency | Baseline | 3.5x improvement over Orin |
| Weight | 85 kg | 88 kg |
| Payload capacity | 5 kg per arm | 15 kg total |
| Battery life | Up to 10 hours | 2 hours (high-performance) / up to 10 hours (standard) |
| Autonomy level | High (with occasional teleoperation) | 100% autonomous (no teleoperation required) |
| Target market | Retail, pharmacy, household | Industrial, automation, advanced retail |
| Price | ~$87,000-$97,000 USD | ~$55,000 USD (estimated) |
| Degrees of freedom | 25 | 25 |
| Height | 173 cm | 173 cm |
| Arm span | 190 cm | 190 cm |
The 7.5x increase in AI compute is the most significant upgrade, enabling the G1 Pro to run larger vision-language-action models in real time while maintaining fluid motion control. The 3.5x improvement in energy efficiency partially offsets the higher computational draw of the Thor platform, though battery life under sustained high-performance operation is reduced to approximately 2 hours compared to the standard G1's 10-hour runtime.[1][4]
The defining technical feature of the G1 Pro is its adoption of the NVIDIA Jetson AGX Thor processor. Announced at the 2025 World Robotics Conference in August 2025, this made Galbot the first company globally to deploy Jetson Thor in a commercially operating humanoid robot.[1]
The NVIDIA Jetson AGX Thor is a robotics supercomputer module built on the NVIDIA Blackwell GPU architecture. Key specifications of the Thor platform include:[6]
| Specification | Value |
|---|---|
| AI compute | Up to 2,070 FP4 TFLOPS |
| Memory | 128 GB |
| Power envelope | 40-130 W |
| CPU | 14-core Arm Neoverse-V3AE |
| GPU architecture | NVIDIA Blackwell |
| Multi-Instance GPU | Yes (Blackwell MIG) |
| Networking | 4x 25 GbE |
The Thor platform delivers 7.5 times the AI compute of the previous-generation Jetson Orin and 3.5 times greater energy efficiency. For the G1 Pro, this translates into the ability to run multiple large AI models simultaneously in real time, including vision-language models, manipulation planners, and navigation controllers, without the latency constraints that limited the base G1's responsiveness in complex scenarios.
Professor Wang He, Galbot's founder and CTO, stated: "Our G1 Premium, now running on NVIDIA Jetson Thor, delivers significant gains in speed and real-time reasoning, enabling our proprietary VLA models to achieve enhanced real-world performance."[1]
In demonstrations at the World Robotics Conference, the G1 Pro showed markedly improved fluidity of motion, precision drifting during navigation, and the ability to recover from intentional disruptions (such as staff kicking the robot or unexpectedly moving cargo) without interrupting its task execution. These capabilities earned the G1 Pro recognition as the "swiftest humanoid robot worker" at the WRC.[4]
The enhanced compute also enables what Galbot describes as "100% autonomous" operation, meaning the G1 Pro completes its assigned tasks entirely without teleoperation or remote human guidance, a claim validated at the 2025 World Humanoid Robot Games where the robot competed fully autonomously and won the gold medal.[7]
The G1 Pro runs the same suite of proprietary embodied AI foundation models as the standard G1, but the increased compute of the Thor platform allows these models to operate at higher resolution, faster inference speeds, and with greater context window for planning.
| Model | Function | Description |
|---|---|---|
| GraspVLA | Manipulation | A vision-language-action model trained on billions of simulated movements. Supports zero-shot learning, enabling the robot to grasp novel objects without task-specific training.[8] |
| GroceryVLA | Retail operations | Specialized for dense retail environments. Handles SKU identification, shelf scanning, and inventory management across thousands of product types. |
| NavFoM | Navigation | Foundation model for autonomous navigation using pure vision-based perception, without QR codes, pre-mapped routes, or external markers. |
| TrackVLA | Object tracking | Real-time tracking of objects and people in dynamic environments, supporting safe human-robot collaboration. |
On the Thor platform, these models benefit from Blackwell MIG (Multi-Instance GPU) technology, which allows multiple AI workloads to run concurrently on a single GPU without resource contention. This means the G1 Pro can simultaneously process visual inputs, plan manipulation trajectories, track dynamic obstacles, and generate natural language responses for voice interaction, all within the latency budget required for fluid real-time operation.[6]
Galbot's AI models are developed using a Sim2Real methodology that relies heavily on NVIDIA Isaac Sim and Isaac Lab, built on the Omniverse 3D graphics platform. The company's proprietary simulation dataset contains over 10 billion high-quality robotic action data points, covering grasping, navigation, deformable object manipulation, and multi-step task planning.[9]
The DexGraspNet dataset, developed in collaboration with NVIDIA, contains 1.32 million ShadowHand grasps on 5,355 objects across 133 categories. An updated version, DexGraspNet 2.0, provides dexterous grasping data for cluttered scenes and achieves a 90.70 percent real-world success rate for dexterous manipulation tasks.[10]
Galbot co-developed the OpenWBT_Isaac platform with Tsinghua University and the Shanghai Qi Zhi Institute. This simulation platform enables whole-body teleoperation of humanoid robots and serves as a key development tool for training and validating the G1 Pro's autonomous behaviors. The platform integrates the OpenWBT whole-body control system with virtual-real fusion capabilities, powered by NVIDIA L20 and RTX 5880 Ada GPUs. While the G1 Pro is designed for fully autonomous operation, OpenWBT_Isaac allows developers to collect high-quality human demonstration data for training new manipulation skills.[1]
The G1 Pro's increased payload capacity (15 kg) and enhanced AI compute make it better suited for industrial environments compared to the base G1. Target applications include:
The G1 Pro is positioned as a more capable alternative to the base G1 for partners like Bosch and Toyota, who require robots that can handle heavier components and operate in more dynamic factory environments while still fitting within flat-floor facility layouts where a wheeled chassis is advantageous.
While the base G1 handles the majority of retail deployments, the G1 Pro offers advantages in high-density retail environments where its superior AI compute enables faster, more reliable operation. Galbot's retail deployments include:
In pharmacy settings, both the G1 and G1 Pro can handle approximately 5,000 different SKUs. The G1 Pro's enhanced processing allows for faster identification and retrieval of medications from multi-tiered shelving systems, and its increased payload means it can carry heavier bulk items or multiple medications in a single trip.
One of the most significant commercial partnerships involving the G1 Pro platform is the joint venture between Galbot and Bosch. On June 17, 2025, Boyuan Capital (the market-oriented investment platform under the Bosch Group) and Galbot announced the formation of the Boyin Innovation Alliance, a joint venture focused on deploying embodied AI robots in industrial settings.[13]
Key elements of the partnership include:
| Component | Details |
|---|---|
| Joint venture name | Boyin Innovation Alliance |
| Partners | Boyuan Capital (Bosch Group), Galbot |
| Focus | Embodied AI in high-precision manufacturing |
| Strategy | "Global design, local production" |
| Target markets | Europe, North America, Southeast Asia |
As an early milestone, the Boyin Innovation Alliance signed a memorandum of understanding with UAES (United Automotive Electronic Systems) to establish RoboFab, a joint laboratory dedicated to developing automotive-specific expertise in embodied AI. RoboFab aims to create standardized frameworks for deploying humanoid robots on automotive production lines, transitioning from traditional rule-based automation to AI-driven adaptive systems.[13]
Dr. Ingo Ramesohl, Managing Partner of Bosch Ventures, stated: "Embodied AI holds transformative potential to redefine manufacturing processes." Professor Wang He added: "The future of manufacturing lies in intelligent, adaptive systems that can learn from real-world data."[13]
The Bosch partnership gives Galbot a pathway to international markets, particularly in European and North American automotive manufacturing, where Bosch has extensive existing relationships with OEMs.
Meituan, one of China's largest online services platforms, has been Galbot's most important deployment partner since the company's founding. Meituan led Galbot's angel funding round of approximately 700 million CNY (~$96 million) in June 2024 and has continued investing in subsequent rounds.[14]
The Meituan partnership drives Galbot's pharmacy and retail deployments. Beginning in late 2024, the two companies developed the 24-hour unmanned pharmacy concept that has since expanded across Beijing. The G1 and G1 Pro robots in these pharmacies perform the full cycle of retail operations: scanning shelves to monitor stock levels, autonomously restocking when items run low, locating and picking items for customer orders, and delivering products directly to customers or delivery couriers.
In March 2026, China's first fully robot-operated pharmacy went into service in Beijing, with the Galbot robot handling all medication sorting, retrieval, and dispensing tasks without human assistance.[11]
The G1 Pro and the Galbot S1 represent two different approaches within Galbot's product lineup. The following table compares the two platforms.
| Feature | Galbot G1 Pro | Galbot S1 |
|---|---|---|
| Type | Wheeled semi-humanoid | Bipedal humanoid |
| Locomotion | Omnidirectional wheels | Bipedal legs |
| Payload capacity | 15 kg | 50 kg continuous dual-arm |
| Battery life | 2-10 hours (task dependent) | Up to 8 hours (autonomous battery swap) |
| Stair climbing | No | Yes |
| Primary market | Industrial automation, advanced retail | Heavy-duty manufacturing, logistics |
| Key deployment | Smart pharmacies, Bosch JV, retail stores | CATL battery production lines |
| Navigation | Vision-based, no QR codes | Vision-based, 360-degree obstacle avoidance |
| Processor | NVIDIA Jetson AGX Thor | Not publicly disclosed |
| Price | ~$55,000 USD | Not publicly disclosed |
The G1 Pro is best suited for environments with flat floors and medium-weight handling requirements, where its wheeled chassis provides superior stability and energy efficiency. The S1 fills the gap for heavy-duty applications where bipedal mobility and 50 kg payloads are necessary, such as battery manufacturing at CATL.[5]
The G1 Pro (Premium) variant earned the gold medal at the 2025 World Humanoid Robot Games, held in Suzhou, China. Competing against 21 other teams, many of which relied on remote teleoperation, the Galbot robot operated with 100 percent autonomy throughout all rounds, from preliminaries to the finals.
In the Pharmaceutical Sorting Challenge, the robot autonomously scanned six tall shelves, identified nine specific medications from hundreds of items, and delivered them in 10 minutes and 22 seconds. The final score was 336 points, finishing over 160 points ahead of the second-place competitor. The victory demonstrated the G1 Pro's ability to perform complex, multi-step tasks in real-world-like settings without any human intervention.[7]
The G1 Pro was officially unveiled at the 2025 World Robotics Conference (WRC) in Beijing, where it was recognized as the "swiftest humanoid robot worker." During live demonstrations, the robot showcased fluid navigation, precision drifting maneuvers, and resilient autonomous operation under intentional disruption. The WRC debut coincided with the announcement of the NVIDIA Jetson Thor integration.[1]
While the Spring Festival Gala appearance on January 25, 2026 featured the Galbot platform broadly (not specifically the G1 Pro designation), the demonstration capabilities, including voice interaction, clothes folding, precise object manipulation, and walnut cracking, reflected the advanced AI and dexterity that the G1 Pro platform enables. The broadcast on February 16, 2026, triggered a surge in public interest: robot searches on JD.com increased 300 percent within two hours, customer service inquiries rose 460 percent, and order volume jumped 150 percent.[15]
The Galbot G1 Pro is listed at approximately $55,000 USD based on early third-party listings, making it significantly less expensive than the standard G1 (which lists at 699,700 CNY or roughly $97,000 USD on JD.com). This lower price point may reflect differences in configuration, a deliberate pricing strategy for industrial bulk orders, or regional market adjustments. Galbot has not published an official differentiated price list for the Pro variant as of early 2026.[2]
Galbot has indicated plans for dramatic cost reductions across its product line as production scales up. The company has discussed offering the G1 platform in sets priced at approximately 50,000 CNY (~$7,130 USD), representing a potential reduction of nearly 90 percent from current retail pricing. Such pricing would bring the platform into range for small business and household adoption.[16]
As of April 2026, the G1 Pro is available primarily in China. International availability is expected to follow Galbot's planned Hong Kong Stock Exchange IPO and expansion through the Bosch joint venture into European, North American, and Southeast Asian markets.
Galbot's funding trajectory has been among the most aggressive in the global humanoid robotics industry. As of early 2026, the company has raised over $800 million in total funding across multiple rounds, with a reported valuation of approximately $3 billion.
| Round | Date | Amount | Key investors |
|---|---|---|---|
| Angel round | June 2024 | ~$96M (700M CNY) | Meituan, Qiming Venture Partners, IDG Capital |
| Series A | June 2025 | ~$153M (1.1B CNY) | CATL (lead), Bosch (Boyuan Capital) |
| Series B | December 2025 | ~$300M | Beijing Robotics Industry Fund, GGV Capital, CICC Capital |
| Additional round | Early 2026 | ~$363M (2.5B CNY) | CATL, CMG Media Fund, Suzhou Venture Capital |
The angel round of 700 million yuan set a record for the embodied AI sector in China at the time.[14][17]
Galbot was founded in May 2023 by Wang He (王鹤), who serves as the company's founder and Chief Technology Officer. Wang He holds a bachelor's degree from Tsinghua University (2014) and a Ph.D. in computer science from Stanford University (2021), where he studied under Professor Leonidas J. Guibas. Before founding Galbot, he was a tenure-track Assistant Professor at the Center on Frontiers of Computing Studies (CFCS) at Peking University and a research scientist at the Beijing Institute for General Artificial Intelligence (BIGAI).[18]