Galbot

Galbot: Humanoid Robot and Embodied AI Company — Complete Guide

Design and Physical Specifications

The Galbot G1 is a wheeled semi-humanoid mobile manipulator — a design class that combines a human-like upper body with two dexterous arms on a wheeled mobile base, rather than using bipedal legs for locomotion. This design choice reflects a pragmatic engineering decision: for the environments where the G1 is deployed — retail stores, manufacturing floors, hospitals, warehouses — flat surface locomotion using wheels provides greater stability, energy efficiency, and reliability than bipedal walking, while the human-like upper body with dual 7-degree-of-freedom arms provides the dexterous manipulation range needed to access shelves, handle objects, and interact with human-scale workspaces.

The G1's 10-hour battery runtime is a critical operational feature for retail and manufacturing deployments where the robot must sustain continuous operation across a full commercial shift without recharging interruptions. The pure vision-based navigation system — operating without LiDAR — reduces hardware cost and simplifies the sensor maintenance profile while relying on Galbot's AI models to provide robust spatial awareness from camera data alone.

Dexterous Manipulation and Generalizable Grasping

One of the G1's defining engineering achievements is its ability to handle approximately 5,000 different types of goods without task-specific programming or parameter adjustments for each product category. This generalizable manipulation is the direct product of the GraspVLA foundation model and its underlying training on synthetic simulation data.

The robot can grasp deformable snack bags, rigid bottles, glass containers, plastic packaging, and pharmaceutical packaging — the full range of materials encountered in retail and pharmaceutical environments — without requiring an operator to configure separate handling routines for each product type. A 95 to 97 percent grasping success rate across diverse materials and configurations represents a level of reliability that is practically viable for commercial deployment in live customer-facing environments.

Technology and AI Systems

Sim2Real Training Methodology

Galbot's central technical differentiation is its Sim2Real training methodology, which enables the company to train robot manipulation policies using massive quantities of synthetic simulation data at a fraction of the cost of real-world data collection. Co-founder Zhang Zhizheng describes the approach as "massive pre-training on large-scale, high-quality synthetic datasets followed by surgical fine-tuning with minimal real-world data."

The company generates its synthetic training data using NVIDIA Isaac Sim, building physics-accurate simulations of object interactions, environmental variations, and manipulation tasks. The trained models are then refined with a targeted volume of real-world interaction data — a process Zhang describes as enabling "skills mastered in virtual environments to transfer seamlessly into novel real-world contexts with minimal semantic relabeling."

This approach has enabled Galbot to build what it describes as a "100-billion high-quality dataset" — a training corpus at a scale that would be economically impossible to construct through real-robot teleoperation alone. The company asserts that it is the world's first to achieve full-stack in-house development across hundreds-of-billion-scale high-quality datasets, embodied foundation models, and robotic hardware — a claim it has made as the basis for its competitive positioning against rival embodied AI companies.

GraspVLA

GraspVLA, released in January 2025, is Galbot's first publicly launched foundation model and the company describes it as the world's first end-to-end embodied AI grasping model trained on simulated synthetic data. The model is pre-trained on billions of simulated physical interactions and can achieve zero-shot generalization — the ability to handle new object types it has not explicitly seen before — using only the pre-training data, without additional real-world demonstrations. Galbot says GraspVLA enables robots to understand and adapt to new manipulation tasks without requiring specific training for each new task type.

The model was developed in part through Galbot's collaboration with NVIDIA Isaac Sim and draws on Professor Wang's DexGraspNet research as a foundational dataset for dexterous grasping across a wide object distribution.

GroceryVLA

GroceryVLA is Galbot's retail-specific embodied AI model, designed to solve the two primary manipulation challenges of commercial retail environments: intelligent grasping of large product catalogs and navigation through densely stocked shelves. In a scenario involving complex and densely stacked goods, the model does not need to adjust parameters for each product separately — it enables the robot to accurately grasp items regardless of whether they are soft bags, hard boxes, bottles, glass containers, or plastic packaging. GroceryVLA powers the G1's deployment in Galbot Store and the company's pharmacy and supermarket installations.

TrackVLA

TrackVLA is Galbot's navigation foundation model. The system supports pure visual environment perception, comprehension of natural-language commands, zero-sample generalization in complex scenes, and following of highly dynamic targets. A key operational feature is that TrackVLA can resume tracking after a temporary visual loss of the target — essential for following a human in crowded retail, hospital, or airport environments where the robot's line of sight to its target may be temporarily obstructed. This capability supports hospital guidance, retail escort, and logistics convoy applications.

The three VLA models — GraspVLA, GroceryVLA, and TrackVLA — collectively form the "brain" powering the G1's autonomous operation across different environments and task types.

Commercial Deployments and Partnerships

Industrial Manufacturing

Galbot has secured partnerships with some of the world's largest industrial manufacturers. The company describes itself as the first globally to deploy humanoid robots for real autonomous operations on manufacturing floors.

CATL: The world's largest battery manufacturer is both an investor in Galbot and an operational deployment partner. Galbot robots are running autonomously on CATL's battery production lines, handling tasks that require the combination of dexterity and precision that the G1 provides. CATL is the only major battery company to make a strategic investment in an embodied AI foundation model company, reflecting the depth of the commercial relationship.

Bosch: Galbot and Boyuan Capital — the investment arm of the Bosch Group — formed a joint venture focused on applying embodied AI in high-precision manufacturing, including complex assembly tasks. The venture is designed to drive large-scale industrial deployment of embodied AI in Bosch's global manufacturing network.

Toyota, Hyundai, BAIC Group, SAIC Motor, Zeekr, Great Wall Motor: Galbot has secured deployment partnerships with major automotive manufacturers across both Chinese and international OEM groups, with cumulative orders across all industrial manufacturing clients reaching thousands of units.

Galbot Store — Autonomous Retail

One of Galbot's most publicly visible applications is Galbot Store — a fully autonomous retail solution in which the G1 robot operates without human staff, serving customers directly. As of late 2025, Galbot Store locations had been established in over 30 cities across China.

The Galbot Store concept represents the first large-scale deployment of a humanoid robot as the primary service provider in a retail environment — the robot identifies products on shelves, retrieves requested items, processes transactions, and assists customers with product inquiries, all autonomously. The daily deployment of the G1 in live customer-facing retail settings, handling thousands of customer interactions, provides operational validation for the robot's reliability, safety, and handling precision that no controlled laboratory demonstration can replicate.

Healthcare — Xuanwu Hospital

In collaboration with Xuanwu Hospital, one of Beijing's leading medical centers, Galbot has deployed G1 robots to assist in patient rooms, pharmacies, and hospital guidance systems. The pharmacy deployment is one of the most demanding commercial applications for the G1: medication retrieval requires high accuracy and reliability given the safety implications of dispensing errors, and the pharmaceutical sorting challenge at the World Humanoid Robot Games provided independent competitive validation of the G1's accuracy in this context.

Galbot's pharmacy robots are deployed in more than 10 unmanned pharmacies across Beijing, operating 24/7 to handle medication sorting, customer service, and inventory management.

Warehouse Logistics

Galbot's warehouse solutions have demonstrated stable and continuous 24/7 operations for over a year across multiple deployment locations. Warehouse logistics involves the types of repetitive pick-and-place and transport operations that play directly to the G1's strengths — consistent precision handling across a large catalog of different goods, sustained operation across full shifts, and autonomous navigation through the dynamic, human-shared environment of a working warehouse.

World Humanoid Robot Games — Gold Medal

At the inaugural World Humanoid Robot Games held in Beijing in 2025, Galbot won the gold medal in the Robot Skills Competition. Competing against 21 teams, many of which relied on remote teleoperation rather than autonomous operation, the Galbot G1 achieved a score of 336 points — 160 points ahead of the runner-up.

The competition's Pharmaceutical Sorting Challenge required the robot to autonomously scan six tall shelves, identify nine specific medications from hundreds of products, and deliver them in just over 10 minutes. Galbot's G1 completed this task fully autonomously while competitors relied on human remote control — a distinction that underscores the practical significance of Galbot's autonomous AI approach relative to teleoperation-dependent alternatives.

CCTV Spring Festival Gala Performance

The Galbot G1 appeared on China's CCTV Spring Festival Gala — the most-watched broadcast event in China — performing tasks including rolling walnuts, picking up glass shards, folding clothes, skewering sausages, and retrieving items from shelves. The broadcast appearance provided national-scale public visibility for Galbot's technology and demonstrated the robot's ability to perform fine-grained manipulation tasks for a mass consumer audience.

Advantages and Differentiation

Sim2Real Synthetic Data at Scale: Galbot's ability to train on hundreds of billions of synthetic data points at a fraction of real-world data collection costs provides a scalable foundation for expanding the robot's task and object handling capabilities without proportional increases in training infrastructure cost.

Zero-Shot Generalization via GraspVLA: The ability to handle new object types without additional real-world demonstrations — using pre-training alone — means Galbot can deploy into new retail or industrial environments without extensive customization for each client's specific product catalog.

10-Hour Battery Runtime for Full Shift Operation: The G1's 10-hour endurance enables uninterrupted deployment across commercial operating hours without mid-shift charging, which is directly relevant to retail and manufacturing deployments where continuous uptime is essential.

One-Day Store Deployment: The G1's ability to be deployed into a new retail store environment within a single day dramatically reduces the integration cost and timeline for retail operators, making economic deployment at scale viable.

Validated Performance in Live Environments: Galbot's deployments in CATL battery lines, Bosch factories, live retail stores across 30 cities, Xuanwu Hospital pharmacies, and active warehouses provide production-verified performance data rather than laboratory-only demonstrations.

Strategic Investor-Partner Alignment: CATL's role as both investor and industrial deployer creates an unusual alignment of financial and operational interests, with CATL motivated to support Galbot's technical development as a direct user of its robots on production lines.

Frequently Asked Questions (FAQ)

What is Galbot? Galbot (Galaxy General Robot Co., Ltd.) is a Beijing-based humanoid robotics and embodied AI company founded in May 2023 by Professor He Wang of Peking University. The company develops the Galbot G1 robot — a wheeled semi-humanoid mobile manipulator with dual dexterous arms — and the proprietary AI foundation models (GraspVLA, GroceryVLA, TrackVLA) that govern its autonomous operation. Galbot has raised $800 million in total funding and reached a $3 billion valuation as of December 2025, with cumulative orders for thousands of units from industrial partners including CATL, Bosch, Toyota, and Hyundai.

How does the Galbot G1 robot work? The Galbot G1 navigates environments using its wheeled mobile base guided by TrackVLA, a pure vision-based navigation model that understands natural language commands and follows dynamic targets without LiDAR sensors. Its dual 7-DOF arms, controlled by the GraspVLA manipulation foundation model, identify and grasp objects across thousands of product types using RGB-D cameras and tactile sensors in the hands. GraspVLA was trained on billions of simulated physical interactions through Galbot's Sim2Real methodology — using NVIDIA Isaac Sim for synthetic pre-training — enabling zero-shot generalization to new objects and environments without task-specific programming.

What is Galbot's Sim2Real technology? Galbot's Sim2Real methodology trains robot manipulation AI using massive synthetic simulation data generated through NVIDIA Isaac Sim, then applies targeted fine-tuning with a small volume of real-world interaction data. This approach enables Galbot to build training datasets at the hundreds-of-billions scale that would be economically impossible through real-robot teleoperation alone. Galbot asserts that it is the world's first company to achieve full-stack in-house development across hundreds-of-billion-scale high-quality datasets, embodied foundation models, and robotic hardware.

What are Galbot's main commercial deployments? Galbot has four primary commercial deployment categories. In industrial manufacturing, the G1 operates on production lines at CATL, Bosch, Toyota, Hyundai, and major Chinese automotive manufacturers, with thousands of units on order. In retail, Galbot Store — a fully autonomous robot-staffed store — operates across more than 30 Chinese cities. In healthcare, G1 robots are deployed in more than 10 Beijing pharmacies and assist at Xuanwu Hospital. In warehouse logistics, the G1 has sustained continuous 24/7 autonomous operation for over a year across multiple warehouse facilities.

Summary

Galbot (Galaxy General Robot Co., Ltd.) represents one of the most commercially advanced embodied AI and humanoid robotics companies in operation as of 2026. Its Sim2Real training methodology, GraspVLA zero-shot manipulation foundation model, and production-validated G1 robot have collectively enabled a commercialization trajectory that few robotics companies achieve: live autonomous operations in consumer-facing retail stores across 30 cities, deployment on major automotive and battery manufacturer production lines, award-winning performance at international robotics competitions, and a prime-time national television appearance — all within two years of founding. With $800 million in cumulative funding, a $3 billion valuation, strategic partnerships spanning CATL, Bosch, Toyota, and the China Development Bank, and a technical foundation built on full-stack in-house development from datasets through hardware, Galbot has established itself as a leading figure in the commercialization of embodied AI and autonomous robotics at industrial and consumer scale.