Insider Brief
- Galbot has integrated NVIDIA’s Jetson AGX Thor into its G1 Premium humanoid robot, showcased at the World Robotics Conference, achieving 7.5x more AI compute and 3.5x greater energy efficiency than previous platforms.
- The G1 Premium demonstrated advanced autonomy and fluid motion, earning the gold medal at the World Humanoid Robot Games, and is already deployed in smart pharmacies in Beijing with expansion planned to 100+ locations.
- Galbot’s progress is underpinned by its Sim2Real methodology and proprietary embodied AI foundation models, enabling zero-shot grasping, autonomous navigation, and real-world adaptability across healthcare, retail, and industrial environments.
Beijing-based Galbot announces it is the first company in China to integrate Nvidia’s Jetson AGX Thor chip into a humanoid robot. The company said it will create a faster, smoother, and more adaptive autonomy that sets a higher bar for general-purpose robotics.
“Our G1 Premium, now running on Nvidia Jetson Thor, has already demonstrated remarkable advancements in speed and improved real-time reasoning capability,” said Professor Wang He, Founder & CTO of Galbot. “This early adoption empowers us to push our proprietary VLA models to new levels of real-world capability.”
According to Galbot, the Jetson Thor upgrade delivers more than seven times the AI compute of the previous generation and over three times the energy efficiency. In practice, that means the G1 Premium can plan, move, and react with a fluidity that looks far closer to human capability than earlier versions. At the World Robotics Conference (WRC), Galbot demonstrated this leap with precise navigation, real-time reasoning, and even controlled drifting—earning it the reputation as the “swiftest humanoid worker” on display.
Performance in the lab is one thing. Performance in competition is another. The company pointed out Galbot proved both, clinching gold at the World Humanoid Robot Games by completing a fully autonomous challenge in just over 10 minutes—more than 160 points ahead of its nearest competitor. The robot’s ability to identify and retrieve medications from complex shelving demonstrated it’s abilities in healthcare settings and Galbot noted its robots are already serving customers in more than 10 smart pharmacies in Beijing, with plans to expand to over 100 nationwide by year’s end.
Earlier this month, they introduced the “Galbot Capsule,” an autnomous, 24/7 retail system that serves snacks and drinks. In parallel, partnerships with manufacturers like Bosch Group are bringing its embodied AI systems into industrial production lines, replacing rigid robotic arms with more adaptive, general-purpose machines.
Behind these achievements is a technical foundation built on simulation-to-reality learning, the company said. Using Nvidia’s Isaac Sim and Isaac Lab, Galbot pre-trains its robots on billions of synthetic grasping and manipulation scenarios before fine-tuning them with smaller amounts of real-world data. The result is a family of proprietary embodied AI models—GraspVLA, TrackVLA, and GroceryVLA—that give Galbot’s machines zero-shot capabilities in grasping, navigation, and retail operations.
The company is also pushing collaborative platforms. In partnership with Tsinghua University and Shanghai Qi Zhi Institute, Galbot introduced OpenWBT_Isaac, a teleoperation system for humanoid robots in simulation. By pairing Nvidia GPU acceleration with virtual-real fusion, the platform allows rapid data collection and cross-robot compatibility, providing researchers with a scalable testbed for embodied AI.
