What does it take for a humanoid robotics startup to justify a valuation of $3 billion? This question is answered by Galbot, a Beijing-based humanoid robotics startup that achieved the rare fusion of technical expertise, scale adoption, and whole-stack embodied AI. This move has now attracted over $300 million in new funds, taking the total amount to $800 million, setting a record in the industry.
Galbot’s strategy involves in-house development at all levels of its embodied AI stack, from proprietary datasets in the order of hundreds of billions, foundation models adapted for physical interaction, all the way to custom humanoid robotics hardware. This allows it to offer functionalities which remain out of reach for most firms, such as multi-task generalization, full-body control, cross-embodiment autonomous navigation, and manipulation using its hands, among others. These are more than just marketing features; they represent years of engineering on coordinated motion planning, cross-embodiment learning architectures, among other functionalities.
In the manufacturing domain, the humanoids developed by Galbot are already setting up autonomous operations at facilities owned by CATL, Bosch, Toyota, and Hyundai. This will require their robots to deal with tasks such as tote transfer, parts, and tool manipulation in a human-like manner. This will require sophisticated algorithms regarding the control of the entire body, including the chassis, arms, and torso, similar to the algorithms used for the possible energy of the arm in the study of humanoid robots designed for mobility.
The retail implementations under the brand name “Galbot Store” demonstrate another application of the technology: autonomous customer service. In more than 30 cities, the Galbot G1 humanoid robot’s capability to navigate complicated store layout configurations, naturally interact with customers, and execute the fine motor skills of handling merchandise as well as making payments is in use. Such autonomy is enabled by cross-embodiment-based navigation systems, wherein the system ignores the differences of the hardware implementations and enables the same policy for controlling different robots.
In warehouse logistics, the systems of Galbot operate in a functioning mode for more than a year and carry out their duties on a 24/7 basis. This is made possible by the presence of advanced motion planning and obstacle avoidance algorithms, combining task space and joint space controls for smooth manipulation and transfer of products. There is also the use of dynamic movement primitives to adapt the trajectories in real time.
Healthcare pilots with hospitals such as Xuanwu Hospital integrate humanoid robots into patient rooms, pharmacies, and guiding stations. In this case, fine manipulation in the hands becomes an important area, as tasks involving the manipulation of packets in healthcare involving patient drugs require manipulation skills not exceeding a sub-centimeter range. The foundation models by Galbot use particle-world models to ensure high-level manipulation tasks can be transferred to other equipment using human and robotic data sources. Transfer between models is an advantage in the healthcare sector where equipment needs to be seamlessly integrated into healthcare systems when an equipment upgrade takes place.
Investment-Wise, the fact that Galbot has been employing such large scale and diversity evidences their readiness for rapid ramp-up for productions. Galbot’s capability to buy contracts for thousands of units for manufacturing, perform self-guided retail operations in dozens of cities, and maintain operations within warehouses for one whole year reads not just their technical maturity but marketing savvy as well. For technologists working within the realm of robotics, the combination of sophisticated control software strategiescoupled with mass-productible hardware manufacturing exemplars a model for the realization of cognitive AI.
For AI researchers, Galbot’s project exemplifies the potency of integrated state-action space and large dataset training for realizing general-purpose humanoid abilities. With its latest round of funding, Galbot is poised to not only enhance its technology offering by expanding into other industries, but also break the barriers that currently limit what can be accomplished by humanoid robots. It is certainly one that will be followed with keen interest by everyone interested in the development of embodied AI.
