Countries need physical AI solutions and autonomous robots to realize and scale their manufacturing needs. Studies show that growth in industries where large parts are assembled — space and automotive among them — is constrained more by manufacturing limits than by engineering limits.
Built For Human Environments, Not Around Them
We see a future where robots that can operate in environments already designed for human working conditions bring immediate effects to manufacturing. Large machines like cranes can still be operated by humans, and generally don’t need to be produced at scale. But a factory robot, produced at scale, can have immediate effects across industries constrained by human labor.
Five-Fingered Hands, and Teaching Them to Work
Five-fingered dexterous hands are essential for robots to perform the wide range of tasks found in human environments, since most tools and everyday objects are designed for human use. But modern robotics still lacks true autonomy, because no standardized software framework exists that lets robots independently learn and execute unfamiliar tasks.
To replicate the learning capabilities of a 20-year-old human without requiring decades of experiential learning, we’ve built a wearable robotic system that lets humans transfer their knowledge and experience directly into robotic systems. It lets an operator communicate, move, and feel the same forces sensed by the robotic hands — a more intuitive and efficient way to collect training data.
Wheels, Not Legs
Current robotic actuator technology isn’t yet as efficient as human joints. So rather than directly replicating the complex structure of human legs and feet — generally less important than hands for most industrial tasks — our robot uses a wheel-based mobility system for greater stability and energy efficiency. That design trades away stair-climbing, so we’ve added a height-extension mechanism that lets the robot reach up to nine feet, accessing elevated work areas in factory environments without needing to navigate staircases.
Built to Run Without Stopping
To maximize productivity and minimize downtime, we’ve built a hot-swappable battery system that lets the robot autonomously replace depleted batteries with fully charged ones. The batteries live in the wheelbase, adding weight low in the robot and improving overall stability.
The wheelbase also includes an adjustable center-of-gravity mechanism that can shift the batteries horizontally, keeping the robot balanced while lifting or carrying heavy objects — particularly when its arms are extended away from its central axis.
The AI Stack
AGI needs systems that can continuously collect real-world data and learn at scale. To be truly useful in the physical world, it will need a humanoid form factor — our environments, tools, and infrastructure are designed around the human body.
America needs its own equivalent of Unitree. Scaling physical AI research in the US requires accessible, capable robotic platforms. Buildo provides the infrastructure to collect real-world data and train models tailored to your specific use case. The Buildo AI model is highly scalable and modular.
We’ve built a three-layer architecture that handles everything from high-level reasoning down to sub-100ms fine-motor control:
System 2 (Server) — Vision-Language Action Model. Understands human instructions and decomposes complex tasks into sequential steps. Perceives the environment and plans actions accordingly. Provides high-level reasoning, decision-making, and long-horizon planning with sufficient generalization for robots to operate across diverse scenarios. Operates at ~1 Hz.
System 1 (On-board computer) — Foundation model responsible for motion planning and coarse action control. Ensures the robot approaches objects in an optimal way prior to contact. Operates at ~10 Hz. Most application-related training runs on System 1 using public or private domain data.
System 0 (On-board microcontroller) — Super high-frequency model responsible for instantaneous interaction via fine-motor control. Leverages tactile feedback in real time to continuously readjust hand and finger positions during contact with objects. Operates at ~100 Hz.
Build Your Robot App
Consider a simple pick-and-place warehouse robot. In the US, a single robot AI model can generate $20 per operating hour — $3,200 per month. A warehouse deployed with 20 robots generates approximately $64,000 per month, or $768,000 per year, in revenue.
The Team
At StarForge, we are engineers with firsthand experience in liquid bi-propellant engine design, systems engineering, and full-stack humanoid robots. We are building “Made in US” robotic and aerospace infrastructure from scratch — knowledge that is extremely rare in the world.
If you are interested in learning any part of this, contributing to the mission, or working alongside one of the most ambitious teams, reach out.
This pre-seed is funded by the founder. We are open to offering equity to early supporters and engineers who wants to build the infrastructure layer for orbital data centres and, eventually, a Type II civilisation.