Real data
Slow and expensive to collect
Recording robots doing hand tasks takes people, hardware and time, for every task and every hand.
The infrastructure for robot dexterity
Teaching robots
to use their hands.
Galexia builds the data and algorithms robots need to learn fine motor skills: high-fidelity simulation, grounded in real-world data.
The problem
Robots can walk and see. But they still can’t do hand tasks that are simple for us. Even chopping vegetables is hard for a robot. Fine motor control is the missing piece for robots to work in the real world.
Why it’s hard
The bottleneck is
data and algorithms.
Simulation
Hard to make real enough
Environments with the right physics for grip, contact, breaking and liquids are complex to build.
Algorithms
Hungry for data
Today’s learning methods need a lot of data to learn even one skill.
What we build
An engine for
robot dexterity.
Galexia gives teams both: the data, simulated and real, and the algorithms to train robots on hard hand tasks.
- 01
Pick a task
Crack an egg, slice a tomato, insert a cable.
- 02
Simulate it
High-fidelity physics for contact, breaking and liquids.
- 03
Ground it
Real-world data tunes the materials and checks the results.
- 04
Train it
Data-efficient algorithms, for any robot hand.
Where we start
The hard cases
first.
Robots practise these thousands of times in simulation before they touch real hardware.
Fragile
Objects that need a gentle grip
Breakable
Things that crack and come apart
Liquids
Fluids that pour and spill
Team
The founders.
Shine Tokas
Co-founder & CEO
Finishing a business degree at Shenandoah University, also studying computer science and AI. Builds AI products.
shine@galexia.aiKevin Ndumia
Co-founder & CTO
Five years in software and AI engineering, building software and machine learning systems, including LLMs and speech models.
kevin@galexia.aiGet in touch
Robots that need
hands?
We’re working with our first robotics teams. If your robots need to handle the hard stuff, let’s talk.