We’re accelerating robot development by generating and testing 1000s of designs in simulation
When building a novel robot, you don’t know how it should look.
Which drivetrain is best? Tracks, differential drive, skid steer, swerve drive?
How should the robot be articulated? Joints between wheels, arm or other structure to end effector?
Should the robot be split into multiple ones? With same or different end effectors?
What sensors can guarantee safe operation? Where should they be placed?
Fabricator answers all these by trying 1000s of designs in simulation, automatically trains optimal controls for each of them and presents you with videos of how each design would behave, and metrics like speed, accuracy or cost.
We developed tile grouting robots in record time
We developed tile grouting robots, deployed in Singapore and ready to scale. They are operated by a single button, fully autonomous, small and easy to move around and hence able to boost productivity 5x.
Our goal is to make robot development 100x more efficient
Have you ever looked around and wondered where all the robots are? It's because new robots currently take around 5-10 years to develop. But that can be changed with recent advances in AI and simulation.
Every design choice, from drivetrain selection to sensor placement, has unpredictable ripple effects. Right now, the only way to get it right is trial-and-error: build, test, tweak—repeat. With evolving customer needs and inevitable surprises, the iterations pile up.
We want to revolutionize this process.
The mature version of our platform will generate robust, deployment-ready hardware design and software for any task and environment ― optimized for task completion, cost, ease of use and longevity.
This will make robotics 100x more efficient, unlocking thousands of new robot applications even in oceans, deserts and space.
%20(1).webp)
By having our own robots in production, we can guarantee the design automation platform makes accurate predictions
If we only worked in simulation, there wouldn’t much reason to trust the platform’s predictions, even if it’s from a physics simulation. But we can try what the platform predicts for our grouting robot. We can say “What if we switched to mecanum wheels?” Fabricator might predict it’s 5% faster but 5% less accurate but then we can actually check that in real life on real construction sites. And the prediction accuracy should transfer well to at least wheeled robots in solid environments, there we can be very confident about the prediction intervals given.
Want to change the future of robotics with us?
We are always looking for brilliant people! In Europe we are hiring for reinforcement learning and software engineering positions. In Singapore we are hiring for electrical, mechanical, firmware and software engineering. In both locations we are hiring full-timers, interns and limited remote positions.