One model. Any arm you already run.
DecoDune's models are trained for contact-rich manipulation and deployed across industry-standard robotic arms — no proprietary hardware required.
The hard part of manipulation isn't picking things up. It's everything after.
Deformable objects, contact-rich assembly, and part-to-part variation remain some of the most persistent unsolved problems in industrial robotics. A cable that tangles differently every time. A connector that needs to be felt into place, not just positioned. These tasks have resisted fixed automation for over a decade — not because manufacturers haven't tried, but because the underlying problem is genuinely hard.
Three steps, one continuously improving model.
We build simulation environments for each task domain and collect real-world data from every deployment, creating a dataset that grows more valuable over time.
Our models learn manipulation skills that generalize across part variation — not fixed programs, but policies that adapt to what they encounter.
The same trained model deploys across supported arm platforms, so capability isn't locked to a single piece of hardware.
We validate on open infrastructure. We deploy on certified industrial hardware.
DecoDune's models are developed and tested using open robotics platforms, giving us faster iteration and full auditability during development. For production deployment, the same models run on certified industrial arms already trusted on manufacturing floors — so what you deploy has been proven before it reaches your line.