Robots trained by examples, not programming.
We build software that lets existing robots learn new tasks, handle real-world variation, and adapt to new environments with minimal human involvement.
Existing robots, new skills
We focus on making deployed robot hardware useful in more flexible and messy workflows.
From demos to autonomy
We aim to automate the data collection, model training, and deployment process.
How VLA systems learn a robot task
A simple three-step view of how we go from human demonstrations to an autonomous robot skill.
1. Data collection
A human demonstrates the task through teleoperation so the robot can gather grounded examples of what good behavior looks like.
2. Model training
Those demonstrations become training data for a vision-language-action policy that learns how to connect what it sees to the right motion.
3. Autonomous robot
Once trained, the robot can repeat the skill autonomously and handle more variation than a rigid scripted workflow.
What we are building
Software that helps robots become more useful in flexible, changing, and difficult-to-automate work.
Teach new robot skills through demonstrations instead of specialist programming.
Adapt robots to new tasks and changing environments with less manual rework.
Reduce the human effort needed for deployment, retraining, and switching tasks.
Why it matters
Many valuable tasks remain difficult to automate with conventional robot programming, leaving people to perform repetitive, physically demanding, or labour-intensive work. Advances in robot learning now make it possible to train adaptable skills from demonstrations.
This can reduce the engineering time and cost required to automate new tasks, make smaller production runs more economical, and help manufacturers respond faster when products or processes change.
This matters beyond individual factories. Europe faces labour shortages, demographic pressure, and a growing technology gap with the United States and China. Building robotics and physical AI in Europe is essential to maintaining industrial capability, economic resilience, and the ability to shape technology around European priorities such as well-being, sustainability, and social responsibility.
Come meet us
Who we are
MAD Robotics is being built by doctoral researchers at the University of Oulu working on adaptable, real-world robotic manipulation.
Adriell Amor
Doctoral researcher in BISG at the University of Oulu, focused on learning-based control for robotic manipulation, teleoperation-driven data collection, vision-language policies, and embedded deployment.
Miika Malin
Doctoral researcher in BISG at the University of Oulu, focused on data- and compute-efficient machine learning for spatiotemporal control, training pipelines, and runtime performance on constrained hardware.
Contact
Want to continue the conversation?
If you are working on a real robot task, building in robotics or AI, investing early, or curious about joining the team, reach out directly.
Our long-term goal is robotic intelligence that can adapt with minimal human involvement, from industrial environments on Earth to remote operations in space.
Send an email