AI-Based Autonomous Systems Research Lab

To build AI systems that earn the right to act. That means models which learn across distributed data without centralising it, stay robust when some of their inputs are wrong or adversarial, and report their confidence honestly enough that a clinician, a driver or a teacher can decide when to trust them.

The lab works at the point where machine learning meets real deployment — edge hardware with real limits, networks that drop packets, sensors that drift, and data that cannot legally leave the building.

Federated and multimodal intelligence that a hospital, a school district or a highway authority can adopt without surrendering their data or their judgement.

Over the next few years the lab is pushing toward models that unlearn on request, that carry calibrated uncertainty end to end, and that hold up under adversarial conditions rather than only on benchmark splits.