All ProjectsContact Dr. Hossain
UASLifelong LearningAutonomy
Self-evolving Unmanned Aerial Systems
Most autonomous aerial systems are frozen at deployment. The policy that flies the aircraft is the policy that was trained in simulation months earlier, while conditions drift, airframes wear, and missions change, so the gap between the training distribution and the operating environment only widens. This project studies aerial systems that keep learning in the field, updating their own policies from the data they generate while flying, without catastrophic forgetting and without drifting outside the behavior their operators approved.
Research Objectives
- Develop on-platform learning that adapts control and planning policies from operational data
- Prevent catastrophic forgetting so adaptation on a new mission does not erase earlier competence
- Keep adaptation inside verified behavioral bounds so autonomy does not drift out of certification
- Share learned experience across a fleet without moving raw sensor data off the aircraft
Methods & Techniques
- Continual and lifelong reinforcement learning with experience replay
- Sim-to-real transfer and online domain adaptation
- Federated learning across fleets under bandwidth and privacy constraints
- Runtime monitors that constrain policy updates to an approved operating envelope
Interested in this research?
Get in touch to discuss collaboration or graduate opportunities.