Keenan Albee

Research Summary

Increasingly capable algorithms are required to continue taking dynamic mobile robots out of the lab and into the field. Operating under challenging conditions with real-world constraints, these robots face situations that are uncertain, unknown, or unstructured with limitations in perception, real-time demands on compute, and often with complex or stochastic dynamics. I develop methods for adaptive decision-making under uncertainty for these systems, drawing from control theory, reinforcement learning, and stochastic modeling, with an emphasis on a "theory to practice” philosophy including field hardware validation: my work has flown on the International Space Station and, in the coming months, to the Moon. My overarching goal is to make autonomous robotic operations safer and more efficient when human-in-the-loop operation becomes infeasible, risky, or wasteful. To this end, some of my current research areas include:

1. Infusing learning-based tools into planning and control to improve efficiency and safety under imperfect knowledge.
2. Providing or enhancing safety guarantees for uncertain dynamical systems, including when new information is revealed online.
3. Considering perception, localization, and information gain explicitly in robotic planning.
4. Autonomy frameworks incorporating the above for exploration, space, and other extreme environment robotics applications.