My research goal is to understand the computational mechanisms that enable people to plan while building models of the world and to instantiate these mechanisms on robots. Planning and building models of the world are tightly coupled problems since agents must plan in spite of their uncertainty about how the world works. Indeed, agents must choose actions that improve their world models so that they may achieve their goals. I aim to provide an account of how this is possible by developing computational models of planning and world modeling, testing them against human behavior, and deploying them on robots.
My research principally draws upon the tools of probabilistic generative modeling, Bayesian inference, and model-based planning. Probabilistic generative models provide a flexible representation for modeling the world, Bayesian inference gives a framework for maintaining beliefs over these models, updating them as new evidence arrives, and model-based planning provides a mechanism for selecting actions to achieve a goal given the agent’s beliefs. Together, I believe these tools can provide a framework for modeling how people act while learning how the world works.
I expand upon these ideas in my research statement.