My research objective is to develop robots and decision-making systems that can learn from limited data and rapidly generalize by studying the cognitive foundations of human intelligence. I aim to understand the computational principles that enable people to plan, infer goals, and solve novel problems, and to recreate these abilities in robotics and AI systems. Humans exhibit remarkable data efficiency: we can infer structure, adapt to new situations, and recombine known skills to solve previously unencountered tasks. In contrast, many state-of-the-art neural-based AI systems rely heavily on imitation and large-scale data collection and have an arguably limited ability to adapt to new tasks.
My research seeks to overcome these limitations by uniting ideas from cognitive science and robotics—specifically, by combining Bayesian inference, which enables agents to form and update beliefs about the world, with robot planning, which enables them to act to achieve novel goals. I am now interested in applying program induction to robotics, allowing robots to synthesize flexible, structured “programs” that describe how to achieve a task rather than following fixed action sequences. This approach aims to give robots the ability to represent and reuse behavioral structure, improving both their generalization and interpretability.