About Me

I am a doctoral student in Computer Science at UMass, where I work with Andrew Barto and Rod Grupen. I study a subfield of Artificial Intelligence called Developmental Robotics; for more about my research, see below. I graduated from Swarthmore College in 2003, and from the Illinois Mathematics and Science Academy in 1999--both were amazing, wonderful, formative places.

Research

Developmental robotics. Artificial intelligence, robotics, computational reinforcement learning, machine learning, development, artificial life, evolutionary computation, self-organization, complex adaptive systems, agent-based modeling.

Through my research I seek to understand the computational principles of systems that display continual, open-ended, endogenous adaptation, by synthesizing similarly autonomous and adaptive artificial systems.

My driving interest is in discovering representations and algorithms for constructive, grounded, long-term, domain-general cognitive development in robots. Toward this end, I study the role of intrinsic motivation in computational reinforcement learning: an agent's motivational system provides reward signals to other learning systems (in contrast to the traditional practice of hand-crafting a reward function for a specific problem). I think that such motivation--which encompasses issues of curiosity, exploration, and attention--helps to coordinate cognitive development and is essential for continual learning. I am eager to prove the utility of a developmental approach to endowing real robots with useful abilities.

I have a sustained secondary interest in evolutionary computation and self-organization; this interest is most at home in the multidisciplinary field of Artificial Life.

Publications

  • Andrew Stout, George Konidaris, and Andrew Barto - Intrinsically Motivated Reinforcement Learning: A Promising Framework For Developmental Robot Learning. Proceedings of the AAAI Spring Symposium on Developmental Robotics, 2005. [pdf]
  • Andrew Stout and Lee Spector - Validation of Evolutionary Activity Metrics for Long-Term Evolutionary Dynamics. Proceedings of the Genetic and Evolutionary Computation Conference (GECCO), 2005. [pdf]
  • Julie Corder, Oliver Hsu, Andrew Stout, and Bruce Maxwell - A Modular Software Architecture for Heterogeneous Robot Tasks. Proceedings of the AAAI Workshop on Mobile Robotics, 2002.