Developmental Robotics

This project, at the nexus of cognitive science, machine perception, and artificial intelligence, develops physical agents that must solve many of the same problems that must be solved by the human brain. The focus is on learning sensory concepts from very little supervision, or from unsupervised, curiosity-driven active exploration. As part of this effort, ergalic theories are articulated that identify what specific software architectural features are needed to enable such learning and predict when that learning will occur in human beings.

The deeper goal of this project is to understand the computations that happen in the brain to form "meaning" out of the "blooming, buzzing confusion" that impinges on our sensory apparatus. We hypothesize that by building artificial tools that do these human-like tasks, we may be able to shed light on what architectures and computations may actually be present in the brain when solving those tasks. Of course, we also wish simply to have tools that can do these things, regardless of what it says about human biology. But the two are somewhat inextricable since we tend to approach tasks that are of interest to people, not to aliens.

More Information