Joshua Eckroth

PhD Student

My aim is to develop strategies that improve the ability of an intelligent agent (such as a robot) to perform challenging reasoning tasks even when the agent's knowledge about the world is significantly limited or the sensors that provide information about its domain (e.g., video cameras and microphones) are sometimes inaccurate, misreporting, or otherwise malfunctioning. My approach is to provide the agent with a self-reflective "metareasoning" capability that enables it to detect its own states of confusion and perform repairs in its reasoning processes. To demonstrate the effectiveness of this new facility, I experiment with software simulations, and compare the performance of agents with this facility to those without this facility. In order to support the claim that the metareasoning capability is effective across several different task domains, the experimental agents perform a variety of tasks such as plan recognition and tracking moving objects with simulated video cameras.

Picture for joshuaeckroth.1

 395 Dreese Laboratories2015 Neil AveColumbus, OH 43210