This project is developing computational agents that operate for extended periods of time in rich and dynamic environments, and achieve mastery of many aspects of their environments without task-specific programming. To accomplish these goals, research is exploring a space of cognitive architectures that incorporate four fundamental features of real neural circuitry: (1) reinforcing behaviors that lead to intrinsic rewards, (2) executing and learning over mental, as well as, motor actions, (3) extracting regularities in mental representations, whether derived from perception or cognitive operations, and (4) continuously encoding and retrieving episodic memories of past events. A software framework called Storm facilitates this exploration by enabling the integration of independent functional subsystems, allowing researchers to easily plug in and remove different subsystems in order to assess their impact on the overall behavior of the system.
Cognitive architectures are being tested by exposing them to a wide variety of novel environments with unpredictable (and non-repeatable) extrinsic rewards, but in which many actions could lead to intrinsic rewards (e.g., surprise). To assess flexibility, an automated environment generator exposes agents to environments that are unknown in advance to the artificial agent or human researcher. To assess robustness, cognitive systems are being exposed to many variants of the same environment to ensure that the systems can learn from past experience and generalize when appropriate. And to assess autonomy, systems' must operate effectively for extended periods of time in a dynamic environment.
In the longer term, flexible and robust cognitive architectures being devloped under this research will have application as the 'brains' of robotic and software systems in emergency, miltary, and a wide variety of other societal and service realms.
This award is funded under the American Recovery and Reinvestment Act of 2009 (Public Law 111-5).
