An Architecture for Situated-learning Agents.

Mitchell, Matthew
An Architecture for Situated-learning Agents
Ph.D. Thesis, Monash University, Australia, 2004. (PDF - 1.9 MB)

Abstract: This thesis looks at the problem of situated learning agents operating in real-world environments. It presents a reinforcement learning system which dynamically develops a connectionist model of its environment while learning. The learned model addresses both input-generalisation as well as hidden-state (temporal learning). The system is intended for non-stationary, non-deterministic environments which contain hidden-state. The system is demonstrated on some common input-generalisation and hidden-state problems and also on a simple problem which demonstrates its potential for on-line planning.