Simulation with learning agents

We propose that learning agents (LAs) be incorporated into simulation envir onments in order to model the adaptive behavior of humans. These LAs adapt to specific circumstances and events during the simulation run. They would select tasks to be accomplished among a given set of tasks as the simulatio n progresses, or synthesize tasks for themselves based on their observation s of the environment and on information they may receive from other agents. We investigate ail approach in which agents are assigned goals when the si mulation starts and then pursue these goals autonomously and adaptively. Du ring the simulation, agents progressively improve their ability to accompli sh their goals effectively and safely. Agents learn from their own observat ions and from the experience of other agents with whom they exchange inform ation. Each LA starts with a given representation of the simulation environ ment from which it progressively constructs its own internal representation and uses it to make decisions. This paper describes how, learning neural n etworks can support this approach and shows that goal-based learning,may be used effectively used in this contest. An example simulation is presented in which agents represent manned vehicles, they are assigned the goal of tr aversing a dangerous metropolitan grid safely and rapidly using goal-based reinforcement learning with neural networks and compared to three other alg orithms.