A Generic Distributed Simulation System for Intelligent Agent Design and Evaluation

Using a simulator to design and evaluate intelligent agents in realistic environments places enormous demands on a simulation tool: everything from supporting multiple agents and their interactions, to providing detailed control over trials in an environment, to accurate perception within computational bounds. While the computationally intensive nature of this process is the most obvious reason to consider distributed simulation, we have also found that distributed simulation provides solutions to timing and perceptual problems that are particularly difficult in single-system simulation. This paper describes ongoing work on DGensim, a distributed version of the Gensim single-system simulator, and the significant advantages that distribution brings to the simulation process in this case. We also discuss the difficulties of preserving the generic aspects of a simulator in a distributed setting.