Learning Virtual Agents for Decision-Making in Business Simulators

In this paper we describe SIMBA, a simulator for business administration, as a Multi-Agent platform for the design, implementation and evaluation of virtual agents. SIMBA creates a complex competitive environment in which intelligent agents play the role of business decision makers. An important issue of SIMBA architecture is that humans can interact with virtual agents. Decision making in SIMBA is a challenge, since it requires handling large and continuous state and action spaces. In this paper, we propose to tackle this problem using Reinforcement Learning (RL) and K-Nearest Neighbors (KNN) approaches. RL requires the use of generalization techniques to be applied in large state and action spaces. We present different combinations in the choice of the generalization method based on Vector Quantization (VQ) and CMAC. We demonstrate that learning agents are very competitive, and they can outperform human expert decision strategies from business literature.