A report the difference features of a multi-agent using an overlay knowledge in the fire panic problem

This report of this paper shows a performance of the heterogeneous multi-agent system using the overlay knowledge technique. The overlay knowledge is one kind of layered rule-based system. Basic idea is the tile-coding rule on reinforcement learning. But In the tile-coding rule, all reinforced rules are spread any categorical states of rule table. Proposal style of overlay knowledge is from a tile-coding. We examined the features of the learning principle about the heterogeneous multi-agent. It was proved that the overlay knowledge was useful sufficiently for the heterogeneous multi-agent. The robust characteristic was the cause which outputs absolute behavior. By changing of the robust levels, behaviors of the agent expected a change into the flexible movement.