An Architecture for Integrating BDI Agents with a Simulation Environment

We present Simulated Autonomous Vehicle Infrastructure (SAVI), an open source architecture for integrating Belief-Desire-Intention (BDI) agents with a simulation platform. This allows for separation of concerns between the development of complex multi-agent behaviours and simulated environments to test them in. We identify and address the impedance mismatch between modelling and simulation, where time is explicitly modelled and differs from "wall clock" time, and BDI systems, where time is not explicitly managed. Our approach avoids linking the environment's simulation time step to the agents' reasoning cycles, relying instead on real time simulation where possible, and ensuring that the reasoning module does not get ahead of the simulation. This contributes to a realistic approximation of a real environment for the simulated BDI agents. This is accomplished by running the simulation cycles and the agent reasoning cycles each in their own threads of execution, and managing a single point of contact between these threads. Finally, we illustrate the use of our architecture with a case study involving the simulation of Unmanned Aerial Vehicles (UAVs) following birds.