Real-Time BDI Agents: a model and its implementation

TL;DR

This paper introduces a real-time extension to the BDI agent model, incorporating explicit time management to improve responsiveness and effectiveness in real-world, time-constrained environments.

Contribution

It redefines the BDI control loop with real-time algorithms and provides an implementation validated in a resource-collection video game scenario.

Findings

Enhanced agent responsiveness under real-time constraints

Effective management of goals, plans, and actions with respect to time and resources

Successful validation in a resource-collection video game scenario

Abstract

The BDI model proved to be effective for developing applications requiring high-levels of autonomy and to deal with the complexity and unpredictability of real-world scenarios. The model, however, has significant limitations in reacting and handling contingencies within the given real-time constraints. Without an explicit representation of time, existing real-time BDI implementations overlook the temporal implications during the agent's decision process that may result in delays or unresponsiveness of the system when it gets overloaded. In this paper, we redefine the BDI agent control loop inspired by well established algorithms for real-time systems to ensure a proper reaction of agents and their effective application in typical real-time domains. Our model proposes an effective real-time management of goals, plans, and actions with respect to time constraints and resources availability. We propose an implementation of the model for a resource-collection video-game and we validate the approach against a set of significant scenarios.