Enhanced agent-oriented programming for robot teams

Iago de Oliveira Silvestre, Leandro Buss Becker, Michael Fisher, Jomi Fred Hübner, Maiquel de Brito

Abstract: The employment of Belief-Desire-Intention (BDI) agents offers a range of appealing features in the design of autonomous systems, including autonomy, adaptability, reactivity, and cooperation. However, when considering using BDI agents for controlling teams of multiple robots, few works explore the full potential of this approach. In this paper, we aim to integrate various BDI agent features to control a cooperative team of autonomous firefighting Unmanned Aerial Vehicles (UAVs), utilizing a layered architecture that combines the Jason BDI language with the Robot Operating System (ROS). In addition, we also present a new Mission-Management (MM) library, whose internal mechanisms enable the declarative definition of various missions, such as search-fire, extinguish-fire; assist other UAV, and refill. Missions can be temporarily suspended, resumed, or even interrupted. The underlying mechanism orchestrates the missions’ execution, making it simple to navigate between them. Our contention is that this library helps the developer to further exploit an important feature of BDI agents: the balance between reactiveness and proactiveness. An evaluation scenario was developed extending our previous firefighting-UAV application with significantly more functionality, incorporating wind direction and the finite nature of both battery-power and fire-retardant. Our new agent also includes a more sophisticated agent-collaboration algorithm based on the well-established Contract Net Protocol (CNP). Despite these significant extensions, the MM library ensured that the extended agent did not dramatically increase (around 17%) the code size of the original agent. Our experiments also show that the proposed solution does not present performance issues when scaling the application to more agents.

External IDs:dblp:journals/eaai/SilvestreBFHB25