Towards a Jason Infrastructure for Soccer Playing Agents

AgentSpeak and its practical interpreter Jason represent an excellent framework for implementing complex, reasoning agents. This paper discusses an ongoing research dedicated to extending Jason with the support for soccer playing agents. The end goal is to design an efficient infrastructure, capable of deploying and running BDI agents in the RoboCup soccer simulation league. 1 Intelligent agents playing soccer RoboCup is an annual, internationally-recognized competition of football/soccer playing robots [5]. By providing a formidable challenge in a fun environment, its main goal is to support and further motivate the development of various artificial intelligence techniques. Many concepts of the multi-agent technology, including autonomy, pro-active behaviour, coordination and cooperation, fit naturally into requirements of the RoboCup competition. These concepts are directly supported by the complex, Belief-Desire-Intention (BDI ) agent architecture [6]. The BDI architecture has a strong mathematical basis and is widely supported by a number of agent development frameworks [2]. Our previous work on deploying BDI agents in RoboCup simulations [4] was based on the agent-oriented programming language AgentSpeak and its accompanying interpreter Jason [1]. The main reasons Jason was selected as for this task include its direct support for BDI , and a high level of customizability. By analyzing the inner workings of Jason and the simulator, it was concluded that both systems support agents that operate in sense-think-act cycles. This fact simplifies the integration process significantly. To deploy Jason agents, it ? This work was partially supported by Ministry of Education, Science and Technological Development of the Republic of Serbia, through project no. OI174023: ”Intelligent techniques and their integration into wide-spectrum decision support” Towards a Jason Infrastructure for Soccer Playing Agents 295 is sufficient to extend and modify the following set of the interpreter’s subcomponents: – Simulated environment : a model of the game that enables the agents to sense their surroundings, and act accordingly. The environment was extended with custom parser and generator components which, respectively, extract agent’s belief literals from the simulator’s set of percepts, and transform agent actions into concrete effectors; – Execution control : handles Jason reasoning cycles. Development of a custom execution control was necessary for several reasons, including the support for key-framed motions. Key-framed motions often span across multiple Jason reasoning cycles. The execution control assures that the appropriate sets of commands are sent to the simulator as the motion progresses; – Agent architecture: a link between the simulated environment and the remaining components. Our custom implementation of these components was evaluated using a concrete implementation of a soccer playing agent [4]. The results have shown that Jason is perfectly capable of satisfying strict time constraints imposed by the official RoboCup simulator. However, further improvements and extensions are needed in order to implement and deploy agents that exhibit more complex behaviour. Our ongoing work is dedicated to designing and re-implementing the remaining parts of the Jason infrastructure [1]. This step is necessary in order to fully integrate Jason into the RoboCup simulator, allowing Jason agents to actually compete against other teams, and to do so by relying on extensively researched and well-understood concepts and methodologies of the multi-agent technology. In the long run, the plan is to further extend the infrastructure with MOISE+, an advanced Jason-compatible framework for organizational modelling that has already been tested in virtual soccer simulations [3].