Abstract:

In recent years the teaching of subjects related to Artificial Intelligence has grown notably in higher education degrees. This is the case of the discipline of multi-agent systems, which usually is part of the majority of master's degrees in Artificial Intelligence. Multi-agent systems (MAS) offer solutions for distributed decision making, where a set of autonomous intelligent agents must reach an agreement to solve a problem. These types of problems are usually complex and distributed, difficult to abstract and simplify for classroom teaching. The main problem that teachers of this subject have to face, is to be able to integrate the whole set of related techniques and algorithms in a practical example that is easy to understand and address within the framework of the planning of a course.

This paper deals with the use of the "learning by doing" methodology in a subject of multi-agent systems in the Master's Degree in Artificial Intelligence at the Universitat Politècnica de València. This methodology is applied by avoiding master classes to focus on practice. The classes become a scientific-technological experience. The students and the teacher are a team working with a common purpose, seeking to achieve a goal.

To do this, the whole course has been reformulated, proposing the students to solve different typical problems of the MAS area on the same domain, in this case the improvement of urban mobility and the efficient use of energy in the cities. It is considered to be a sufficiently current topic that can motivate the student to participate and propose solutions.

To achieve this objective, a multi-agent system tool has been developed that allows students to simulate the different situations proposed and develop solutions. The tool provides them with an urban simulation environment where they can easily introduce their own strategies to be carried out by each simulation agent. In this way, students are proposed different challenges where they can develop negotiation strategies to simulate the operation of urban taxi fleets, and cooperation strategies, where different agents help each other to achieve a common goal.

This tool, called SimFleet, has been developed in an open way and published as open source, so that it can be used by any teaching team that wishes to do so, and even receive external contributions and improvements thanks to its open character.

This learning by doing methodology supported by the SimFleet simulation tool has been applied in two consecutive academic years obtaining better results in student assessment and learning than in previous courses. Furthermore, the results of the student satisfaction surveys have shown a notable increase when using these technologies, which reinforces the idea that this type of learning is more useful and more satisfactory for students.

Keywords:

Artificial intelligence, multi-agent system, learning by doing, simulation.