( Entry by Daniel Furelos, Artificial Inteligence and Machine Learning Research Group (
In this blog entry I will explain my experience as a Maria de Maeztu scholar at the Artificial Intelligence and Machine Learning Group. The project in which I participated was Enhancing usability and dissemination of planning tools.
I joined the AI-ML group in December 2016 while I was doing my master’s degree at Universitat Pompeu Fabra. Since I was an undergraduate, I had always been interested in the stuff that people in this group was producing. Actually, I did my bachelor’s thesis on reinforcement learning with Anders Jonsson, the head of the group. Besides, at that time, I was wondering whether I would like to opt for a research career in the short term. Therefore, when I was told to join the group I did not think twice! It was a huge opportunity to do research in a field that I really liked. In the end, all my work would form my thesis.
For many years, the members of the AI-ML had been developing software described in diverse journal and conference papers. However, some of them required further development (mainly documentation and usability) in order to be adopted by more researchers in the community. More importantly, this software had to be open so that other researchers could reproduce the results in the corresponding papers and make modifications easily to test new ideas.
My work focused on three different topics: multiagent planning, temporal planning, and the application of temporal planning to carpooling. In the case of multiagent planning, we developed a parser capable of reading the standard format for multiagent planning problems. Besides, we also developed a new method for solving multiagent planning problems by compiling them into classical planning problems. We will present this work at ICAPS (International Conference on Automated Planning and Scheduling), specifically during the DMAP (Distributed and Multi-Agent Planning) workshop. You can find the code here: https://github.com/aig-upf/universal-pddl-parser-multiagent/.
In the case of temporal planning, we uploaded existing planning methods developed by the group to GitHub. We also developed a new method for solving temporal planning problems involving simultaneous events. This work will be presented in the COPLAS (Constraint Satisfaction Techniques for Planning and Scheduling) workshop at ICAPS. Furthermore, all the existing temporal planning algorithms developed by the group were combined into a single planning portfolio algorithm that will participate in the next International Planning Competition (IPC). You can find the code of all planners here: https://github.com/aig-upf/temporal-planning.
Finally, we collaborated with Antonio Bucchiarone, a researcher from Fondazione Bruno Kessler (Trento, Italy), to apply temporal planning to carpooling. The resulting work was accepted at AAMAS (International Conference on Autonomous Agents and Multiagent Systems) in the main track as an extended abstract and in the demo track. You can find the code here: https://github.com/aig-upf/smart-carpooling-demo.
All in all, I consider my experience as a research assistant at the AI-ML group very rewarding. I have had the opportunity to be involved in the research process: I discussed ideas with researchers, tested them empirically and submitted to workshops and conferences. If I was in December 2016 again, I would definitely join the group again!