Artificial intelligence can make people's lives easier in many ways. However, how such an artificial intelligence agent makes decisions can be somewhat difficult for the public to imagine. Even for AI developers themselves, it is often difficult to read these decision-making processes solely from the available logs. That is why the PANSim project was created, which visualises the behaviour of AI agents when planning in a non-deterministic environment, i.e. an environment where external factors that the agent cannot influence but must take into account come into play. These "interventions of nature" are one of the main challenges in robotics and automation today. The PANSim project thus helps both laymen and developers themselves to understand the "thought processes" behind artificial intelligence decision-making and planning. Thanks to this visualisation, developers can more easily detect errors and subsequently fine-tune their agents.
Currently, PANSim displays two scenarios: a simulation inspired by the classic game Perestroika and the operation of underwater drones between ships. In both cases, the agent must find a guaranteed safe path to the goal despite obstacles caused by external factors. To find such a path, the agent uses the techniques developed by the team of doc. Lukáš Chrpa at the Department of Industrial Informatics at CIIRC CTU, where a student of FEE CTU began working on the project. Currently, in addition to laymen, the tool is mainly beneficial for researchers working with one of these two scenarios. However, Erol Medenčević intends to expand the project to include other scenarios.
"In the future, I would like PANSim to become domain-independent, i.e. to function as a universal tool in which researchers can simulate any scenario they can imagine," describes Bc. Medenčević. At the same time, however, he admits that this is still a very ambitious plan. In the near future, he will focus more on making the programme interactive so that it becomes attractive to people outside the field and arouses greater public interest in how artificial intelligence works. "Users could try to solve a specific problem themselves in PANSiM and then compare their results with how the computer performed in that situation," outlines Bc. Medenčević.
The project originated as a bachelor's thesis
Erol Medenčević currently works at the Department of Industrial Informatics at CIIRC CTU, specifically in the team of doc. Lukáš Chrpa, where he continues to work on the project during his studies at FEE CTU. Mr Medenčević would like to link the further development of the PANSim tool to his master's thesis, as he did with his bachelor's thesis, in which PANSim was created.
At its core, the project integrates the Unity game engine, which provides smooth graphic simulation, and a significant number of programmes where all calculations and logical planning take place, each of which is completely different and written in a different programming language. "I was intrigued by the opportunity to combine my field of study, graphics, with the field of artificial intelligence. Until then, I hadn't had much experience with planning, so at first I focused mainly on understanding it, while the graphics part took a back seat. The first visualisation was therefore relatively simple, but even so, I liked the fact that, despite its simplicity, it provided much greater insight into the simulation and behaviour of agents than a standard command line output," describes Bc. Medenčević.
The bachelor's thesis was supervised by Jakub Med, with whom Medenčević continues to work closely at PANSiM. "The project is primarily complicated by the fact that it integrates many algorithms from the scientific community that are not yet ready for practical application, as is often the case with scientific prototypes. Erol often faced problems with approaches that were not fully compatible and needed to be properly integrated. Despite this, he proactively came up with ideas on how to improve the tool and where to take it next," says Ing. Med, who is a doctoral student at FEE ČVUT and a researcher at the Department of Industrial Informatics at CIIRC ČVUT.
Participation in the AAAI 2026 conference
At the beginning of the year, between 20 and 27 January 2026, the 40th edition of the prestigious international AAAI conference on artificial intelligence, organised by the Association for the Advancement of Artificial Intelligence, took place in Singapore. It is one of the highest-rated conferences in the field of AI in the world. As part of the demo track, i.e. the practical demonstration section, the PANSim tool was also accepted at the conference, with only 28% of the submissions accepted for this section at this year's conference. "I am extremely happy about this and, to be honest, it was a big surprise for me. It showed me that the project not only fulfilled my study obligations, but that it also has a broader impact and that the work is really useful, which motivates me to continue working on the tool," says Bc. Medenčević.
The response to PANSim during the conference was very positive. There was continuous interest in the stand where the practical presentation of the project took place. "It seemed to me that the project was of interest mainly to people from the feedback learning and planning communities. We exchanged contacts with some of them. We hope that the presentation will lead to some kind of collaboration," comments Ing. Med, who presented the project on site.
Due to the conference coinciding with the exam period, the first author, Mr Medenčević, was unable to attend the conference in person. Even so, he sees the acceptance of the project at the AAAI conference as a great success. "It made me want to learn much more about planning. I enrolled in a course on this topic this semester, so in the future I could deal not only with the graphic side of things, but also with the behaviour of agents themselves," adds Bc. Medenčević.
More information about PANSim, including a scientific article, can be found in the project repository at https://gitlab.com/automated-planning/tools/pansim.
