AI, Pathfinding and Multi Agent System

For the last assignment of the AI course, in my third year in ESAT (Valencia), we were encouraged to develop a Multi Agent System to simulate a scenario with some phases: assault, rescue and scape, showing the result in a visual SDL environment.

To carry out this project, we had to use our own pathfinding A Star algorithm. The design and programing of this A* algorithm was done in the first assessment phase of the AI course, under the tutelage of Toni Barella.

I have to say that I still very satisfied with the results of my pathfinder code. I have refactored and used this algorithm in different projects with different aims and I still enjoying with the scalability and possibilities of this code.

This first file, it's the visual result of my AI pathfinder. The visual test made with SDL was able to read from a pixel map with colour codes and generate a maze from it. In case it cannot find a path, the console shows the result. If a way is possible, a spider will run the maze through the optimal path and the algorithm will draw a cobweb in every scanned area.

A* Spider Demo

You have only to execute the .exe file in the bin folder and you can also replace the "borderMapAI.png" for another from the maps folder renaming the new file file with the same name ("borderMapAI.png").

This second demo show the visual result of the Multi Agent System. Attacks are disabled. The phases are: assault, rescue and scape. The prisoners are working until the intruder alarm is activated. The guards switch their state to alarmed after one of them watch the intruders and active the base alarm.

It's important to say that the 30 agents are processing their own paths for free in real time and the performance is excellent. Some of the agents uses a control point system to reach a better computation performance. There are some grided AI maps under the final map which are used to recognise the transitable areas and reduce the number of nodes to scan.

Multi Agent System Demo