About Our Interactive Demo

This online interactive demo allows you to create arbitrary multi-agent pathfinding (MAPF) instance on square grid worlds, by placing obstacles, agents (robots), and corresponding goals, before running a trained PRIMAL model in real-time to see agents plan individual collision-free paths to their goal. The full PRIMAL paper is available at this link, while the full training and testing code is available on github. More information and resources about MAPF can be found at the main MAPF community website.

Important disclaimer: PRIMAL does not provide any completeness/optimality guarantees, so you may very well find cases that the trained model cannot complete. The MARMot Laboratory at NUS is still actively building upon/extending it, see this paper and this other one for recent progress along this line of work. We will also be continuing to improve this interface and add functionalities (e.g., non-square grids, other models such as PRIMAL2/SCRIMP, etc.).

Start by reading the instructions by clicking on the “How to play” button, and have fun (stress-)testing PRIMAL!