About

The goal of this project was to create and solve a 2D puzzle inspired by MIT’s modular M-blocks. The puzzle consisted of robots, walls, and a portal. The agents must learn to work together by climbing over each other to pass as many of themselves through the portal.

The problem is solved using deep-QL with hindsight experience replay. Below is an example policy that the agents learn. The red squares are walls and the square highlighted white is the goal portal.