Flatland is a toolkit for developing and comparing multi agent reinforcement learning algorithms on grids.
The base environment is a two-dimensional grid which allows to place many agents. Each agent must solve a task in the grid world.
Agent can navigate from cell to cell. The transitions between neighbour cells can be restricted.
For each cell all possible transition can be separately defined. If no transition restriction is defined, the agent can move fully free at given cell.
The base environment is a two-dimensional grid which allows to place many agents. Each agent must solve one or more tasks in the grid world.
In general agents can freely navigate from cell to cell. But the navigation from cell to another cell can be restricted by transition maps.
Each cell can have a different transition map aligned. If no transition map is defined, the agent can move freely form given cell to it's neighbor cells.
That means the agent can navigate to all eight neighbour cells (go up and left, go up, go up and right, go right, go down and right, go down, go down and left, go left).
The general purpose of the implementation allows to write in very short time many different environments.
