The task is to put down a wild fire by controlling a helicopter.
The fire is simulated by a Forest Fire Automaton [Drossel and Schwabl (1992)].
At each time step the helicopter moves to an adjacent cell and if it is a fire it changes it to a empty cell.
The reward is
The system has no termination state.
The actions are the natural numbers from 0 to 8, each representing a direction: 0. Left-Up
- Up
- Right-Up
- Right
- Don't move
- Left
- Left-Down
- Down
- Right-Down
Forest Fire Automaton Drossel and Schwabl (1992)
Three type of cells: TREE, EMPTY and FIRE At each time step and for each cell apply the following rules (order does not matter).
- Lighting Rule With probability f:
- TREE turns into FIRE
- Propagation Rule If at least one neighbour is FIRE:
- TREE turns into FIRE
- Burning Rule Unconditional:
- FIRE turns into EMPTY
- Growth Rule With probability p:
- EMPTY turns into TREE