Figure 7: Cleanup, a Sequential Social Dilemma Game from Vinitsky et al. [2019]

Neural Information Processing Systems 

Agent 2 (a) The initial setup with two agents and two river (b) The result of both agents perform the "clean" tiles. When the river tiles become dirty, they are action. Both river tiles can be are cleaned since shown as a brownish color instead. Agent 1's action is resolved first. Agent 2 (a) If there are no dirty river tiles in the path of the (b) If there is a dirty river tile in the path of a cleaning beams, the beams will extend to the full beam, the beam will stop at the tile, changing it to length of five tiles. Figure 8: An example of Agent 1 using the "clean" action while facing East.