1)The agent encounters a dead-end in the maze 2)The agent runs into a node it has already visited 3)The agent reaches the destination node (the algorithm is complete) When one of the first two cases occurs, the tank performs the same task. It returns to the previous node in the list of visited nodes and paints the edge red along the way. This way, it knows to never return to that node and the learner can see that the agent now knows that path cannot lead to (at least not redundantly) the path for which it is searching. This process repeats until the third and always-final case is reaches, in which the tank agent knows the final path and exits the maze at the destination node victoriously.