In this paper we develop an influence map based agent for Pac-Man 
and analyze its performance compared to a previous implementation. 
The major particularity comes from the fact that the proposed model 
limits the influence area of the components in the game. This 
difference proves to be crucial for the lifetime of the Pac-Man 
agent, as he is more $careful$ when compared to his predecessor. 
A thorough comparison between the two approaches, under similar 
circumstances, reveals that our expectations are met and the 
proposed model is competitive or better when considering several 
outcomes. Experiments also reveal how important rewards and enemy 
agents are and how their influence should be propagated. Overall, 
results are promising, showing that limited influence propagation 
is a valid solution in some cases, managing to solve some of the 
problems of the previous agent.