The contribution gives a micro-structural insight into the pedestrian decision process during an egress situation. A method how to extract the decisions of pedestrians from the trajectories recorded during the experiments is introduced. The underlying Markov decision process is estimated using the finite mixture approximation. Furthermore, the results of this estimation can be used as an input to the optimization of a Markov decision process for one `clever' agent. This agent optimizes his strategy of motion with respect to different reward functions, minimizing the time spent in the room or minimizing the amount of inhaled CO.