Fig. 2: Example of the constrained grid decision scenario (left part of the figure).
Black squares represent states with penalties. Penalties are also generated when the agent moves top left, bottom left, right, and bottom right (see black squares in top right grid), and when it moves to a blue or a green state (see black squares in bottom right of the figure). The red lines describe a set of trajectories generated by the agent (all with the same start point S0 and end point SG). The strength of the red color for each move corresponds to the number of trajectories employing such a move.
