We conducted statistical analyses on the agents’ behavior in the same way as for humans, bearing in mind that the sources of intrinsic variability were different (relating exclusively to the random initial seed and the structure of the map experienced) and that the larger quantity of data rendered our results highly significant.
← all excerpts
Goal-directed navigation in humans and deep reinforcement learning agents relies on an adaptive mix of vector-based and transition-based strategies.
1
—
—