Navigation and generalisation in a four-rooms environment

Humans performed a computerized task. They learned to navigate a 3D maze by pressing buttons to obtain rewards that were hidden in boxes. Mirroring the classic problem in AI research the rooms were arranged into a square with connectors linking adjacent boxes. Participants were required to forage for two rewards within a fixed time period. They viewed a context cue prior to each trial onset which indicated that rewards were either to be found in two horizontally arranged (cue A) or vertically arranged (cue B) rooms. This allowed them to infer the location of the second reward after the first had been located. Participants were not informed of this contingency but required to learn it by trial-and-error before generalising to a new environment.