We propose that spatial inferences made during planning and executing a route influence the learning of relative locations through wayfinding. In Experiment 1, separate and combined route plans were compared. The results suggest that inferring multiple directions during the initial stage of planning leads to more accurate representations of relative locations than planning a single route. In Experiment 2, regular and irregular updating modes during the execution phase were compared. The results suggest that irregular updating, which involves multidirectional self-to-object updating, also leads to more accurate representations than regular updating. We conclude that the requirement to make spatial inferences about multiple multidirectional metric interconnections in egocentric reference frames during wayfinding facilitates spatial learning.