Humans make efficient plans during everydaynavigation and natural spatial search, while these tasks still remainchallenging for algorithms. Which mental computationalmodels do we have that makes this possible? We investigatethree computational principles that may be leveraged by people— approximate expected utility maximization, discountedutility, and probability weighed utility—in the context of a novelspatial Maze Search Task. These computational principles arewell studied in classic bandit tasks and monetary gambles, butthey have not been evaluated on naturalistic spatial tasks thatinvolve sequential decision making. We found that accountingfor a combined effect of these three principles explains aggregatehuman behavior better than models that include justone, or two of these principles, or any of the four behavioralheuristics. We also found substantial individual differences,revealing that humans are best explained by a diversity ofplanning strategies rather than a single best model. Our resultstake a step toward uncovering common computational qualitiesof human spatial planning that may generalize to naturalhuman behaviors in daily life.