Human navigation requires to learn and exploit spatial regularities through predictive motor adjustments. Using virtual reality, we investigated how humans learn environmental statistics and accordingly adjust their eye and body movements. Participants navigated through a virtual museum and had to find the shortest path to one of two exit doors while avoiding a suddenly appearing virtual guest blocking either the left or right path. We manipulated the expected reward of a path by varying obstacle frequency in a block-wise manner (high frequency-90% vs. low-frequency-10% path blocking). In control blocks, path blocking occurred equiprobably. We hypothesized that participants learn and capitalize on the statistical regularity in the world (obstacle appearance), and as a result predictively adjust their eye and body movements towards the less frequently blocked path in later trials of a block. We found that only around one-third of the participants adjusted their movements to the spatial regularities of the environment. Many of the other participants walked in a straight line until obstacle appearance, delaying their decision until the last moment. While this approach resulted in greater adjustments upon obstacle appearance compared to scenarios where participants had committed to the correct pathway earlier, it minimized the risk to make even larger movement adjustments if the initially chosen path was blocked. In a follow-up experiment, we eliminated the opportunity to employ this strategy by separating the pathways right at the start, forcing earlier decisions. Now, a majority of participants consistently selected the path presumed to be free, however, roughly one-third still chose the wrong path even in later trials of a block. These findings highlight that navigational behavior does not simply arise from the presence of environmental regularities but reflects an adaptive process of sensorimotor decisions shaped by task demands, environmental structure, individual strategy preferences, and learning capabilities.