Practice fixed the measuring tasks, not the mental map itself
The clearest gains came from the simplest ingredient: repeated attempts with feedback. Participants walked one route per week for six weeks and then did tasks testing what they knew about it. With feedback alone, their straight-line distance estimates and hand-drawn sketch maps became more accurate by the sixth session, reaching a level comparable to people with an average sense of direction [1]. That is a real improvement in how well they could report the layout of a route they had walked.
A second group got the same feedback plus extra practice at allocentric spatial updating, meaning practice at keeping track of locations relative to the environment rather than relative to one's own body. This improved the accuracy of direction estimates, but the effect was limited: it was not large enough to bring the poor-sense-of-direction group up to the average group's performance [1]. So the extra training helped a specific judgment without erasing the underlying difference between the groups.
Other approaches aim at the same problem from the outside
Training is not the only route being explored. A mixed-reality system called LandmarkLens was designed to guide people with a poor sense of direction toward the landmarks that skilled navigators rely on, and a follow-up study reported performance improvements from this guided landmark attention [4]. That is a tool that scaffolds navigation in the moment rather than a course that rebuilds the underlying skill, and the work does not fully examine whether the benefit carries over to other tasks [4].
Work with blind users points in a similar direction: exploring multimodal maps at different orientations was studied for its effect on cognitive maps and sense of direction [2]. The available summary does not address whether any such effect transfers to other tasks or contexts [2], so it cannot yet be counted as evidence for or against transfer. Animal work offers a looser analogy: goldfish trained in allocentric and egocentric maze procedures learned the tasks accurately within a week, and later transfer tests probed what they had learned [3]. That is a different species and a different question, so it informs the general idea that spatial training can be tested for transfer without settling the human case.
What this means if you are the one who gets lost
The practical picture is modest but not discouraging. Deliberate practice with feedback on real routes appears to sharpen the specific judgments you rehearse, such as estimating distances and sketching layouts [1]. Adding practice at tracking your position relative to the environment may improve direction estimates a little [1]. What the evidence does not support is the expectation that a few weeks of training will make you navigate like someone with a naturally good sense of direction, because the flexible switching between viewpoints that navigation demands proved hard to train [1].
The comparison group matters for reading these results: the study included 40 adults with a poor sense of direction and 20 with an average sense of direction, and the feedback gains reached the average group's level on distance and sketch tasks but the updating gains did not [1]. So the honest answer to whether training transfers is: to closely related judgments, yes; to the core perspective-shifting skill, not reliably in this evidence.
Sources used in this answer
Improving cognitive mapping by training for people with a poor sense of direction.
In 40 adults with a poor sense of direction, six weeks of route practice with feedback improved straight-line distance estimates and sketch maps to the level of an average-sense-of-direction comparison group, while added allocentric spatial updating practice improved direction estimates only modestly and did not close the group gap.
Investigating the effect of exploring multimodal maps at different orientations on blind user's cognitive maps and sense of directions
The study examined how exploring multimodal maps at different orientations affects blind users' cognitive maps and sense of direction, but did not address whether effects transfer to other tasks or contexts.
Performance of goldfish trained in allocentric and egocentric maze procedures suggests the presence of a cognitive mapping system in fishes
Goldfish trained in allocentric and egocentric maze procedures learned the tasks with high accuracy after one week, and subsequent transfer tests probed what they had learned.
LandmarkLens: Predicting and Presenting Effective Landmarks for Mixed-Reality Urban Exploration
A mixed-reality system guiding people with a poor sense of direction toward landmarks used by skilled navigators was associated with performance improvements in a follow-up study, though transfer effects were not fully examined.