Abstract
Navigation in the natural world is a challenging problem that engages many cognitive systems, including cognitive maps, attention, motor control, and planning. However, most fMRI navigation studies use highly simplified environments and tasks that are unlikely to engage all navigational processes. Thus, they cannot create detailed functional cortical maps of the many different types of information that are likely relevant for natural navigation.
To recover detailed cortical maps of navigation-related information, we used fMRI to record whole-brain activity while subjects performed a taxi driver task in virtual reality. The pilot environment is a 1×1 km town without other agents. The main environment is a 1×2 mile city with traffic, pedestrians, and various neighborhoods and off-road areas. Subjects drove using an MR-compatible steering wheel and pedals constructed in our lab. One subject participated in the pilot for 130 minutes and the main environment for 260 minutes. A second subject participated in the pilot for 90 minutes.
We applied the voxelwise modeling framework to the data. We extracted stimulus and task features from the experiment, and used banded ridge regression to find optimal weights for each feature for every voxel in each subject. We evaluated 16 feature spaces that captures various aspects of navigation. We used a separate dataset to test statistical significance and generalization of models in each subject.
The recovered cortical maps show the PPA, RSC, and OPA represent information about roads, buildings, and boundaries. RSC and precuneus tracks route progression. Visual motion-energy is represented across visual cortex, including the posterior parts of RSC, OPA, and PPA. FFA and EBA represent information about pedestrians and other vehicles. These results show that naturalistic navigation elicits rich cortical activity and navigation information is represented in distributed networks of brain regions.