The problem
Imagine a familiar room with a table moved across your usual route to the door. You know the room and the table, but their new arrangement changes the way through. How does the brain combine familiar places and obstacles into a usable map of a new layout?[1]
Consider two familiar barriers brought together to block a corridor. Their joint effect on possible routes differs from the effects of either one alone. How do neural circuits combine stored spatial components and correct for those interactions to plan through a new layout?
A predictive map represents how places can be reached over future steps, rather than only which places are next to one another. In Piray and Daw’s 2025 model, a baseline map of open space is altered by reusable representations of obstacles. This compositionality—building a new whole from familiar parts—provides an initial approximation; learning adjusts it when nearby obstacles interact.[1]
Scope
Neural implementation of composing reusable spatial object and barrier representations into predictive maps for flexible navigation.
What remains open
A computer model can reuse the effects of individual obstacles and then correct for how they interact. We do not yet know how brain circuits perform that combination, especially when several obstacles change the same routes.[1]
The model reproduces selected entorhinal response properties, including activity patterns associated with objects and spatial position. These matches do not show that neurons perform the proposed computation. The authors suggest hippocampal replay—sequences of reactivated spatial-cell activity—as a candidate for learning object interactions; that implementation remains untested. Occlusion and changing, partly observed layouts are further limits.[1]
Why it matters
A map assembled from reusable parts could support a new route before extensive practice.[1]
The corridor example exposes a useful tradeoff: storing every possible layout would waste learning, while simply remembering each barrier separately can give the wrong route. Explaining how the brain reuses knowledge while correcting its combinations would connect spatial memory to flexible planning. Broader connections to composing thoughts are promising proposals, not results established by this navigation model.[1] [2]
Definitions
2 terms
predictive map
An internal representation of how places can be reached over future steps. In the cited model it summarizes reachability under a baseline movement rule; it is not a photograph of the environment.[1]
Sources
- Reconciling flexibility and efficiency: medial entorhinal cortex represents a compositional cognitive map
Notes on Reconciling flexibility and efficiency: medial entorhinal cortex represents a compositional cognitive map
Introduces a computational predictive-map model and explicitly leaves the neural basis of multi-object composition unclear.
- Cognitive maps as a medium for thought
Notes on Cognitive maps as a medium for thought
Opinion article proposes how dynamic map representations could support compositional thinking; it does not establish a complete biological mechanism.