Spatial Navigation Torus

Spatial navigation torus is the torus-shaped population-activity geometry discussed in Claire Isabel Webb & Nina Miolane: The Geometry of Consciousness. Nina Miolane uses it as the talk’s main worked example: neurons involved in spatial encoding can produce activity that looks like a donut-shaped manifold when projected from high-dimensional firing rates into a lower-dimensional geometry.

Key Claims

• Each point on the torus corresponds to a two-dimensional location of the animal or agent.
• The torus appears because relevant navigation neurons have periodic firing patterns rather than simple one-location codes.
• Similar torus structures have been observed in biological systems and in artificial networks trained to predict position from self-motion cues.
• The torus can deform around reward or salient locations, giving more representational resolution where it matters for the task.
• Adding social complexity, such as another agent, breaks the simple account and remains an open research area.

Connections

• Population Coding and Neural Geometry - representational method that exposes the torus.
• Fourier Spatial Encoding - candidate explanation for why periodic spatial codes are efficient.
• Mathematical Theory Of Intelligence - theory-building role of the torus example.
• World Models and Multimodal Intelligence - adjacent AI themes involving spatial representation, prediction, and action.
• Consciousness Measurement - related sleep-state and replay examples in the same source.