How Grid Cells Map Space in the Entorhinal Cortex | Moser Research

The hippocampus contains multiple types of spatial cells working together to create a complete cognitive map: (1) Place cells identify specific locations, (2) Head direction cells indicate the animal's orientation relative to the environment, (3) Grid cells create a metric system with evenly spaced firing fields across the environment, and (4) Boundary cells detect proximity to environmental boundaries or objects. Together, these cells provide all necessary information for spatial navigation.

The hippocampal formation contains multiple specialized spatial cell types working together: (1) Place cells identify specific locations; (2) Head direction cells track orientation using an internal compass; (3) Grid cells provide coordinate systems with hexagonal firing patterns covering the entire environment; (4) Boundary vector cells encode distances to environmental boundaries; (5) Object vector cells encode distances to specific objects. These cells collectively provide the information necessary to construct a complete cognitive map of the environment.

The cognitive map theory predicted specific neural mechanisms that were subsequently discovered: (1) Place cells, identified by O'Keefe, fire when an animal occupies specific locations; (2) Head direction cells, discovered in the 1980s, provide directional information regardless of position; (3) Grid cells, discovered by Moser and colleagues, create a coordinate system for spatial positioning. These cell types work together to form a complete system for spatial representation. The discovery of these neural correlates validated the theoretical predictions and provided the physiological basis for understanding how cognitive maps are constructed and utilized in the brain.

The hippocampus contains specialized neurons that collectively construct a complete cognitive map of space. Place cells fire when animals occupy specific locations, creating a distributed representation of the environment. However, a complete map requires encoding spatial relationships between locations through vectors representing distance and direction. Additional cell types contribute: head direction cells provide orientation information functioning as an internal compass, and boundary-distance cells encode proximity to environmental edges. Grid cells in the entorhinal cortex fire at multiple locations arranged in hexagonal patterns, providing a metric for measuring distance and direction. Critically, these patterns depend on environmental geometry - beautiful symmetrical grids emerge only in symmetrical environments. In asymmetrical environments, grid patterns become distorted, demonstrating that spatial computation integrates both geometric constraints and self-motion information. Together, these cell types construct a flexible system capable of calculating optimal routes and generating alternatives when paths are blocked.

The brain encodes spatial information through specialized cell types including place cells (which fire at specific locations), grid cells (which create hexagonal spatial maps), head direction cells, border cells, and object vector cells, all working together in the hippocampus and entorhinal cortex to form an internal cognitive map; these spatial coding mechanisms operate on a toroidal (donut-shaped) manifold structure that persists even during sleep, while time encoding in the lateral entorhinal cortex shows progressive drift and sequence generation, both contributing to episodic memory formation and being implicated in Alzheimer's disease pathophysiology.

Scientists at University College London are developing virtual reality navigation tests to help diagnose Alzheimer's disease earlier by assessing spatial memory and navigation abilities, which are affected by the disease before severe dementia symptoms appear; this cognitive test complements biological markers like blood tests and offers a culturally neutral assessment method that could identify the disease years before traditional diagnostic methods.

Virtual reality technology enables researchers to test spatial orientation skills and simultaneously record brain activity, revealing that early changes in the brain's internal GPS system (including reduced grid cell signals and tau protein deposits) occur years before clinical Alzheimer's diagnosis, making VR-based spatial navigation tests potentially more sensitive for early detection than traditional memory assessments.

Spatial disorientation represents one of the earliest symptoms of Alzheimer's because the disease initially affects brain regions containing cells responsible for spatial navigation. Scientists developed Sea Hero Quest, a gaming app that measures population-level navigation behavior using immersive virtual reality. With over 4.5 million downloads across 63 countries, this approach revealed that age, sex, and nationality strongly influence individual navigation skills. Understanding these demographic factors is crucial for accurately diagnosing Alzheimer's-related navigation deficits.

Spatial navigation, particularly path integration—the ability to monitor self-motion cues to update one's position—serves as a sensitive behavioral readout for detecting early stages of Alzheimer's disease, as this cognitive function relies on the entorhinal cortex grid cell circuit which is among the first brain regions affected by Alzheimer's pathology; research demonstrates that path integration deficits appear in subjective cognitive decline patients (who report memory concerns but show normal cognition on standard tests) and MCI patients, with performance correlating with amyloid pathology levels and outperforming standard cognitive tests in diagnostic accuracy.

Spatial navigation difficulties are among the first clinical signs of Alzheimer's disease, appearing before memory loss occurs; researchers have developed a global navigation database using a video game with over 4 million participants to study how age, gender, and culture influence navigation patterns, enabling personalized assessment of preclinical Alzheimer's patients by comparing their orientation abilities with matched controls to distinguish normal navigation from pathological signs, potentially allowing earlier diagnosis and intervention.

Navigation fundamentally requires answering 'Where am I?' (localization) and 'How do I get from here to there?' (path planning). While SLAM algorithms enable basic robot navigation, researchers seek how navigation abilities emerge in learning systems. Three desiderata for navigating agents include: communication using symbolic representations grounded in visual observations, adaptation to new environments, and learning from experience. Reinforcement learning enables agents to acquire observations and take actions with extrinsic (goal-reaching) and intrinsic (exploration) rewards. Auxiliary prediction tasks improve learning efficiency but agents initially only learned procedural skills in static environments. Neuroscience-inspired approaches leverage place cells (hippocampus, encode position), head direction cells (encode orientation), and grid cells (entorhinal cortex, encode coordinate system with periodic triangular firing patterns). Grid cells serve as spatial basis functions summing to unique positions, enabling artificial agents to learn grid cell-like representations through path integration and classification.

RatSLAM is a robotic navigation algorithm inspired by grid cells in the brain, which uses continuous attractor networks connected to visual input and self-motion cues to help robots simultaneously localize themselves and build maps of unknown environments, addressing the classic SLAM (Simultaneous Localization and Mapping) problem where robots must create environmental maps while determining their own position within them.

RatSLAM combines pose cells (encoding both location and orientation) with experience maps (graphical readouts of neural representations). Experience maps enable path planning, error distribution, and memory management. The system achieved over 1,000 navigation trials across two buildings over 1.5 weeks, autonomously recharging and resuming after being moved. Grid cells represent hundreds or thousands of locations at multiple scales, creating overlapping maps. Some grid cells have conjunctive pose properties. This multi-scale representation enables efficient spatial encoding and was discovered in 2004-2005, earning the 2014 Nobel Prize. The journey from fundamental neuroscience to capable SLAM technology took over a decade, demonstrating that biological inspiration provides principles but practical deployment requires combining bio-inspired approaches with conventional algorithms.

This section connects neuroscience discoveries to robotics. The 2014 Nobel Prize went to O'Keefe and the Mosers for discovering place cells (specific location firing) and grid cells (periodic coordinate system firing) in the entorhinal cortex. These cells form the brain's 'GPS system' for local positioning and are critical for forming long-term memories. The speaker explains SLAM as a cornerstone of autonomy, where robots generate maps from noisy local data while estimating position. Loop closing corrects accumulated drift errors. The speaker advocates for dense 3D mapping using GPU-accelerated volumetric representations (TSDF) and pose graph optimization, demonstrating how biological systems achieve representational harmony that robots struggle to replicate.

John O'Keefe discovered place cells in 1970-71—hippocampal neurons that fire when an animal occupies specific locations in an environment, creating a cognitive map. More recently, Edvard Moser and colleagues discovered grid cells in the entorhinal cortex, which fire in hexagonal patterns across space, providing an intrinsic coordinate system for navigation. These findings have inspired new navigation algorithms, including DARPA-funded projects, demonstrating how understanding neural mechanisms can directly translate into engineering applications.