Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

BrainSLAM: SLAM on Neural Population Activity Data

Kipp Freud, Nathan Lepora, Matt W. Jones, Cian O'Donnell

TL;DR

BrainSLAM tackles the challenge of inferring internal cognitive maps from neural activity by decoding velocity and location signals from wide-band $LFP$ data (across HPC, PFC, and PC) using a CNN on Morlet-wavelet scalograms, and feeding these signals into a RatSLAM-inspired system with a $3$-D pose cell network, view cells, and an experience map. The approach yields high-accuracy neural decoding (location MAEs around $1.6$–$2.2 ext{ cm}$; direction MAEs around $6^ ext{o}$–$12^ ext{o}$) and produces faithful, loop-closed maps of the environment from only $LFP$ input, demonstrated in three rats navigating a $1.3 ext{ m} imes 1.7 ext{ m}$ maze. Key contributions include an end-to-end pipeline that (i) decodes odometry and loop-closure cues from neural data without spike sorting, (ii) integrates these signals in a RatSLAM framework with a $3$-D pose cell network and loop-closure mechanisms, and (iii) visualizes cognitive maps inferred solely from brain activity, revealing a new modality for mapping environments and probing spatial cognition. The work advances bio-inspired navigation and BCI research by showing that neural population activity can support robust SLAM and environment representation. The findings imply broader significance for understanding cognitive maps and their role in navigation and decision making, and suggest avenues for semantic-space mapping and cross-environment generalization.

Abstract

Simultaneous localisation and mapping (SLAM) algorithms are commonly used in robotic systems for learning maps of novel environments. Brains also appear to learn maps, but the mechanisms are not known and it is unclear how to infer these maps from neural activity data. We present BrainSLAM; a method for performing SLAM using only population activity (local field potential, LFP) data simultaneously recorded from three brain regions in rats: hippocampus, prefrontal cortex, and parietal cortex. This system uses a convolutional neural network (CNN) to decode velocity and familiarity information from wavelet scalograms of neural local field potential data recorded from rats as they navigate a 2D maze. The CNN's output drives a RatSLAM-inspired architecture, powering an attractor network which performs path integration plus a separate system which performs `loop closure' (detecting previously visited locations and correcting map aliasing errors). Together, these three components can construct faithful representations of the environment while simultaneously tracking the animal's location. This is the first demonstration of inference of a spatial map from brain recordings. Our findings expand SLAM to a new modality, enabling a new method of mapping environments and facilitating a better understanding of the role of cognitive maps in navigation and decision making.

BrainSLAM: SLAM on Neural Population Activity Data

TL;DR

BrainSLAM tackles the challenge of inferring internal cognitive maps from neural activity by decoding velocity and location signals from wide-band data (across HPC, PFC, and PC) using a CNN on Morlet-wavelet scalograms, and feeding these signals into a RatSLAM-inspired system with a -D pose cell network, view cells, and an experience map. The approach yields high-accuracy neural decoding (location MAEs around –; direction MAEs around –) and produces faithful, loop-closed maps of the environment from only input, demonstrated in three rats navigating a maze. Key contributions include an end-to-end pipeline that (i) decodes odometry and loop-closure cues from neural data without spike sorting, (ii) integrates these signals in a RatSLAM framework with a -D pose cell network and loop-closure mechanisms, and (iii) visualizes cognitive maps inferred solely from brain activity, revealing a new modality for mapping environments and probing spatial cognition. The work advances bio-inspired navigation and BCI research by showing that neural population activity can support robust SLAM and environment representation. The findings imply broader significance for understanding cognitive maps and their role in navigation and decision making, and suggest avenues for semantic-space mapping and cross-environment generalization.

Abstract

Simultaneous localisation and mapping (SLAM) algorithms are commonly used in robotic systems for learning maps of novel environments. Brains also appear to learn maps, but the mechanisms are not known and it is unclear how to infer these maps from neural activity data. We present BrainSLAM; a method for performing SLAM using only population activity (local field potential, LFP) data simultaneously recorded from three brain regions in rats: hippocampus, prefrontal cortex, and parietal cortex. This system uses a convolutional neural network (CNN) to decode velocity and familiarity information from wavelet scalograms of neural local field potential data recorded from rats as they navigate a 2D maze. The CNN's output drives a RatSLAM-inspired architecture, powering an attractor network which performs path integration plus a separate system which performs `loop closure' (detecting previously visited locations and correcting map aliasing errors). Together, these three components can construct faithful representations of the environment while simultaneously tracking the animal's location. This is the first demonstration of inference of a spatial map from brain recordings. Our findings expand SLAM to a new modality, enabling a new method of mapping environments and facilitating a better understanding of the role of cognitive maps in navigation and decision making.
Paper Structure (14 sections, 4 equations, 3 figures, 1 table)

This paper contains 14 sections, 4 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methods
  3. Dataset & Data Preprocessing
  4. Deep neural decoding for odometry and loop closure modules
  5. RatSLAM
  6. Pose Cells
  7. View Cells
  8. Odometry
  9. Experience Map
  10. Results
  11. Neural Decoding Results
  12. SLAM Results
  13. Discussion & Further work
  14. Conclusion

Figures (3)

  • Figure 1: In the research presented here, local field potential data collected from the dorsal CA1 of hippocampus (HPC), prefrontal cortex (PFC), and the parietal cortex (PC) of rodents as they completed a simple behavioural task within a maze is transformed into wavelet decomposition images (a). This data is used to train a deep neural decoding system with 13 downsampling layers and separate fully connected layers for each output; this network predicts both odometry values (speed and direction) and location (b). These predictions are used to shift activation within a competitive 3D attractor network, powering a RatSLAM architecture which produces graphical representations of the environment being explored (c). Panel c adapted with permission from milford2010persistent
  • Figure 2: Angular histograms showing the distributions of estimated directions for each true direction value for one rat; (a) North, (b) East, (c) South, and (d) West. We see that the system is able to decode all directions with good accuracy. We note that perfect alignment with cardinal angles is not expected due to imperfections in the maze shape.
  • Figure 3: The true shape of the environment being explored by the rats (a), and inferred cognitive maps from each of the three rats as generated by the system presented here from $\sim 6$ minutes of test data (b,c,d). The shape and scale of the inferred maps was accurate in all cases. A photograph of the true maze is shown in Figure \ref{['fig:bigfig']} for comparison.