Long-range navigation by path integration and decoding of grid cells in a neural network

Abstract

Abstract-Neural modelers in the domain of robot navigation, e.g. within the fields of neurorobotics and neuromorphic engineering, can benefit from a wealth of inspiration from neuroscientific research in the hippocampal formation-cell types such as place cells and grid cells provide a window into the inner workings of high-level cognitive processing, and have spawned many interesting computational models. Grid cells are thought to participate in path integration and to implement a general coordinate system, both of which are useful features in a neural navigation model. Continuous attractor networks are a computational model that can embody both aspects of grid cells, and in previous work we showed that a neural network can successfully decode the outputs of such networks in order to implement vector navigation. That work assumes that the grid cell system represents long distances by employing a geometric progression in its spatial scaling of successive submodules, in such a way that "nested" grid cell decoding can be performed. For long-range navigation this requires that the continuous attractor networks can implement sufficiently long geometric progressions of grid scales, but this turns out to trigger the issue of "pinning". In this paper we demonstrate conditions under which pinning occurs as well as its consequences for the grid cell-based navigation model. We propose and assess several candidate solutions to the problem, in particular based on differential adjustment of neurons' update rates in the model. We finally demonstrate that the system is able to perform long-range navigation using our chosen solution.

Navigating with grid and place cells in cluttered environments

“…Note that the model of combined vector and topological navigation proposed below is indifferent to the particular workings of the grid cell-decoding mechanism, or indeed to the origin of the grid cell signal itself-we assume only that vector navigation can be performed through the readout of grid cells. Our specific implementation used here builds on the implementation from Edvardsen (2017), where grid cell decoding is performed according to a "nested" view of the grid cell system (Stemmler, Mathis, & Herz, 2015). Feedforward decoder neurons receive inputs from the two grid cell populations and are preconfigured to detect specific patterns of directional offset between the two inputs.…”

An On-chip Spiking Neural Network for Estimation of the Head Pose of the iCub Robot

“…Loop closure detection was realized based on visual template matching (Gu and Yan, 2019), and multi-sensor fusion was shown to provide more accurate odometry and precise cognitive mapping (Zhang et al, 2019). Neural networks of grid cells have been shown to perform long-range navigation through path integration in the 2-dimensional plane (Edvardsen, 2017), and a model that was established through the Neural Engineering Framework confirms the attractor map implementation of path integration and proposes that the head direction signal can be used to modulate allocentric velocity input (Conklin and Eliasmith, 2005).…”