TL;DR This tutorial demonstrates how to simulate a navigating agent (RatInABox), collect its visual data (Blender), build a Recurrent Neural Network (RNN) to model hippocampal-like spatial representations (Pytorch), and analyze its latent representations.
INTRODUCTION The intricate interplay between sensory perception and the "cognitive map" of space (O'Keefe & Nadel, 1979) allows animals to navigate complex environments, interact with them in purposeful ways, and adapt to new situations. In the hippocampal formation, various classes of spatially modulated neurons support navigation by integrating sensory inputs to construct flexible internal models of the world (Behrens et al., 2018). Vision, in particular, plays a crucial role in guiding movement and shaping neural representations of space. However, since time is linear and irreversible, animals must rely on sequential observations to infer environmental structure. These experiences are then transformed into internal models that capture the relationships between elements in the world (Buzsáki & Tingley, 2018; Stachenfeld et al., 2017).
DESCRIPTION This tutorial introduces computational methods for studying these processes. First, we demonstrate how to simulate the trajectories of virtual agents resembling animals (such as rodents and ferrets) with RatInABox and how to automatically collect visual information while agents navigate a highly customisable virtual environment using the Blender Python API. Next, we explore how shallow RNNs can be trained (using Pytroch) to develop biologically-plausible allocentric tuning curves in their latent space representations, resembling the activity of spatially selective neurons in the hippocampal formation. These models enable us to predict how specific experiences or stimuli may facilitate or hinder the emergence of spatial neurons in the hippocampus, for example in the context of development. Moreover, they can serve as surrogate models for evaluating navigational performance under different sensory conditions or targeted neuronal lesions.
