Robust self-localisation and navigation based on hippocampal place cells

A computational model of the hippocampal function in spatial learning is presented. A spatial representation is incrementally acquired during exploration. Visual and self-motion information is fed into a network of rate-coded neurons. A consistent and stable place code emerges by unsupervised Hebbian learning between place- and head direction cells. Based on this representation, goal-oriented navigation is learnt by applying a reward-based learning mechanism between the hippocampus and nucleus accumbens. The model, validated on a real and simulated robot, successfully localises itself by recalibrating its path integrator using visual input. A navigation map is learnt after about 20 trials, comparable to rats in the water maze. In contrast to previous works, this system processes realistic visual input. No compass is needed for localisation and the reward-based learning mechanism extends discrete navigation models to continuous space. The model reproduces experimental findings and suggests several neurophysiological and behavioural predictions in the rat. (c) 2005 Elsevier Ltd.

A connectionist model of spatial learning in the rat

Thomas Strösslin

When animals explore an environment, they store useful spatial information in their brains. In subsequent visits, they can recall this information and thus avoid dangerous places or find again a food location. This ability, which may be crucial for the animal's survival, is termed "spatial learning". In the late 1940s, theoretical considerations have led researchers to the conclusion that rats establish a "cognitive map" of their environment. This spatial representation can then be used by the animal in order to navigate towards a rewarding location. In 1971, researchers have for the first time found direct evidence that the hippocampus, a brain area in the limbic system, may contain such a cognitive map. The activity of neurons in the hippocampus of rats tends to be highly correlated with the animal's position within the environment. These "place cells" have since been the target of a large body of research. Apart from spatial learning, the hippocampus seems to be involved in a more general type of learning, namely in the formation of so-called "episodic memories". Models of hippocampal function could thus provide valuable insights for the understanding of memory processes in general. Insights from animal navigation could also prove beneficial for the design of autonomous mobile robots. constructing a consistent map of the environment from experience, and using it for solving navigation problems are difficult tasks. Incorporating principles borrowed from animal navigation may help building more robust and autonomous robots. The main objective of this thesis is to develop a neural network model of spatial learning in the rat. The system should be capable of learning how to navigate to a hidden reward location based on realistic sensory input. The system is validated on a mobile robot. Our model consists of several interconnected brain regions, each represented by a population of neurons. The model closely follows experimental results on functional, anatomical and neurophysiological properties of these regions. One population, for instance, models the hippocampal place cells. A head-direction system closely interacts with the place cells and endows the robot with a sense of direction. A population of motor-related cells codes for the direction of the next movement. Associations are learnt between place cells and motor cells in order to navigate towards a goal location. This study allows us to make experimental predictions on functional and neurophysiological properties of the modelled brain regions. In our validation experiments, the robot successfully establishes a spatial representation. The robot can localise itself in the environment and quickly learns to navigate to the hidden goal location.

EPFL2004

Slowness and Sparseness Lead to Place, Head-Direction, and Spatial-View Cells

We present a model for the self-organized formation of place cells, head-direction cells, and spatial-view cells in the hippocampal formation based on unsupervised learning on quasi-natural visual stimuli. The model comprises a hierarchy of Slow Feature Analysis (SFA) nodes, which were recently shown to reproduce many properties of complex cells in the early visual system [1]. The system extracts a distributed grid-like representation of position and orientation, which is transcoded into a localized place-field, head-direction, or view representation, by sparse coding. The type of cells that develops depends solely on the relevant input statistics, i.e., the movement pattern of the simulated animal. The numerical simulations are complemented by a mathematical analysis that allows us to accurately predict the output of the top SFA layer.

Public Library of Science2007

Is there a geometric module for spatial orientation? Insights from a rodent navigation model

Angelo Arleo, Ricardo Andres Chavarriaga Lozano, Wulfram Gerstner, Thomas Strösslin

Modern psychological theories of spatial cognition postulate the existence of a geometric module for reorientation. This concept is derived from experimental data showing that in rectangular arenas with distinct landmarks in the corners, disoriented rats often make diagonal errors, suggesting their preference for the geometric (arena shape) over the nongeometric (landmarks) cues. Moreover, sensitivity of hippocampal cell firing to changes in the environment layout was taken in support of the geometric module hypothesis. Using a computational model of rat navigation, the authors proposed and tested the alternative hypothesis that the influence of spatial geometry on both behavioral and neuronal levels can be explained by the properties of visual features that constitute local views of the environment. Their modeling results suggest that the pattern of diagonal errors observed in reorientation tasks can be understood by the analysis of sensory information processing that underlies the navigation strategy employed to solve the task. In particular, 2 navigation strategies were considered: (a) a place-based locale strategy that relies on a model of grid and place cells and (b) a stimulus–response taxon strategy that involves direct association of local views with action choices. The authors showed that the application of the 2 strategies in the reorientation tasks results in different patterns of diagonal errors, consistent with behavioral data. These results argue against the geometric module hypothesis by providing a simpler and biologically more plausible explanation for the related experimental data. Moreover, the same model also describes behavioral results in different types of water-maze tasks. (PsycINFO Database Record (c) 2009 APA, all rights reserved)