A Model for Learning Representations of 3D Objects Through Tactile Exploration: Effects of Object Asymmetries and Landmarks

In this paper, we develop a neural network model that learns representations of 3D objects via tactile exploration. The basic principle is that the hand is considered as an autonomous ‘navigating agent’, traveling within the ‘environment’ of a 3D object. We adapt a model of hippocampal place cells, which learns the structure of a 2D environment by exploiting constraints imposed by the environment’s boundaries on the agent’s movement, and perceptual information about landmarks in the environment. In the current paper, our focus is on 3D analogues of these 2D information sources. We systematically investigate the information about object geometry that is provided by navigation constraints in a simple cuboid, and by tactile landmarks. We find that an asymmetric cuboid conveys more information to the navigator than a symmetric cuboid (i.e., a cube) – and that landmarks convey additional information independently from asymmetry.

[1]  M. Shapiro,et al.  Prospective and Retrospective Memory Coding in the Hippocampus , 2003, Neuron.

[2]  Stephen Grossberg,et al.  Coordinated learning of grid cell and place cell spatial and temporal properties: multiple scales, attention and oscillations , 2013, Philosophical Transactions of the Royal Society B: Biological Sciences.

[3]  B. McNaughton,et al.  Dead Reckoning, Landmark Learning, and the Sense of Direction: A Neurophysiological and Computational Hypothesis , 1991, Journal of Cognitive Neuroscience.

[4]  Bruno B Averbeck,et al.  Representing spatial relationships in posterior parietal cortex: single neurons code object-referenced position. , 2007, Cerebral cortex.

[5]  Hiroshi Ando,et al.  Automatic representation of a visual stimulus relative to a background in the right precuneus , 2015, The European journal of neuroscience.

[6]  T. Kohonen.  Self-organized formation of topographically correct feature maps , 1982 .

[7]  Bruce L. McNaughton,et al.  Path integration and the neural basis of the 'cognitive map' , 2006, Nature Reviews Neuroscience.

[8]  Kate Jeffery,et al.  Retrosplenial and postsubicular head direction cells compared during visual landmark discrimination , 2017, Brain and neuroscience advances.

[9]  Steven Mills,et al.  A neural network model for learning to represent 3D objects via tactile exploration , 2018, CogSci.

[10]  S. Tipper,et al.  Object-based attentional mechanisms: Evidence from patients with unilateral neglect. , 1994 .

[11]  M. Moser,et al.  A prefrontal–thalamo–hippocampal circuit for goal-directed spatial navigation , 2015, Nature.

[12]  Kate J. Jeffery,et al.  Neural encoding of large-scale three-dimensional space—properties and constraints , 2015, Front. Psychol..

[13]  Barbara Hammer,et al.  Merge SOM for temporal data , 2005, Neurocomputing.

[14]  Lorenzo Rosasco,et al.  Active perception: Building objects' models using tactile exploration , 2016, 2016 IEEE-RAS 16th International Conference on Humanoid Robots (Humanoids).

[15]  Teuvo Kohonen,et al.  Self-organized formation of topologically correct feature maps , 2004, Biological Cybernetics.

[16]  Stephen Grossberg,et al.  How Entorhinal Grid Cells May Learn Multiple Spatial Scales from a Dorsoventral Gradient of Cell Response Rates in a Self-organizing Map , 2012, PLoS Comput. Biol..