RatSLAM uses grid cell techniques to map subdivision

hexagons
slam
grid-cells
robotics

#1

I know many of you have heard about SLAM and already read this, but here is the paper, and here is a link to the video where it gets really cool:


Every time I think about grid cells for too long…

DGkHxOfUQAIXEXe


An Apical Depolarization for Numenta: How to Generate the Allocentric Location Signal
#2

Cool Video for my breakfast! Thanks Matt!


#3

The link you gave to the paper does not work.


#4

I’m not sure if this is the same paper from Matt’s post, but this is the one I usually send people: http://www.itee.uq.edu.au/think/filething/get/790/Milford_PersistantNavigationAnfMap.pdf


#5

Hi there,

While still booting on the HTM theory I ran into this interesting Deepmind paper. I think I read that it postulates that the grid cells are used for planning rather than localisation:

https://www.biorxiv.org/content/biorxiv/early/2017/07/25/097170.full.pdf


#6

Check out the answer by Ajit Rajasekharan.
Very nice answer:
https://www.quora.com/What-are-the-major-differences-between-grid-cells-and-place-cells-and-why-do-we-need-them-both

What does this have to say to SDR and HTM?
First - think of the hippocampus as a sort of encoder for time and space. Patient HM shows that this is not the space where long-term memory is stored but we can’t form these without the hippocampus. Grid/place/boundary/path/head direction are all things that are parsed out and somehow passed to the cortex. Once in the cortex, it is usable “as is” so the cortex must duplicate this spacial map in some way.

Speculation: The vestibular “self” is not mentioned anywhere in this but it absolutely must be part of time and space calculations. The codes are envirocentric. There must be personal space in there somewhere. The tissue between must be the part that relates you in the environment.