I was wondering if there has been any implementation of Nengo that uses a system similar to grid cells and place cells in the brain in order to integrate velocity/direction inputs. The closest thing to what I’m looking for is: Spiking ratSLAM, however the document itself is not that detailed.
To give some more information, I would like to use some kind of attractor network, where each cell represents a particular location or orientation. Is there an efficient way to use such kind of ‘place code’ instead of the distributed representation Nengo normally uses? I imagine it should be possible by having an ensemble that has a function/weight matrix that ‘bins’ the incoming vector representation.
Thanks a lot!
Hi @MatthijsPals. Coincidentally, @yaffa just got a paper accepted to CogSci on this topic of grid cells in Nengo. Perhaps she can give you some suggestions for things to try or places to look? @brent may also have some suggestions as he has been investigating localization and navigation in Nengo.
Thank you @arvoelke, I will have a look at their work/try to get in touch with them!
Hey @MatthijsPals, just a quick update here:
Thanks a lot! Looking forward to delving into those papers
In case anyone has a similar question and stumbles across this topic http://compneuro.uwaterloo.ca/files/Conklin.PathIntegration.pdf also has a nice implementation of a “grid cell like” network for path integration (thanks to yaffa for suggesting it).