Roddy Grieves

Shortcuts as a behavioural signature of cognitive map use? My awesome colleagues Eleonore Duvelle & Roddy Grieves revisit Tolman’s sunburst maze and are not convinced…

Amazing work. And ouch, a heavy blow to one of my favourite literature "classics".

There are still many other Tolman classics to choose from, just don't read Ciancia's (1991) review on the detour task and you will be fine...

pdf4pro.com/view/tolman-...

Your monograph on the Tolman sunburst maze never explains how templates explain the shortcutting behaviour. It just describes several examples of motor learning.

Your templates cannot be used to solve the water maze, the animals learn in one or two trials and can then navigate directly to the platform from novel start positions.

What you are describing is procedural memory, which requires many trials and depends on different brain regions.

OpenRatSLAM: an open source brain-based SLAM system - Autonomous Robots RatSLAM is a navigation system based on the neural processes underlying navigation in the rodent brain, capable of operating with low resolution monocular image data. Seminal experiments using RatSLAM...

ratSLAM paper: link.springer.com/article/10.1...

Algorithms that do this (like ratSLAM) learn about landmarks by initially estimating their relative positions in a topological map.

They then use self-motion information to iteratively update this topological map to a metric, topographical one.

I think the brain works in a similar way...

I do think that self-motion is used to help generate a map under normal conditions though.

In a new place, an animal has to simultaneously localise itself and create a map. In robotics this is known as the SLAM problem.

However, the second part - the allocentric reference frame - is a bit difficult based on self-motion alone.

Any resulting map would always have to be relative to something (a nest, a starting point, a feature).

So, I think you need more than JUST self-motion.

Ultimately, a (spatial) cognitive map represents the relationships (vectors) between features in an allocentric reference frame.

It is possible to learn these vectors using self-motion information alone, although that would take a lot of time and limits you to features you can interact with.

No. Animals use the constellation of distal cues, not one cue.

Often there is no cue next to the platform.

Animals can also solve the task after a curtain has been pulled around the entire platform quadrant - pulling a curtain around the whole maze results in chance performance (Morris, 1984) 👇

JOVE's AI-voiced video describing the Morris water maze (app.jove.com/v/5418/spati...) :

"Visual cues are arranged around the maze to associate each quadrant with a specific symbol" 😕

"The rodent should learn to associate the location of the platform with the location of a specific visual cue" 🤦

Thanks, hope you like it!

Thanks!

I would also recommend reading this review:
doi.org/10.1142/S021...

Does a great job explaining the different types of navigation and the evidence for them.

Motor Programs: Concepts and Issues | 3 | Attention and Performance Xi Lashley introduced a number of arguments that sequences of human motor activity are guided by plans and not by peripheral or central associations between one

I concentrate on the supermarket example because it involves navigation. Your other examples are what we call motor programs or motor sequences.

Not really my field, but good paper to read here:
www.taylorfrancis.com/chapters/edi...

Consistent with some of your ideas and also highlights issues.

Another question about your supermarket example: the path you show is extremely efficient (it never back tracks or crosses itself).

How would a naive person derive that motor sequence in the first place? Without other forms of navigation how could they know which items are close to each other?

Some navigation is cetainly what we call procedural or route-based, where animals execute a pre-learned sequence of actions.

These sequences may nor may not be associated with cues such as landmarks by the way - they need not be entirely self-motion based.

A motor program could not be used to guide navigation (find the detour) in this case.

This was one of the central arguments of Tolman's cognitive map theory - some behaviours cannot be explained by sequential motor actions.

To use your supermarket example: we may have a motor sequence for visiting items in a supermarket (seems unlikely though, who buys the same products every time?), but what happens if one day one of the aisles are blocked?

Most people would be able to take a detour to the next item they want.

Your manuscript describes many motor programs (starting a car, playing piano, entering a sequence of numbers etc), but it never addresses why these are an issue for Tolman's results or theory?

Motor sequences AND cognitive maps can both exist, they are not mutually exclusive.

Thanks Arne! And thanks for the peer review 😜

I’m not fully sold on transparent peer review as a reviewer, but it’s great from the author side of things.

We talked a lot about your comment: needing more on shortcutting post-Tolman - we will need to write another paper for that though!

Thanks Jörn, glad you liked it!

Perhaps, although that is a bit of a self-fulfilling prophecy...

Did they learn the sequence of turns or associate a visual cue with reinforcement and beacon to that in the test?

I'm not sure it has ever been specifically tested if what is learned latently is actually map-based.

That's funny because she says the exact same about you on Mastodon!

Our paper is specifically about shortcutting in Tolman's sunburst maze experiment.

We are not trying to argue that cognitive maps do or do not exist.

Just that the sunburst maze experiment is not a good demonstration or test of this.

Latent learning is very much a finding that has been studied across many groups.

It is also consistent with current place cell research - we know place/grid cells form a representation of space even in the absence of reward. This map can then inform behaviour when reward is introduced.

Haha, good to know it is useful and thanks for motivating us to finally get around to writing it!

I would recommend checking out Aidan's preprint:
Cognitive maps are flexible, dynamic, (re)constructed representations
doi.org/10.31234/osf...

There are other studies and lines of research that would be better cited in its place.

Perhaps we will write something more broadly on shortcutting and cognitive maps very soon...