You really should release parts as parametric or at least the source files. I see everything is an STL and STLs are just a pain to work with. Suppose we want to try with mice? Or what about my cat? I do not expect just scaling in my slicer is going to end up with a good result, I'll need to redo everything from scratch. But parametric parts? That gives us a lot faster iteration. That gives you a lot faster iteration too! I highly recommend taking that approach when designing and I find it is worth it more often than not.
Could you add cost estimates to the BOM? These never need to be accurate but I always find it helpful when estimating a project. You're just saving people from the time it takes to click every single link and throw them into a calculator. And informs people very quickly what to innovate on to drive costs down. (Sorry, BOMs without cost estimates are a big pet peeve of mine)
# Questions:
- Do the rats enjoy playing Doom?
- Are there specific games the rats like to play?
I've never thought about what types of videogames other animals would enjoy, but damn if you didn't just open Pandora's Box here. I actually think we could learn a lot about them (and even their specific personalities) from this question. It gives a whole other level of refinement than just knowing what my cat's favorite toys and games are...
And also, thanks for open sourcing this! I'm excited to see what comes of it!
As an evolutionary cousin of a rat, the half-second delay after firing my weapon and the response would make me want to eat my whiskers.
This would completely kill any potential reward (and replace it with the opposite, frustration) you're trying to train me with, please fix immediately.
I'm opposed to this project because it involves using animals in medical experiments, which I believe is never ethically justifiable. It goes against basic moral and ethical principles regarding animal treatment. If the project were designed to allow animals to choose whether or not to participate, it would be more acceptable. Some scientists have already explored such approaches. By not giving animals a choice, you're limiting their freedom and potentially exposing them to physical or psychological harm through your simulation. As someone who advocates for animal rights, I'd prefer to see alternative methods that don't involve animals or allow them to participate voluntarily
>We reached the point of rat habituation but didn’t start training. Our rats (Todd, Kojima, Gabe) aged out before full testing. The setup works, but behavioral validation is pending.
Ah man, what a pity. That VR rig is awesome, but it doesn't really seem to me they are planning to continue these experiments, or do they?
The year is 2034. Countless attempts at re-producing the sophisticated wetware of the brain have failed. Modeling research has proved unfruitful, with the curse of dimensionality afflicting every attempt at breaking the walls of general intelligence. With only a few million of capital left, and facing bankruptcy, they knew that only one option remained.
I think there is probably an anthropocentric flaw in the rat display.
Human eyes are side-by-side and forward, with a big binocular overlap and a clear vanishing point. A forward-facing curved screen fits that geometry well.
Rats' eyes, by contrast, are lateral. They have a much wider field of view, a tiny binocular zone, and use motion and contrast more than neat perspective lines. A single human-style "cinema screen" isn't laid out for a rat's optics or brain.
Perhaps if the scenes were rendered with a much wider, 250° FOV, it would help the rat understand what it was seeing better.
Or even rendered with two virtual cameras offset and angled apart, then stitch their outputs into one extra-wide view wrapped onto the curved display. That would approximate the rat’s much wider horizontal field of view and reduce the mismatch between where its eyes are actually looking and where the important visual information appears.
There are other differences in perception of color and motion, but fixing the FOV would be an immediate and relatively easy software fix.
Doom aside (very impressive), I love the concept of putting rats on a roller ball instead of in a wheel for exercise. It would be better on their backs.
A huge shortcoming with this setup is that rats don't just see in front of them, they see all around them. The screen right in front of his nose doesn't really simulate his environment the way it does for a human. Luckily doom has command line options for screens on the left and the right. But I imagine a lot of his awareness comes from what's above him.
In the year 2032, the rodrone wars broke out. What started as innocent video game experiments, had taken a dark turn. After the rats had perfected playing the classic Doom games in 2026, they were easily trained on more complex simulations. In 2028 the first rodent controlled surveillance drones were tested. The year after weaponised. And the first real deployments in warfare in 2030. They easily outperformed OpenAIs latest battle systems. Home robots were soon after remote controlled by rats, known as Rodots. Rodent intelligence escalated quickly as lab selections bred only the smartest specimens. It was only a matter of time before the takeover. Now rodots were building steadily more capable drones and bots. Long before humans could foresee the need for a defense against the rats, the rodrone attack on all humankind, was a fact...
Bad news: no video of them playing on this setup, just on the previous version. We iterated on v2 too long, our pet rats grew old and couldn't be trained. We open-sourced the hardware and software so others can build upon it. You can TLDR the whole thing in this thread: https://x.com/yolorun_capital/status/1996632980903620886?s=2...
...Also, here's my personal X, dm me if you have any questions, or would want to build it for a lab or for yourself: https://x.com/viktor_thoth
Here is a thought: Maybe we are all just living in some gigantic experiment, that some incomprehensibly advanced civilization has created, to make us indirectly play DOOM, without us even realizing. Any progress we make is just one step further in their experiment, a reward, every issue like war, climate change, corruption, put in place by them as a challenge, while behind it all, in higher dimension, it is translated all into input for their version of DOOM, while they are watching us perform for them. One of them found it funny or interesting, to give us our own DOOM to play with, and now their scientists are speculating, whether it is always the case, that when you give a test population a DOOM, that they will pass a DOOM onto other beings, recursively.
I did not expect this to be as thoughtful as it is. Kudos to the thought put into the setup and training. Building custom hardware for the rat is amazing.. I need something like this for my own setup.
Very cool... but why not use something with higher res graphics, if you're going to go through the trouble of building the mechanical system. You could go ultra 4K and see if a rat begins to see... reality.
Is this related to the Tank lab experiments at Princeton?
I’ve seen this before with the practical application being that you can keep the animal stationary while it thinks it’s moving, in order to record brain activity.
Wonderfully crazy. Should the rats have multiple weapons at their disposal, which do they prefer? How many demons escape their fate? Are there videos of gaming sessions? So many questions!
Whats the fascination with Doom? I keep seeing running Doom on this and that and now Rats playing Doom. I mean ok, Doom runs on everything and now rats playing doom. So what?
Rats Play DOOM
(ratsplaydoom.com)428 points by ano-ther 12 December 2025 | 159 comments
Comments
# Suggestion:
You really should release parts as parametric or at least the source files. I see everything is an STL and STLs are just a pain to work with. Suppose we want to try with mice? Or what about my cat? I do not expect just scaling in my slicer is going to end up with a good result, I'll need to redo everything from scratch. But parametric parts? That gives us a lot faster iteration. That gives you a lot faster iteration too! I highly recommend taking that approach when designing and I find it is worth it more often than not.
Could you add cost estimates to the BOM? These never need to be accurate but I always find it helpful when estimating a project. You're just saving people from the time it takes to click every single link and throw them into a calculator. And informs people very quickly what to innovate on to drive costs down. (Sorry, BOMs without cost estimates are a big pet peeve of mine)
# Questions:
- Do the rats enjoy playing Doom?
- Are there specific games the rats like to play?
I've never thought about what types of videogames other animals would enjoy, but damn if you didn't just open Pandora's Box here. I actually think we could learn a lot about them (and even their specific personalities) from this question. It gives a whole other level of refinement than just knowing what my cat's favorite toys and games are...
And also, thanks for open sourcing this! I'm excited to see what comes of it!
This would completely kill any potential reward (and replace it with the opposite, frustration) you're trying to train me with, please fix immediately.
Ah man, what a pity. That VR rig is awesome, but it doesn't really seem to me they are planning to continue these experiments, or do they?
"Bring me the rats."
Human eyes are side-by-side and forward, with a big binocular overlap and a clear vanishing point. A forward-facing curved screen fits that geometry well.
Rats' eyes, by contrast, are lateral. They have a much wider field of view, a tiny binocular zone, and use motion and contrast more than neat perspective lines. A single human-style "cinema screen" isn't laid out for a rat's optics or brain.
Perhaps if the scenes were rendered with a much wider, 250° FOV, it would help the rat understand what it was seeing better.
Or even rendered with two virtual cameras offset and angled apart, then stitch their outputs into one extra-wide view wrapped onto the curved display. That would approximate the rat’s much wider horizontal field of view and reduce the mismatch between where its eyes are actually looking and where the important visual information appears.
There are other differences in perception of color and motion, but fixing the FOV would be an immediate and relatively easy software fix.
A Carmack might've kept the system stable long enough
Great project btw!
Bad news: no video of them playing on this setup, just on the previous version. We iterated on v2 too long, our pet rats grew old and couldn't be trained. We open-sourced the hardware and software so others can build upon it. You can TLDR the whole thing in this thread: https://x.com/yolorun_capital/status/1996632980903620886?s=2...
...Also, here's my personal X, dm me if you have any questions, or would want to build it for a lab or for yourself: https://x.com/viktor_thoth
This could give a whole new meaning to "the rat race"
Perhaps I’ve just done too much miniature wargaming.
Guys, you should do it with a cockroach ^^ https://makeagif.com/gif/fifth-element-remote-controlled-coc...
https://en.wikipedia.org/wiki/Project_Pigeon?wprov=sfti1#
That will feel next level bad when it happens.
Missed opportunity to name them Neo, Morpheus and Trinity.
I’ve seen this before with the practical application being that you can keep the animal stationary while it thinks it’s moving, in order to record brain activity.
hehe very important
although... if the walls moved and touched the whiskers in conjunction with the game could be something
Fairly benign, compared to all the other things that get done to rats in labs, I would think.
[1]: https://archive.ph/tCAGu
[2]: Live Run ISO: https://archive.org/details/msdos_Super_Solvers_Operation_Ne...