Riddle me this. Dingus is in fact really happy! If you think about it, the “reward system” is like the dopamine response animals get when they do something enjoyable (they feel good/happy)… So if dingus continues to push the bolder up this map for the rest of existence, he will always be happy.
I think it’d be cool if he had control of his hands so he truly masters pushing the boulder by having to slightly adjust it with his hands instead of mostly just walking forwards and turning after he finds the boulder at the start. And of course just like everyone else I’d like to see randomized or procedurally generated courses and starting positions so that he can learn to master boulder pushing in general. Great video!
The concept of a 'human playable' version of this is interesting because technically Dingus has no reaction time or playback speed. You just advance a frame, ask Dingus what it wants to do in this state, and it outputs a response. So you could make it run as fast or as slow as you want but Dingus would do the same things, however a human would find it way harder or easier.
Dingus Edit: Your videos are really cool! This video specifically was like a follow-up to your other Dingus pushing boulders up hills but in a slightly cooler scale!
is dingus learning how to go through a preset course, or learning what to do in each situatuon, based off environmental inputs? If it is the second one, why don't you use the AI from the last boulder pushing video / why do you reset dingus' AI every time he goes to a new level in ur other videos?
It’s Definitely just learning to adapt to this course. He even points out himself it’s just learning which action it needs to take roughly at which times. Idk how much input it receives from its enviroment either way.
I wonder what would happen if dingus or the boulder started at an ever so slightly different starting location. I bet it would generalize a lot more and not "memorize" the course if it had different starting locations
His rotation is randomised, but starting locations aren't. My gut tells me he'd still very likely still overfits because he'd likely get the ball and himself to a specific location and then just take the exact same actions afterwards.
@@Dingus_Labs That is a really good point! especially in some of the narrower parts. Forces single state system. As a completely separate project it would be interesting to see a Dingus vs Mingus Where Mingus is a level creator trying to make dingus fail. (Some nice adversarial networking)
@@Dingus_Labs-- Shuffling the course segments could reduce the reliance on overfitting, but I'd love to see a procedural course. Ideally you should be able to throw Dingus at any hand made course and train him to success much faster if he started with general training.
Putting A.I through different challenges like this is super fun and pretty entertaining. I could imagine a free for all situation where all A.Is are trying to complete different goals in a race
See if you can get a dingus to explore an open environment searching for Something to bring back to a set point. Maybe these items could be obscured by simple platforming puzzles, or evading enemy dingus'.
Suggestion: For reference (and because I enjoy seeing others fail) I would love to see you try your own challenges. The reason is to get a reference point for the controls Dingus has to deal with and how hard (or easy) it would be for a normal thinking human.
Yeah I want to play around with "team" and "adveserial" AIs. I'm exploring something like that at the moment but hard to tell if it will work like I want until I'm in deeper.
AI can't learn the actions for random hills because it keeps changing but maybe if there was some sort of image recognition (I don't know if the AI called "Dingus" has any sort of camera or image recognition) for the bumps maybe the AI could learn to avoid them.
The Max speed (when not going down a slope) isn't all that high, it could definitely be interesting if I made the max speed higher so there was room for further skill expression by completing quickly.
Imagine if dingus has to learn to cook, like taking food out before it gets too hot and it burns or using the right ingredients lol, that’d probably be hours
sometimes it´s easier to make AIs follow paths generated through some calculations on random terrain, slowly reducing the importance of the path spline for pattern recognition or association (at the very least theoretically... this is how we humans learn - we get a task, solution and some explanation of their correlation, then make some observations, associations and our approximations get better and better. some people can do integrals, derivatives, roots and more in their heads even without other signs of neurodivergence.) there are lots of different AI types and many training methods, eg. parameter spaces, liquid neural networks and spiking neural networks, each better and better, there´s letting the AI go at it, giving it rewards as it gets closer to completing its tasks and giving it examples to learn from. there are also handmade models, indirect controls, etc. which can lead to better results, less erratic movements and more. you have to make the physics a tiny bit inconsistent or add in miniscule variations in the level so the runs aren´t all the same. and also you should be proud of the dog if he can do similar things with similar toys. if there are dog races, there have to be dog logic / maths competitions as well.
This is such an interesting development of content we used to have rage games that lets players would do. Now we have content about training an ai to do that couse with commentary it almost feels like second person
@@CertifiableGuy i don't think so, you could just include the old code for as dingus gets closer to the finish he would get more rewards, then just make dingus spawn at random x,y coordinates that you handpicked (could be like 5) on the map, i dont remember how exactly dingus "sees" the world, but if its through ray casts then it should be possible
It would be fun to see a more open ended task, where it learns strategies For example Dingus could stack things to try to reach places But it would be even more fun to have two(or more) AIs compete
I’m thinking you should add some randomness to the maps in order for dingus to not just learn the exact movements, but to actually learn how to complete the challenges.
I wonder how dingus would have dealt with moving platforms. I feel like the ai might have a tougher time with the idea of stopping it’s movement. Awesome vid tho :O
100% agree I think he'd have a very hard time waiting, these trials all let him go at his pace, the concept of waiting would be completely foreign and would likely need a lot of specific training to learn.
aren’t the boulderless escapades an indication of a massive oversight? Dingus doesn’t care about the boulder at all, and that kind of makes his learning void if he’s just learning a set of moves to repeat. maybe add a punishment for greater distances to the boulder, and maybe variation where the boulder spawns? i love the series, that just struck me as odd!
Do you have a way to run the original boulder pushing Dingus on this map to see how well it'd do on the new map, and vice versa by putting this Dingus that learned this map on the old maps? Basically comparing their training by seeing how well they'd do on each other's hills?
One shall imagine Dingus Happy.
Waiting for the "you beat me to it" type comments
Was gonna type "One must imagine Dingus happy", but yours is so close it'd feel like copying lol
@@revimfadli4666 you beat me to it
Dont forget again
@@revimfadli4666waiting for the "waiting for X" type comments
I’m so proud of Dingus for having hands again
When will dingus learn to use a gun?
soon
@@user-qv2mc3dw5o oh no.
👀
He kinda has in the old Videos playlist
As a tank
I love all the falling boulders in the background, implying that there are other Dingi making attempts above our POV.
Yea
This is equivalent of "Getting over it" for the dingus.
Sisyphus would be proud.
lil
lol
one must imagine sisyphus proud
Oh no, Dingus has dingers
lol
Again!
Don't call it that.
@@juliewinchester1488 oi. Get ya minda outa the gutter.
you know, i’m still half convinced that this is a secret dankpods channel.
FRANK!!
No, there's too little yelling
the endless suffering of dingus, he learn only pain, but we like to watch this do it Dingus
I love that after loosing the ball he gives up and walks off the cliffs great simulation of depression
Riddle me this. Dingus is in fact really happy! If you think about it, the “reward system” is like the dopamine response animals get when they do something enjoyable (they feel good/happy)… So if dingus continues to push the bolder up this map for the rest of existence, he will always be happy.
“Should I be proud or frustrated?” Yes!
But then, Dingus got this run:
Great background music and volume!
You've improved 🤘
Thank you, I spent more time on the audio and mixing this time
This is like the AI version of Getting Over It.
Don’t worry Dingus, I’d find this hard too
It might've helped to punish dingus for being too far away from the Boulder so he won't try to go without the Boulder
He doesn’t get any points, so it’s probably punishment enough
Randomize start boulder position so he has to learn to look for it.
Dingus is getting too good at this
The enviroments in the background are REALLY pretty and interesting, good job on them! could be a world map for something
I think it’d be cool if he had control of his hands so he truly masters pushing the boulder by having to slightly adjust it with his hands instead of mostly just walking forwards and turning after he finds the boulder at the start. And of course just like everyone else I’d like to see randomized or procedurally generated courses and starting positions so that he can learn to master boulder pushing in general. Great video!
hes doing sick flips again!! :3
The concept of a 'human playable' version of this is interesting because technically Dingus has no reaction time or playback speed. You just advance a frame, ask Dingus what it wants to do in this state, and it outputs a response. So you could make it run as fast or as slow as you want but Dingus would do the same things, however a human would find it way harder or easier.
Dingus.
Dingus
Edit: Your videos are really cool! This video specifically was like a follow-up to your other Dingus pushing boulders up hills but in a slightly cooler scale!
You did really good on the scenery in this one
you should do this again but with procedurally generated courses so dingus masters the art of boulder pushing, and doesn't just memorize everything
I've had a few requests for this, will look into potential approaches.
Will likely be a while though
is dingus learning how to go through a preset course, or learning what to do in each situatuon, based off environmental inputs? If it is the second one, why don't you use the AI from the last boulder pushing video / why do you reset dingus' AI every time he goes to a new level in ur other videos?
Curious about this too! Surely the more you do with it the less over fitting there will be and the more like.. universal it'll be
Though, I'd guess the training from the start is a big part of the fun of these
It’s Definitely just learning to adapt to this course. He even points out himself it’s just learning which action it needs to take roughly at which times. Idk how much input it receives from its enviroment either way.
Both, but mostly the former
It's more interesting to watch probably
I wonder what would happen if dingus or the boulder started at an ever so slightly different starting location.
I bet it would generalize a lot more and not "memorize" the course if it had different starting locations
His rotation is randomised, but starting locations aren't. My gut tells me he'd still very likely still overfits because he'd likely get the ball and himself to a specific location and then just take the exact same actions afterwards.
@@Dingus_Labs That is a really good point! especially in some of the narrower parts. Forces single state system.
As a completely separate project it would be interesting to see a Dingus vs Mingus
Where Mingus is a level creator trying to make dingus fail. (Some nice adversarial networking)
@@Dingus_Labs-- Shuffling the course segments could reduce the reliance on overfitting, but I'd love to see a procedural course. Ideally you should be able to throw Dingus at any hand made course and train him to success much faster if he started with general training.
One must imagine dingus happy
This was amazing! Thank you for sharing the link too! Dingus did great
oh no dingus has HANDS
edit: DINGUS HAS HANDS
thank you for the edit, it was very helpful
dingus and your whippet are both adorable
I enjoy these vids so much! Please keep them coming
How is this channel not more popular?
im always happy to watch dingus try to figure stuff out
A truly glorious run. Definitely interested in expanded versions of old ideas.
this is an extreamely underrated channel, Keep up the awesome videos Mr Labs!
What can Dingus perceive though? The location of the ball? Does he also see the environment?
Putting A.I through different challenges like this is super fun and pretty entertaining. I could imagine a free for all situation where all A.Is are trying to complete different goals in a race
See if you can get a dingus to explore an open environment searching for Something to bring back to a set point. Maybe these items could be obscured by simple platforming puzzles, or evading enemy dingus'.
He’s done something like this already, but the enemy part would be cool to see.
ua-cam.com/video/UOvWqU1SBog/v-deo.htmlsi=LepZ91ptQ0-e0Phb
I wonder if training would improve if some of the agents were running on a mirrored map to force a bit of general knowledge and reduce overfit.
A good question, I could see it becoming more generically capable but I'm not sure if it would be faster, I'll have to play around with that
Dingus labs is possibly one of the most perfect, unimaginably, unequivocally, preposterously, coolest dudes ever
Dingus is the drift king 9:35
i love your content, Thank you for making each video quality content :)
Suggestion: For reference (and because I enjoy seeing others fail) I would love to see you try your own challenges. The reason is to get a reference point for the controls Dingus has to deal with and how hard (or easy) it would be for a normal thinking human.
So what I've learned is that Dingus may be a masochist.
He just wants to see if he can take it.
dingus is such an inspiration, training in limbo to become the boulder dribbler of all time
I'm glad we got to revisit Sisyphingus, it was entertaining
i'd be VERY interested in seeing how well dingus would cope with having to work together with another dingus. can AI learn teamwork?
Yeah I want to play around with "team" and "adveserial" AIs. I'm exploring something like that at the moment but hard to tell if it will work like I want until I'm in deeper.
Yeah!! Noice!! Dingus is a very cool dude.
May Dingus find peace and finally become a Restful Cube one day.
What about randomised hills?
AI can't learn the actions for random hills because it keeps changing but maybe if there was some sort of image recognition (I don't know if the AI called "Dingus" has any sort of camera or image recognition) for the bumps maybe the AI could learn to avoid them.
You need to do one of these where you run it over night and train for speed and just see how fast dingus can go.
The Max speed (when not going down a slope) isn't all that high, it could definitely be interesting if I made the max speed higher so there was room for further skill expression by completing quickly.
1:08 mabye hes a massochist who knows
Dingus
Translate it to English.
@@LoafOfCat123 dingus (in english)
@@iamtheonewhoaskedlol
@iamtheonewhoaskedlol
34 minutes ago
@LoafOfCat123 missing (in english)
That's what showed up in the translation
@@LoafOfCat123 dingus is missing!!
we love dingus
This channel reminds me of another channel that does something similar to this, the name of the channel is AI warehouse
I've had a few people point it out, after looking I definitely agree! They have real clean editing too!
New Dingus video, let's go!
i can confirm that this hill is indeed insane
The background is awesome
always fun wathcing dingus finding the way to heaven
One must imagine dingus happy 😔
Imagine if dingus has to learn to cook, like taking food out before it gets too hot and it burns or using the right ingredients lol, that’d probably be hours
Hear me out. Dingus learns to play Rainworld.
That is going to require a much bigger Dingus!
Dingus has ascended.
when do we get a video on dingus doing a depressive 9 to 5 job
this IS his 9-5
@@Artificer_ its not, 9 to 5 should be boring, this atleast has some action
Nice terrain and clouds!
Sisyphus is proud of him
Dingus looks like he's doing the Mr. Beast pose in the thumbnail XD
4:13
dingus is truly evil sisyphus
Dankpods would love this channel
Dingus for President! /J
I love this silly little cube boi
sometimes it´s easier to make AIs follow paths generated through some calculations on random terrain, slowly reducing the importance of the path spline for pattern recognition or association (at the very least theoretically... this is how we humans learn - we get a task, solution and some explanation of their correlation, then make some observations, associations and our approximations get better and better. some people can do integrals, derivatives, roots and more in their heads even without other signs of neurodivergence.)
there are lots of different AI types and many training methods, eg. parameter spaces, liquid neural networks and spiking neural networks, each better and better, there´s letting the AI go at it, giving it rewards as it gets closer to completing its tasks and giving it examples to learn from.
there are also handmade models, indirect controls, etc. which can lead to better results, less erratic movements and more.
you have to make the physics a tiny bit inconsistent or add in miniscule variations in the level so the runs aren´t all the same.
and also you should be proud of the dog if he can do similar things with similar toys.
if there are dog races, there have to be dog logic / maths competitions as well.
This is such an interesting development of content we used to have rage games that lets players would do. Now we have content about training an ai to do that couse with commentary it almost feels like second person
dingus !!! my sweet boy!! he's back!!
dingus over it
maybe start dingus in random spots of the map so that dingus will learn how to play the map instead of 'remembering' the map
Would be super hard to code
@@CertifiableGuy i don't think so, you could just include the old code for as dingus gets closer to the finish he would get more rewards, then just make dingus spawn at random x,y coordinates that you handpicked (could be like 5) on the map, i dont remember how exactly dingus "sees" the world, but if its through ray casts then it should be possible
I have a cat named Dingus
His full name is Megamind Pander Dingus, but we just call him Dingus for short
I have to wonder if the dingus from the other boulder video would have more luck solving this
your videos are amazing
You have an adorable doggo!
Ringus Mc.Dingus
I'm going to remake this in gmod but replace dingus with Maxwell cat because some people call Maxwell dingus
You should make dingos have friends and play multiplayer games like Mario Wii
Oh actually, a game of Dingus kart would be good. That's possibly doable within a reasonable amount of work, I will have to investigate!
Aaaand todayyyyy...
Dingus become...
Sisyphus!!!
I love dingus™
Make a challenge where he is randomly assigned a speed, to see if he can handle it
Oh dingus you are a dingus
I’m dingusing all over the place rn
10:32
and then, Hefest got this run
I like the background
King Dingusphus
It would be fun to see a more open ended task, where it learns strategies
For example Dingus could stack things to try to reach places
But it would be even more fun to have two(or more) AIs compete
man that's a pretty background scene
dingyphus
I’m thinking you should add some randomness to the maps in order for dingus to not just learn the exact movements, but to actually learn how to complete the challenges.
I wonder how dingus would have dealt with moving platforms. I feel like the ai might have a tougher time with the idea of stopping it’s movement. Awesome vid tho :O
100% agree I think he'd have a very hard time waiting, these trials all let him go at his pace, the concept of waiting would be completely foreign and would likely need a lot of specific training to learn.
dingus prime
aren’t the boulderless escapades an indication of a massive oversight? Dingus doesn’t care about the boulder at all, and that kind of makes his learning void if he’s just learning a set of moves to repeat. maybe add a punishment for greater distances to the boulder, and maybe variation where the boulder spawns? i love the series, that just struck me as odd!
Do you have a way to run the original boulder pushing Dingus on this map to see how well it'd do on the new map, and vice versa by putting this Dingus that learned this map on the old maps? Basically comparing their training by seeing how well they'd do on each other's hills?
You should livestream dingus learning things!
Hoping to be in a position to do that again soon!