I love the solver AI in this game, it's so cool to see the very specific dance of moves it finds to optimize the puzzles. I designed assignments for a class on this kind of thing once and it's always a blast to watch.
@@SunbleachedAngel It's very easy to just BFS through all the possible game states and cut off any search that reaches a previously seen state since the total number of game states is very constrained (if you have 3 characters on an 8x8 grid that's only 64*63*62 possible game states ignoring walls which would lower it even further). The one exception to this is that bunnies are capable of creating an explosion in the number of game states which is why I think the game crashes sometimes when you have weird things going on with bunny mating.
@@SunbleachedAngel As other people have already replied, it's typical to solve these with some kind of search algorithm on the state space. The simplest method is just breadth-first search (look at every possible move from the current game state, and then every possible move from those new game states, until you find the goal), though there are other more complicated search methods as well. You can also use A*, which is normally thought of as a pathfinding algorithm: A* uses a heuristic to guess at how far away the goal is from the current state (for example, how many moves it would take to get each player to a carrot if you removed all the dynamic obstacles), and this lets you search more efficiently by focusing on more promising-looking states (as it's probably better to focus your search on states where the puzzle looks nearly solved). In general, if you can bake information about the game into the solver, it's possible to make the search run a lot faster. It can seem a bit weird to describe this as AI given the recent focus on machine learning in the AI community, but this is the same type of search algorithm that is used in AI chess players, and in most other turn-based board game AI (with the added assumption that your opponent will pick moves that are good for them). There was an interesting AI for the puzzle game 2048 that I saw using this, which treats the random spawning of new tiles as an "opponent" that you have to assume will make the most harmful moves possible.
It would be cool for this game to have leaderboards like: - fewest tiles altered (with at least completing the puzzle with 0-3 carrots) - longest steps taken
@@F_A_F123 Might be that I didn't notice before or just have bad memory, but I was concerned that with more complex levels finding the exact optimal solution could become tedious.
13:10 the second one is better while the first one is worse due to parity. Statrting with blue makes the cucumber have to double back an additional time.
If I had a penny for every time tyler has utilized bunny mating for puzzle solving I'd have 2 nickels. Which isn't a lot, but it's sus that it happened twice
For everybody who is wondering. The AI is based on an exhaustive BFS in a space of all board positions (as opposed to a space on board tiles) for all characters at once. Therefore, it always finds an optimal solution or exhausts all possible board states to prove that there is no solution. The algorithm is honed specifically for this task and is coded in C++ for faster calculation. So it works well on any potato PC.
Most likely it probably has a move limit as well to avoid infinite loops. Baba is you has an interesting easter egg where if a step takes more than a 100 recursions (I think?) it shows an "infinite loop" or "too complex!" screen.
@@Frn1There's almost certainly something in there where if it features a board state it's already seen, that set of moves is cut out, since it's just wasting moves (e.g. moving up-right-down-left would put the character in the exact same place, but now the move counter is +4, so it throws it out)
Just noticed that the symbols used for the gates/buttons/sensors also have a number "hidden" in the design. The red is hardest to see with 1, green is 2, blue is 3, etc
I discovered this channel today and tbh I already like it too much had just seen the rest of the episodes for this game so I'm happy we get to see more
What I learned from this video: Never let Tyler be a level designer, he WILL make you wait a real-life year for virtual bunnies to finally meet just to get through one level, with no gameplay on your part whatsoever.
13:40 if you move the final red fork closer to the onion knights blue fork and put a green fork right before the goal it makes the larry outside of "The Gauntlet" double back ONE more time for a high score
When trying to get the two bunnies to meet as late as possible, wouldn't the best starting position be one step away from the meet up point (facing away from it)? Ignoring what happens when they meet, they would move in a reliable cycle meeting up at regular intervals, so one step after such a meeting is the longest time before they would meet again.
Funny how he said that the game is giving him false hubris that he can design puzzles, when all you do here is make the "puzzles" as long and tedious as possible. Gonna make the next Cocoon, Tyler? (I enjoyed that game but i do have to admit it was certainly slow)
I imagine you’ve considered it @Aliensrock, but if you haven’t; posting edits of the streams, at least whatever part of a stream gets the most viewership, may be bad for the YT recommendations. In my amateur, educated guessing opinion. I say this because despite watching all of your videos daily, to the point of searching when you aren’t at the top of the home page, I obviously didn’t end up watching the Balatro video as I’d seen the whole thing Live. UA-cam’s algorithmic response? That I must have gotten bored of your content, and am no longer even the slightest bit likely to click on it if recommended. You don’t show up in my recommended now. Not a big deal for someone who will intentionally search because they know your schedule, but if it affects the overall “random” recommendation chance, which it likely will, new viewers may stop seeing your content entirely even after enjoying the Balatro video because UA-cam categorized it as different from the typical content “fans” of your channel like. My point being, that in editing and posting as a video, the most watched part of a recent stream here on UA-cam, due to the watch behavior of fans who already saw it live, the growth of your channel through UA-cam’s recommendation algorithm will likely be harmed. I don’t know how much of your growth it attributes, or more importantly whether you particularly care about that growth, but I wanted to share that thought as I considered it possible that you hadn’t considered it, and may care. Thanks for another great video! Mostly commenting cause I know interaction will do the opposite to recommendation strength, aka help, and want to make up for helping get your recommendation chance lowered.
I love the solver AI in this game, it's so cool to see the very specific dance of moves it finds to optimize the puzzles. I designed assignments for a class on this kind of thing once and it's always a blast to watch.
Can we get a link to that? I’m curious
I am so incredibly curious to see how the AI solver in this game works
@@SunbleachedAngel It's very easy to just BFS through all the possible game states and cut off any search that reaches a previously seen state since the total number of game states is very constrained (if you have 3 characters on an 8x8 grid that's only 64*63*62 possible game states ignoring walls which would lower it even further). The one exception to this is that bunnies are capable of creating an explosion in the number of game states which is why I think the game crashes sometimes when you have weird things going on with bunny mating.
@@SunbleachedAngelnot even tral AI, the grid is small enough to just bfs thr fastest route to each tile
@@SunbleachedAngel As other people have already replied, it's typical to solve these with some kind of search algorithm on the state space. The simplest method is just breadth-first search (look at every possible move from the current game state, and then every possible move from those new game states, until you find the goal), though there are other more complicated search methods as well. You can also use A*, which is normally thought of as a pathfinding algorithm: A* uses a heuristic to guess at how far away the goal is from the current state (for example, how many moves it would take to get each player to a carrot if you removed all the dynamic obstacles), and this lets you search more efficiently by focusing on more promising-looking states (as it's probably better to focus your search on states where the puzzle looks nearly solved). In general, if you can bake information about the game into the solver, it's possible to make the search run a lot faster.
It can seem a bit weird to describe this as AI given the recent focus on machine learning in the AI community, but this is the same type of search algorithm that is used in AI chess players, and in most other turn-based board game AI (with the added assumption that your opponent will pick moves that are good for them). There was an interesting AI for the puzzle game 2048 that I saw using this, which treats the random spawning of new tiles as an "opponent" that you have to assume will make the most harmful moves possible.
It would be cool for this game to have leaderboards like:
- fewest tiles altered (with at least completing the puzzle with 0-3 carrots)
- longest steps taken
The developers have announced that they're going to make per-level highscores soon.
The total steps taken across all levels is a cool leaderboard, but I’d definitely love a per-level one!
@@Aliensrock Per-level highscores are in the game for a few weeks now. You just have to unlock them :)
Imagine someone beats Alien’s 125 step.
@@10EightyPlaysYou can get a few more by adding a diagonal step for the cucumbers, so it can be done :)
I love that the 3 carrot solutions are no longer the literally best you can do but good enough to show you figured it out.
It was the same way previous time Tyler played it
@@F_A_F123 Might be that I didn't notice before or just have bad memory, but I was concerned that with more complex levels finding the exact optimal solution could become tedious.
@@hendrikrasmus They seem to be more like an author medal from trackmania, just the best that the author did.
@@kingacrisius the maker said so in the discord ye. 3 carrots is the solution he was able to find
13:10 the second one is better while the first one is worse due to parity.
Statrting with blue makes the cucumber have to double back an additional time.
I love how Tyler restarts the whole puzzle just to place down one more tile 2:50
Kinda funny
i just realized that this is the SECOND game the tayler played that has bunny-mating as a prominent game mechanic
What’s the first?!
@@markob8934 Paquerette Down the Bunburrows
@@Kore153well technically he played this game before bunburrows
he play the demo@@silksongreactions
If I had a penny for every time tyler has utilized bunny mating for puzzle solving I'd have 2 nickels. Which isn't a lot, but it's sus that it happened twice
Ah yes, the game where you find increasingly extreme ways to slow down speedrunning vegetables
For everybody who is wondering. The AI is based on an exhaustive BFS in a space of all board positions (as opposed to a space on board tiles) for all characters at once. Therefore, it always finds an optimal solution or exhausts all possible board states to prove that there is no solution. The algorithm is honed specifically for this task and is coded in C++ for faster calculation. So it works well on any potato PC.
Does combinatorial explosion mean the algorithm will eventually be slowed to a grinding halt for a board of sufficiently large size?
@@danielchan1668I would imagine it would have to
Most likely it probably has a move limit as well to avoid infinite loops. Baba is you has an interesting easter egg where if a step takes more than a 100 recursions (I think?) it shows an "infinite loop" or "too complex!" screen.
@@danielchan1668 maybe, but A-B pruning could probably cut down a lot
@@Frn1There's almost certainly something in there where if it features a board state it's already seen, that set of moves is cut out, since it's just wasting moves (e.g. moving up-right-down-left would put the character in the exact same place, but now the move counter is +4, so it throws it out)
Just noticed that the symbols used for the gates/buttons/sensors also have a number "hidden" in the design. The red is hardest to see with 1, green is 2, blue is 3, etc
13:07 I knew Tyler would outsmart the game devs eventually
13:44 Feel proud, Tyler. Icely only got 111.
Icelypuzzles
Chess battle advanced.
Your lieing there genetic clones and they have the same brainwaves
@@NoodleNut81 lying*
damn I wonder how much extra time the editor spent on editing these videos with the subtle "highlights", much appreciated
Live Service Developers maximizing player engagement be like
Ah, yes. Cubecumbers from A Monster's Expedition.
I discovered this channel today and tbh I already like it too much had just seen the rest of the episodes for this game so I'm happy we get to see more
love seeing ur vids tyler, watching them with the family!
Ah-rah, dee
Soo-guh-goo-gee-goo-gee
Goo-guh fli-goo gee-goo
Guh fli-goo, ga-goo-buh-dee
Ooh, guh-goo-bee
Ooh-guh-guh-bee-guh-guh-bee
Fli-goo gee-goo
A-fliguh woo-wa mama Lucifer!
Yea
Pretty much
What I learned from this video:
Never let Tyler be a level designer, he WILL make you wait a real-life year for virtual bunnies to finally meet just to get through one level, with no gameplay on your part whatsoever.
Haven’t finished the video yet, but i have to say i love all the edits sprinkled in. Great work :)
Will you go back to that “already solved” game ,it looked really good also I love your videos and I’m glad your feeling better after your break
Maybe whenever I get tired of this game. I think Veggie Quest is the most interesting game in its micro niche so far, though.
This game looks like an abosulute torture for the brain
Thank you for completing the levels for us!
I want the return of that eggplant
13:00 72 steps was enough for 3 carrots... TYLER GOT A FULL 53 EXTRA
Thank u for making my day everyday
: )
Tyler has gotten so good at puzzles he is thinking ahead of the devs!
13:40 if you move the final red fork closer to the onion knights blue fork and put a green fork right before the goal it makes the larry outside of "The Gauntlet" double back ONE more time for a high score
15:32 Wait, is this a confession? 🤨
I feel like you'd get something out of 4D golf. It's not a puzzle game, but it's a great proof-of-concept.
When trying to get the two bunnies to meet as late as possible, wouldn't the best starting position be one step away from the meet up point (facing away from it)?
Ignoring what happens when they meet, they would move in a reliable cycle meeting up at regular intervals, so one step after such a meeting is the longest time before they would meet again.
This is the first puzzle game I've seen that is probably harder to program than is to either design or solve.
Not to say that it's actually easy to design or solve....
So glad he’s back ♥
the idle piano music is quite calm and nice ngl
9:09 pplacing blue to the start seems better
Great game!
I am mind blown 12:56
this game where i don't understand almost everything but i very much enjoy
Petition for us to sdd lore for this game, I mean it's vegetables with ton of characters and collecting other vegetables plenty of room for lore
I would love to see a clock on screen for how long it is taking you to solve these puzzles
cubecumber is my new favourite word
13:08 WOW
13:39 MORE WOW
I love this game so much but I’m literally gonna get this game
What did you do to the middle 5:17
I thin the puzzle at min 10 could be even more if you make a snail about the inner block. So the one cucumber must go 3 steps more that would ad up.
Keep up the good work
18:22 you what?!?!
Least common multiple strats
Cool game
Love this gamies
Funny how he said that the game is giving him false hubris that he can design puzzles, when all you do here is make the "puzzles" as long and tedious as possible. Gonna make the next Cocoon, Tyler?
(I enjoyed that game but i do have to admit it was certainly slow)
6:55 The forks go Red, BLUE and GREEN, but buttons go red, GREEN and BLUE
Pain
I feel like is aw the puzzle at 7:30 in the last video
I've watched so many of his videos that I somehow have gotten to the point where I think of almost the exact same thing as him
if at the long one you had RGBRGB at the top, wouldn't that still be solvable? and that one cucumber needs to double back _again_
Nice.
Alien's rock advanced
chess battle rock
Drop everything you're doing, new veggie quest video
Less than a min. let's go!
Nice
can i has cheese, wisconsone man
How did you score 125 when the 3 carrot amount was around 95. Aliensrock is crazy.
There has to be a way to get even more than that...
You didn’t do all the new levels😢
Hi Tyler❤
13:38
😂😂😂He shold try mosa lina WITHOUT SWEARING😂😂😂😂
125 steps he got 6 carrots
vegtabl
.. PLEASE tell me what tou see at 5:20
Overseen
lol
What happened to the cat Steve that only moves 5 times videos?
Beat the game
@@Aliensrock oh but where are the rest of the videos trying to find the other cats animals’ cryptids all that
I think you can get 146 on the one you got 125 on.
You should play windowkill! It’s incredible and extremely cool!
Dam my notification was 3 minutes latw
I imagine you’ve considered it @Aliensrock, but if you haven’t; posting edits of the streams, at least whatever part of a stream gets the most viewership, may be bad for the YT recommendations. In my amateur, educated guessing opinion.
I say this because despite watching all of your videos daily, to the point of searching when you aren’t at the top of the home page, I obviously didn’t end up watching the Balatro video as I’d seen the whole thing Live. UA-cam’s algorithmic response? That I must have gotten bored of your content, and am no longer even the slightest bit likely to click on it if recommended. You don’t show up in my recommended now. Not a big deal for someone who will intentionally search because they know your schedule, but if it affects the overall “random” recommendation chance, which it likely will, new viewers may stop seeing your content entirely even after enjoying the Balatro video because UA-cam categorized it as different from the typical content “fans” of your channel like.
My point being, that in editing and posting as a video, the most watched part of a recent stream here on UA-cam, due to the watch behavior of fans who already saw it live, the growth of your channel through UA-cam’s recommendation algorithm will likely be harmed. I don’t know how much of your growth it attributes, or more importantly whether you particularly care about that growth, but I wanted to share that thought as I considered it possible that you hadn’t considered it, and may care.
Thanks for another great video! Mostly commenting cause I know interaction will do the opposite to recommendation strength, aka help, and want to make up for helping get your recommendation chance lowered.
End card vid links to this vid
can u please play paquerette down the bunborrows again after the updates
emag elzzup ................
Fun fact: the full game of i wanna lockpick is out!!! you should try it!!!
why does this cost $20???
There’s a TON of puzzles, and they’re really well made.
Anyone knows if it takes some moves for the fork to rise up or is it just instant after they unpress the button
I think it’s instant for buttons but delayed for sensors
The one piece is real
The last time icy puzzle do the puzzle at 14:8 he accidentally make infinite moves oh and the game still somehow allow the puzzle to start
🧅🍴🥕
i wont veji cwest
Bro why do you need 3 carrots? For content? no I don't give a ####
fourk
I think I broke UA-cam 💀
sub day comment
Day two of asking aliensrock to play worldless
It's not a puzzle game tho, he won't
It would be cool but the guy on top of me is right
He might play it. There were some games that he played that were like it.
Day 1236 of telling Tyler that he never finished his minecraft series and that he should finish it.
omg he said the thing
WHY IS THE EGGPLANT HOT
no
bro play vampire survivorssssssssssss
first
Second ig
No views