We're back folks! Sorry to keep you waiting so long. Unexplored has been on my to-do list for a long time and is now available on everything from PC to Switch, so you have no excuse not to try it out! I'll be back soon not only with a new Design Dive episode but another patron-voted episode on another indie game.
Awesome! I bumped into you at a Rezzed event a few years back and you recommended checking out Joris Dormans' work on level generation here. Really interested to see how they apply this to the open-world aspect of Unexplored 2
Oh boy, I've almost forgotten what's it like to just bump into people! 😅 But yeah this has been on my to-do list for a while, so with the sequel coming this year I figured it was time to get it out there.
If you have spent much time with Source SDK, and been on the valve developer wiki. The ideas of Bounce, Loops, and Push level design elements are strongly encouraged and expanded on.
As much as I find this fascinating and elegant, for a randomly generated dungeon crawler like Unexplored it feels overkill. I played a bunch of this game and it never felt much different from other games similar games in terms of progression.
How far down did you get? The first 3 floors or so do feel like traditional floors. (albeit with a lot better pathing, idk how you never noticed that). But get past floor 3 and there is nothing like it at all.
I wish more games were like this: pushing the envelope of what's possible with AI. I will def check out both games, and thank you for another amazing breakdown.
I had been thinking about this problem and did some investigation on games and papers and I think I've had some insights that I still haven't seen anyone come up with. I'd been working on formalizing them, and then real work intruded. One key insight is that all these key-lock problems which I've abstracted to events and the impasses they resolve can be boiled down to a tree of nodes that contain those events and relations (impasses) which cannot be traversed without visiting some boolean combination of nodes containing a set of events required for the impasse. The tree can be generated sequentially, guaranteeing a path to the terminal node. The series of cycles described here can be translated to this. Translating the nodes to temporal/spatial layouts is itself a huge challenge, but flexible. All games with impasses can be reduced to such a graph, I hope to prove.
@@judgedbytime you might consider it doubly directed by default with the potential for directed edges. Or consider all doubly directed with one way traversals described by an unresolveable impass. Cycles can exist in the structure. I'm not sure if my thinking has evolved on it or not or how it is different than the cycles described in the video. I'd have to rewatch and check my notes for what I was trying to say. I think the reduction removes the cycles despite the fact that there may be cyclic paths in space. It breaks the system down to more logical steps where events make impasses traversable.
I just recently thought "Huh. I wonder what Tommy is up to. Didi't see a video from him for quite some time." And then you drop this one..just as I was wondering about you. Great and really ingeresting video as always! 😁
Yeah I was having a tough time getting started again after taking some time off around Xmas. Was just exhausted. That said, I really wanted to get a new episode out for January so got my head down and put this one together.
Just a small unrelated note. I literally watched all the videos on this channel + Smoke and Mirrors (including those I watched previously) and still RUclips showed this video on my newsfeed 3 days after it was uploaded.
This is exactly how I was planning to set up my level design for my game, although my ideas is a FPS and the level would basically be one map with something like 5 of these dungeon areas making up the entire level. Glad to know that it is functional, although I'm not coder and I don't know how far one can push Chat GPT4 and the mod tools of Left 4 Dead 2
Awesome! Purely clever iterative design. Design a system that takes a set of grammars, and uses those to generate tilesets, to generate rooms. Then make an open world game.... that generates GRAMMARS to generate dungeons. Very clever.
Knowing that doom is a 2D top down shooter from a computer's perspective and a 3D first person shooter from player perspective, I really wish that all of these procedurally generated 2D games would use that to their advantage. I love the idea of a (good) procedurally generated first person shooter
Excellent video, as always, Tommy. Extremely interesting system, has me on the lookout for unexplored 2 to see the expansion of the generation and parsing. Very interesting. Have you ever considered doing a video on Monster Hunter World? I feel that the monster AI has a lot of trickery and different decision making rules it follows for what movesets it should use, where it should go, how it interacts to other monsters or the player on a monster-by-monster basis, and it’d make for a very interesting video. Thanks for keeping up the videos all these years
Given I'm currently streaming MH:W on Twitch right now (tune in on Monday), the idea has been sitting in the back of my head. Though certainly the audience have been clamouring for it for a while now. That said, I have done some digging and I've yet to find anything concrete. That said, I may do a Design Dive later in the year.
@@AIandGames thanks for the reply, Tommy. I definitely understand the need for concrete reasons behind the AI behavior. Hopefully you can crack it, but No matter what video game you do it’ll be worth the time, hope you have a good one!
I found this game some Years ago because I read the article about the cyclic generation. Fascinating toppic. I hope more games and developers pick something of it up
I'm wondering if you'd be interested in doing an analysis of the drivatars of Forza Horizon. They have to be the worst racing A.I. I've ever played against in a racing game, and to sum them up in a word they're obnoxious. Getting to know the inner workings of what makes them so un-fun would be interesting.
The dungeons in Moonlighter got very boring for this reason (Binding of Issac and Undermine had similar issues but not as bad and they had more interesting side rooms). There was always just 1 path with branches that would never connect to anything other that the next room in the branch. It made the game less interesting. Interconnectivity is the way to go!
"Wow, this game sure sounds cool! I'm gonna check out the release date for the sequel!" *epic games link* My excitement had never been murdered so mercilessly.
Sounds interesting as concept but to be real nobody would develop an entire engine to generate levels for some style points, instead of making elements of PGC as submodules in unity or unreal environment using hilbert curves or cellular automata or some other basic methods.
Except they did develop an entire engine to develop the high-level design, twice. The original Ludoscope engine is not only a standalone software, but was subsequently ported into Unity for use in the two Unexplored games.
12:00 love the transition from the 2D astethic to the isometric 3D perspective. A great way to show they evolved their game from the first one.
This is the opening shot of the game and it's *really* nice when you see it for the first time.
We're back folks! Sorry to keep you waiting so long. Unexplored has been on my to-do list for a long time and is now available on everything from PC to Switch, so you have no excuse not to try it out! I'll be back soon not only with a new Design Dive episode but another patron-voted episode on another indie game.
@AI and Games is there any way you can look at the rainbow six Vegas ai ??
man this pretty smart game design!!
"Hmm, I wonder how he'll talk about Spelunky today" 🤔
I ask myself this question every day.
@@AIandGames Keep it up! Love it 😁
Awesome! I bumped into you at a Rezzed event a few years back and you recommended checking out Joris Dormans' work on level generation here. Really interested to see how they apply this to the open-world aspect of Unexplored 2
Oh boy, I've almost forgotten what's it like to just bump into people! 😅
But yeah this has been on my to-do list for a while, so with the sequel coming this year I figured it was time to get it out there.
I absolutely love this game and cant wait for the sequel. Thanks for covering this underrated gem!
If you have spent much time with Source SDK, and been on the valve developer wiki. The ideas of Bounce, Loops, and Push level design elements are strongly encouraged and expanded on.
Very informative video. I always look forward to you explaining various implementations of "AI" in games :)
As much as I find this fascinating and elegant, for a randomly generated dungeon crawler like Unexplored it feels overkill. I played a bunch of this game and it never felt much different from other games similar games in terms of progression.
I wonder if making the cycles akin to Dark Souls' could improve things...
How far down did you get? The first 3 floors or so do feel like traditional floors. (albeit with a lot better pathing, idk how you never noticed that). But get past floor 3 and there is nothing like it at all.
I absolutely love this game. I'm glad it's still getting coverage
I wish more games were like this: pushing the envelope of what's possible with AI. I will def check out both games, and thank you for another amazing breakdown.
I had been thinking about this problem and did some investigation on games and papers and I think I've had some insights that I still haven't seen anyone come up with. I'd been working on formalizing them, and then real work intruded. One key insight is that all these key-lock problems which I've abstracted to events and the impasses they resolve can be boiled down to a tree of nodes that contain those events and relations (impasses) which cannot be traversed without visiting some boolean combination of nodes containing a set of events required for the impasse. The tree can be generated sequentially, guaranteeing a path to the terminal node. The series of cycles described here can be translated to this. Translating the nodes to temporal/spatial layouts is itself a huge challenge, but flexible. All games with impasses can be reduced to such a graph, I hope to prove.
You plan to turn the cyclic graph into an undirected graph (tree)?...
@@judgedbytime you might consider it doubly directed by default with the potential for directed edges. Or consider all doubly directed with one way traversals described by an unresolveable impass. Cycles can exist in the structure. I'm not sure if my thinking has evolved on it or not or how it is different than the cycles described in the video. I'd have to rewatch and check my notes for what I was trying to say. I think the reduction removes the cycles despite the fact that there may be cyclic paths in space. It breaks the system down to more logical steps where events make impasses traversable.
I just recently thought "Huh. I wonder what Tommy is up to. Didi't see a video from him for quite some time."
And then you drop this one..just as I was wondering about you.
Great and really ingeresting video as always! 😁
Yeah I was having a tough time getting started again after taking some time off around Xmas. Was just exhausted.
That said, I really wanted to get a new episode out for January so got my head down and put this one together.
Just a small unrelated note. I literally watched all the videos on this channel + Smoke and Mirrors (including those I watched previously) and still RUclips showed this video on my newsfeed 3 days after it was uploaded.
This is exactly how I was planning to set up my level design for my game, although my ideas is a FPS and the level would basically be one map with something like 5 of these dungeon areas making up the entire level.
Glad to know that it is functional, although I'm not coder and I don't know how far one can push
Chat GPT4 and the mod tools of Left 4 Dead 2
Great work! I love nap generation algorithms, a cyclic system looks interesting
Awesome! Purely clever iterative design.
Design a system that takes a set of grammars, and uses those to generate tilesets, to generate rooms.
Then make an open world game.... that generates GRAMMARS to generate dungeons. Very clever.
Knowing that doom is a 2D top down shooter from a computer's perspective and a 3D first person shooter from player perspective, I really wish that all of these procedurally generated 2D games would use that to their advantage.
I love the idea of a (good) procedurally generated first person shooter
Great video! really worth the waiting :D
Excellent video, as always, Tommy. Extremely interesting system, has me on the lookout for unexplored 2 to see the expansion of the generation and parsing. Very interesting.
Have you ever considered doing a video on Monster Hunter World? I feel that the monster AI has a lot of trickery and different decision making rules it follows for what movesets it should use, where it should go, how it interacts to other monsters or the player on a monster-by-monster basis, and it’d make for a very interesting video. Thanks for keeping up the videos all these years
Given I'm currently streaming MH:W on Twitch right now (tune in on Monday), the idea has been sitting in the back of my head. Though certainly the audience have been clamouring for it for a while now. That said, I have done some digging and I've yet to find anything concrete. That said, I may do a Design Dive later in the year.
@@AIandGames thanks for the reply, Tommy. I definitely understand the need for concrete reasons behind the AI behavior. Hopefully you can crack it, but No matter what video game you do it’ll be worth the time, hope you have a good one!
I found this game some Years ago because I read the article about the cyclic generation.
Fascinating toppic. I hope more games and developers pick something of it up
Great video. Nice one Tommy. :)
Awesome! Great video
really good video, thanks
interesting as always; thnx for sharing ... can we have something about Phasmophobia AI please ?
I'm curious if their generation process is abstracted into a graph grammar.
It might be the rum, but I have no idea what is going on, gota re-watch this one
i'm using fsm today in the game jam
I'm wondering if you'd be interested in doing an analysis of the drivatars of Forza Horizon. They have to be the worst racing A.I. I've ever played against in a racing game, and to sum them up in a word they're obnoxious. Getting to know the inner workings of what makes them so un-fun would be interesting.
The dungeons in Moonlighter got very boring for this reason (Binding of Issac and Undermine had similar issues but not as bad and they had more interesting side rooms). There was always just 1 path with branches that would never connect to anything other that the next room in the branch. It made the game less interesting. Interconnectivity is the way to go!
"Wow, this game sure sounds cool! I'm gonna check out the release date for the sequel!"
*epic games link*
My excitement had never been murdered so mercilessly.
Hi, sorry to go off-topic. Can you make a video about AI in Cyberpunk 2077 and why exactly it is so bad?
...
First?
First!
I knew you could do it
...
Sounds interesting as concept but to be real nobody would develop an entire engine to generate levels for some style points, instead of making elements of PGC as submodules in unity or unreal environment using hilbert curves or cellular automata or some other basic methods.
Except they did develop an entire engine to develop the high-level design, twice. The original Ludoscope engine is not only a standalone software, but was subsequently ported into Unity for use in the two Unexplored games.
...