Josh Galecki - Procedurally Generating Puzzles

Puzzles are a common design element in many genres - but not roguelikes. Roguelikes are designed for repitition, and a puzzle that's been solved once provides no challenge and no satisfaction on the third (or thirtieth) playthrough. In order to get around the 'curse' of repitition, Josh leans on some strategies from mathematics in order to create systems for producing puzzles that continually engage players, even across many (many) playthroughs.

[@geisttraft7190]

Interesting talk, but the puzzle examples in the presentation don't seem that engaging to solve repeatedly. They lack any major "a-ha" moment (like the puzzles in Baba Is You or Monster's Expedition) and feel more like the sort of puzzles you'd solve in the Sunday newspaper. That's fine as a small distraction meant to break up the game's pace, but I wonder if there's a way to make more interesting puzzles with some of the elements from the truly great puzzle games.
One issue that comes to mind is that a lot of puzzles in puzzle games are designed around a specific interaction between two or more elements that a player hasn't seen before, causing them to learn something new with every puzzle. That's difficult to replicate in a roguelike, since once the player learns the interaction, they'd be able to solve any puzzle that relies on that interaction pretty easily. You'd need either a massive list of different interactions to use or procedurally generate new interactions with each run.

[@user-cn4qb7nr2m]

Agreed!
I'm obsessed with this idea of generating conceptually novel problems.. I tried to make open world thingie where object types generate dynamically with different rule combinations as you explore, but the project went nowhere.
My solution for the moment is loosening the puzzliness of the problem, opening multiple possible goals that the player could pursue. Working on it...

[@settergren938]

I've been thinking about this as well. What I believe causes the difficulty is that puzzles that teach game mechanics are designed specifically with the intent to convey emotions to the player, and the puzzles are also often integral to the game's story (as in Baba is You and Can of Wormholes). Generating such puzzles would be just as hard as generating meaningful literature, since they have to be coherent with an overall structure in addition to being novel on their own. Frankly, I still think that the puzzles described in this video are meaningful in their own way, since they provide atmosphere and an 'illusion of design' to the game's world. I particularly like the open-ended river puzzle, since the dams interact with each other in interesting ways, which helps prevents the player from finding a simple formula which solves any problem of a given type (which might become an issue with the torch puzzle).

[@user-cn4qb7nr2m]

​@@settergren938 Cool, I want then to make some points about free puzzles. I think strict puzzles are one of the most mature game genres today (top 2, beside logic competetives), love it. But also recently I started thinking about art in general and returning to puzzles, I thought "What a hubris you need to have, to make hundreds of people retrace your exact mental path in this particular interactive system. It kinda assumes that you know what is the most interesting and valuable about the system."
What's the alternative? You make a complex system, you teach the basics in the best way possible and then you let go of the player. Do not set the upper bound on their progress, let them explore the rules at their pace in their chosen direction. Procedural generation is about it. Over time it abstracts mind from details and shows the system itself.

[@moonroofstudios9705]

Hello, speaker here. I agree with your points here. This Moondrop's puzzles are meant to be small challenges rather than any sort of 'storytelling' puzzles which evoke emotions or those moments of deep insight. (Credit to Nat Alison for the storytelling idea - I think you'd really enjoy her Rogulike Celebration talk "In Defense of Hand Crafted Sudoku": ua-cam.com/video/rcIVCrddozQ/v-deo.html.) Puzzle games like Monster's Expedition or the Witness are fantastic, but they are a different breed than procedural-generated roguelikes. They aren't meant to be replayed the same number of times. Different constraints.
For Moondrop, the focus of the puzzle section isn't puzzle-solving, but rather exploration. The puzzles are things you can encounter, but I've seen playtesters not even realize that certain puzzles are puzzles. (I mean, the mazes and such are obvious, but some are more subtle.) And once you realize there's something there, you have to figure out what to do with it. Once you've learned the mechanics of each puzzle, then further repetitions give you something to do, but you've exhausted the "a-ha" moments of the puzzle. At the same time, the variety of initial conditions often means you can't just solve these on autopilot either.