Adaptive Difficulty

[Bells]

One of the things I've always found frustrating about roguelikes (or some roguelikes at least), is how difficult it can be for a beginner to make progress in the game. It takes a large amount of trial and error to get to know which monsters you can attack safely without losing too much health, how long to spend on a dungeon before descending so as you don't run out of food or miss any important items you need later etc.

For some people this is part of the fun and indeed I find it quite satisfying to spend a few days casually playing and learning the basic mechanics of a game and seeing my increasing progress, but having begun designing a roguelike myself with a heavy focus on intelligent and adaptive AI, I'm wondering whether the concept of adaptive difficulty would play well in a roguelike setting or not.

For those that aren't familiar, adaptive difficulty usually works by assigning the player a score based on how many times they die during a set amount of time. So for example, if you die lots throughout the entire game, you will end up with a low score, whereas if you die over and over on a particularly difficult part of a level but are fine through most of the game, you will still have a higher score (and thus a more challenging game).

Obviously, with permadeath, the score would be calculated not on how many times you die but instead on hp. So for example, if you lose a large amount of hp every time you go into combat you will end up with a lower score, than someone who is more skillful in combat and loses fewer hp (speaking theoretically here, an actual technical implementation would be more complicated I imagine).

Anyway, what I'm getting at is the overarching goal of reducing player frustration. So let's say a player isn't very good, we reduce the health of some of the more powerful monsters or spawn fewer of them to make combat 'lighter', the AI is less likely to use powerful attacks, or it neglects to take certain factors into consideration when deciding which action is 'optimal', so for example, the smart AI might notice a door and a corridor (slightly further away) which it can escape to and choose the corridor because that is the most familiar path (anything could be on the other side of the door), whereas the less intelligent AI might choose the door, only to find that it's locked, giving the player more time to attack, more chances to attack.

Just some ideas that have been floating around in my mind anyway, thought I would share them as I'm interested to hear peoples thoughts on this. In reality, I think it would be difficult to implement and I'm thinking that if I did include it as an option in the game, it would be limited to boss battles or strategic fights which can be controlled and balanced at different difficulty levels, with the appropriate difficulty level being selected based on the players 'score'. Although they would still have the option of 'taunting' the monster which would anger it and increase the difficulty level (good for players who'd done the fight before but wanted more of a challenge the second time around).

[rust]

I don't really like the idea. It could enable situations like "monster X is too strong for me, better die a few times to make things easier" as opposed to "monster X is too strong, better gear up or find his weak points to beat him" in other roguelikes. This is assuming that the game would collect the data over several playthroughs. Your original idea about loss of hp would be even easier to exploit by intentionally letting easy monsters bring you to low hp.
As for boss battles, in my opinion it'd be better to somehow educate a player before them (so that they more or less know how to handle said monster) than to decrease the fight's difficulty.

[Bells]

@rot13

I was pretty much going through the same thought process this afternoon but you summed it up far better than I could have. It did cross my mind whether or not you could actually build a game based on 'cheating' or exploiting those kind of mechanics...

[jere]

The only way it would work is if the player didn't know such a system existed and thus wasn't tempted to exploit it. And how are you going to benefit from your adaptive difficulty system without telling players your game has such a neat adaptive difficulty system....

Some forms of adaptive difficulty probably already exist. Ghosts in DCSS for example.

[Vanguard]

Games that lower their difficulty until the player can win are a dead end.  Roguelikes are about learning a little bit more every time until you finally succeed.  There's no need to bother with that if the game is going to beat itself for you.  There's no reason to fear death if it still brings you closer to victory.

The only way it would work is if the player didn't know such a system existed and thus wasn't tempted to exploit it. And how are you going to benefit from your adaptive difficulty system without telling players your game has such a neat adaptive difficulty system....

Roguelikers are tenacious and curious.  If your game is popular, and if your adaptive difficulty system is meaningful, someone's gonna figure it out, and once people know how to exploit it, that's what they'll do.

So if they're going to exploit the system anyway, why not take it in the opposite direction?  Design the game with manipulation in mind, make it fun and interesting to exploit.  Maybe players need to make periodic sacrifices of some kind to keep the difficulty down.  Maybe a high difficulty is desirable in some way - increased experience and better treasure could come with along with the harder enemies.  Maybe the player can alter the difficulty relatively freely, but higher difficulties enable new content or alternate endings.  The equivalent of an ADOM ultra win or a Crawl 15 rune win could be something that only becomes possible if the player deliberately raises the difficulty to its highest setting.  There are tons of things that could be done with the idea.

This isn't even unexplored territory.  There are games that have done these kinds of things and succeeded.  The best example is probably Battle Garegga, a game where nearly every action you take alters the difficulty level, and where mastering that system is necessary to succeed.  It's worth reading about even if you don't play shmups.

[Bells]

A very enlightening post Vanguard, I've actually been thinking of simply having specific areas or routes which I suppose you could call 'gauntlets'. The routes themselves will be challenging, intense strategy areas where there is great risk but also great rewards. Perhaps these could only be unlocked by completing special feats, for example obtaining a kill streak of monsters with level x or higher without losing any hp, etc. I like the idea of collecting some kind of item or 'token' which unlocks a special ending or even further challenge areas, special NPCs, gear, abilities etc. This might be a nice way to add some dynamism and replayability to the static areas I'm featuring in my game.

I hadn't actually heard of Shmups but the system is really interesting, will definitely have to study that a bit later and see whether the mechanics will trigger a streak of inspiration in me.

[UltimaRatioRegum]

I've often thought shmups/RLs have significant overlap, especially with the idea of the TLB (true last boss) in a shmup, vs extended games in RLs as optional but highly challenging extras. And (as dpeg from DCSS suggested to me) they both have a similar "arcade mentality" re: starting/ending a game, permadeath, how many sessions you're likely to play, etc.

[koiwai]

I'd say that the adaptive difficulty is even harder to balance than the fixed difficulty. Even if it sounds like it can make the game easier in some cases, it may also break the game entirely in some other cases, or make it not fun. Now, a player cannot be sure, whether killing a certain uber-boss was his or her real achievement, or the boss got nerfed.

Try to balance the game in such a way that an inexperinced player does not die at the same stage of the game again and again. As long as each new run brings new experience to the player, it should be fine. If the game has enough varaibility, and the difficulty progression is not too steep, even an inexperienced player will learn and get better before they got bored. If it's not enough, you may add an option to resurrect a character, or add gods or something similar to help the player when they are in trouble.

Another thing is psychological. if you want adaprive difficulty, make it the other way around. Start the game easy on the player. And then, if the player is progressing very quickly without a trouble, add difficulty, add difficult encounters, make the game nastier, make it look more bloody/gory, etc. And show this indicator of danger large and at the top of the screen, so the player will be trying to get this difficulty indicator as high as possible. Then, reaching a particularly high difficulty level would be an achievement, and that's what the player will be bragging about.. People get attrackted to such things. 

[Pickledtezcat]

Lots of games have treasure rooms with a special challenge inside or gollems or skeletons which are frozen but only attack if you mess with them, or tombstones... There's usually something you, as a player, can do to make the game more difficult if you want to.

There's an interesting question behind this issue though: Should beating a roguelike be only about gaining meta knowledge and exploiting it to reduce difficulty?

Like for example in Pixel dungeon, where there are several types of treasure room which have a special way of beating them. The room always spawns an appropriate potion to deal with the trap somewhere in the level. Once you know that, you can always beat the treasure room and scoop the reward. Once you acquire that piece of meta knowledge the game becomes instantly easier.

There are many roguelikes where you just play again and again, picking up more knowledge of exploits or shortcuts, it gets kind of boring after a while. Luck should play a part of course, but most games have some mechanic to reduce the impact of luck on player success. Like making powerful items almost useless if collected early because of level or stat restrictions on the item.

Shouldn't there be something else though? Some other factor to decide success or failure?

I enjoy playing regular RPGs, and one of the things I like most is playing through the first time on a fairly easy mode as an "ironman" mode (no save /reload loop, delete save game if I die). If I miss a side quest or do badly on a fight, or level up my charters in a suboptimal fashion, that's OK. It's kind of like life, no second chances. I love finding a puzzle which I've never encountered before and solving it with logic, not with trial and error. I love it that sometimes I can't solve the puzzle and I have to continue without getting the promised reward, simply because I wasn't smart enough. The locked chest will still be lurking there in my imagination, all the more poignant because I don't know what's inside. Later I'll go through the game again on a higher difficulty mode, try to get the stuff I missed, but it won't be quite as fun as the first time around, but theres still the element of challenge from the higher difficulty and trying to get all the secrets and side quests.

Perhaps auto adjusting difficulty has a part to play here.

How about making the game more difficult in areas where you've already had a successful play through before? That means that the game is somewhat easier in areas you're experiencing for the first time. There's more chance to make mistakes but not be brutally punished for it. While exploring new ground the game can be a little less punishing and so more enjoyable. But when you do die and have to begin from the start again, the parts of the dungeon you know well will not be a walk in the park, despite you knowing more about the monsters and traps and game mechanics. This opens up two kinds of challenges. The first time you play you're trying to make good decisions, predict monster vulnerabilities, guess which items or spells are the best look for logical places that traps might be placed, try to predict how a puzzle might be solved. The challenge is intellectual. The second time through an area you're trying to leverage your meta knowledge, pick up every advantage you can so you can get back in to unexplored territory. The challenge is more technical.

[chooseusername]

I remember when Morrowind came out.  Or was it Oblivion?  In any case, whichever one it was had this adaptive difficulty where you could visit a location multiple times and depending on how far your character had progressed by that point, it would be repopulated differently.  It was a design conceit.  It was a conceit chosen as a cheap way to balance and populate the game.  But the effect was that the quality of the game experience also felt cheap and fake, as the experience created did not fit in the context.

Of course, your typical roguelike doesn't aim to create an open world experience.

[reaver]

Adaptive difficulty is prone to exploit, by the very definition of it. Players can underperform on purpose, and on demand. You carefully plan to have a big bad challenging boss with quite a few tricks up its sleeve? Player can exploit the adaptive difficulty to make the boss easy. So, your hard work to create the challenge is diminished, the players sense of reward got (self-)diminished, and you just made it easier for the player to skip content quickly.

Open world is a massively different kind of beast , and the elder scrolls games, while creating experiences like thieves in caves being stronger than dragons and whatnot, did a pretty good job imo considering the task at hand. You can't have it all: non-linear world exploration, being actually able to go anywhere you wish without getting one-shotted and not adapting the difficulty somehow.

[Vanguard]

You can't have it all: non-linear world exploration, being actually able to go anywhere you wish without getting one-shotted and not adapting the difficulty somehow.

You can have it all because high level areas where you get one-shotted are a good and desirable thing.

[reaver]

You can have it all because high level areas where you get one-shotted are a good and desirable thing.

Yes, I agree that they are good things, but if it's impossible to move through an area without getting killed, the experience becomes linear, that's all I'm saying.

[Kevin Granade]

My take on adaptive difficulty is simply letting the player chose their battles.  My game is a sandbox, so I haven't put much thought into how this would interact with a more traditional structured roguelike.
For what it's worth, I get a lot of complaints that the game is too hard, even though combat is nearly always optional, go figure.

[Eudoxus]

And (as dpeg from DCSS suggested to me) they both have a similar "arcade mentality"

It's probably not a coincidence that Linley Henzell went on to make some great shmups after writing crawl...