作者：Josh Bycer

过去几年里游戏开发的最大发展应该要数程序生成了。比起线性体验，现在你能够创造一些可以让玩家不断猜测的内容。像Rogue游戏，生存游戏喝模拟游戏都使用了程序生成去提升自己的重玩价值。一旦程序生成内容能够运行，你便拥有一款拥有无限重玩价值的游戏。但是就像游戏设计的其它元素一样，它也绝非百分百完美，如果设置失衡的话也有可能彻底摧毁游戏。

什么是程序生成？

从基本形式来看程序生成便是游戏中随机生成的进化。随机生成是在游戏并非基于相同顺序呈现内容的时候。这与战斗中的道具掉落一样简单。

随机设计是基于带有某种程度的延展性的硬性价值；如开启一个宝库以及获得一个道具。基于时间玩家可以看到游戏中所有随机元素，就好像它们都是预先设置好的。

而程序生成比这更进一步。即比起提供给游戏引擎或AI一列选择内容，设计师会为AI创造一个公式或一组指示。在这里AI将通过这些指示去创造某些内容同时也会使用一些随机与预设值。设计师所设计的公式将提供给AI创造一些架构，更重要的是将使其成为玩家可以战胜的对象。

程序生成的关键在于设计师不可能100%了解AI将创造出什么。因为是他们设计了公式，所以他们会有不错的想法，但却并非最完美的想法。

许多游戏都同时带有随机生成与程序生成，并且是基于不同程度。就像在《暗黑破坏神3》中，区域设计和结构便保持相对一致。战利品生成是基于程序架构，AI将创造出为了呈现特殊魔法和更高质量的工具的战利品列表。再一次地，战利品列表将为AI在创造工具时明确范围和可行变量。

《以撒的结合》将预设空间作为程序引擎的基础，并使用一些已经做好的空间将游戏关卡整合在了一起。

我知道虽然程序生成设计对游戏开发而言是锦上添花，但它同时也带有一些人和游戏都避免不了的局限性。

Spelunky(from gamasutra)

1.有限的基础

为了让程序生成可以有效运行，玩家与游戏的互动必须保持最基本。因为设计师必须围绕玩家所面对的每一个游戏机制和系统去设计公式。如玩家工具包中拥有更多机制或能力，设计师便更难去创造能够利用所有这些内容的公式。

这便是为何许多基于程序生成的游戏是依据道具而非玩家机制去设定复杂性。就像在《饥荒》中，玩家的互动方式便是使用道具，而这即是一种线性过程。程序生成还伴随着游戏世界是基于生物群落而创造并充斥着各种敌人和可互动对象。

因为有限的互动，游戏还会基于程序设计去限制重玩价值。

因为玩家知道他们真正的局限与能力，所以不管AI展现给他们什么他们都清楚如何游戏。当然了这也需要花费几个小时的时间并且需要玩家投入较高的技能，而这绝非是“无限的重玩性”。

2.无法处理独特性

如果玩家的移动集合越抽象，设计师便越难围绕着它去创造公式。《洞穴探险》在这方面便做得很好，因为AI很容易去处理一款平台游戏的规则和设计。一旦你清楚最大跳跃长度和高度，你便能够将这些限制条件递交给AI去创造关卡。

当我们开始谈论独特的机制时，我们也将尝试着划分程序设计。你不能使用《时空幻境》，《传送门1》或《传送门2》的时间机制去创造AI程序以呈现出真正有趣的关卡。

这时候的AI并不能基于横向思维进行编程。也许哪天当我们拥有一个意外的AI时这种情况会发生改变，但这可能还需要很长的一段时间。当回到第1点的时候你便会知道为什么有那么多程序生成游戏的复杂性被整合到环境中而非玩家的互动中了。

在像《Rimworld》，《Clockwork Empires》或《Dward Fortress》等游戏中，你可以提供给玩家生成顺序然后他们将按照开发者所设计的操作方式去执行它们。而互动方式也将锁定在这些顺序中。

比起添加全新系统或设计，开发者可以基于一些全新的互动内容去探索一些基本系统：提供给游戏更多深度，同时也不会增添玩家所需要面对的控制或系统。

《我的世界》便是另一个典例。从alpha测试到发行，开发者添加了许多内容到游戏中，但却从未改变过玩家与游戏的互动方式。游戏中大多数内容都是基于一个固定基础；这也将确保玩家在整个游戏体验中的控制保持不变。

你可以告诉我你的游戏拥有无界限的环境和关卡，但如果我每次进入游戏都在做着同样的事，那所有的这些便只是纸上谈兵了。

3.程序生成vs手工制作

对于那些不喜欢程序生成内容的人来说，程序生成内容永远不如手工制作的内容。

真正的慧眼都能辨别出程序生成关卡与手动创造的关卡。从本质来看，手动创造的关卡拥有自己的特性。这也是我们过了这么久仍记得马里奥世界1-1的原因。手工设计让设计师能够创造出他们真正想要创造的体验并能够去测试玩家。

而玩过程序生成关卡的玩家会发现这里总是充斥着一些根本不会发生任何事的死胡同。在一些质量较差的游戏中，程序引擎可能会将一些目标放置在一些奇怪的地方。你可能会遇到一个不能通向真正目的地的走廊，敌人可能会自己撞到墙上，更糟糕的是有些关卡还是不可战胜的。

这与《Souls》系列具有很大的不同：因为在这系列游戏中所有环境都是经过精心设计与设置的。而关于特性，《Souls》系列的确拥有特性，并且它也是手工设计优势的典型代表。

当着眼于大多数游戏是如何处理程序生成关卡时，我们总是能看到一些相同的情况。在手工制作的关卡中却不是如此，因为这些设计将拥有自己的特性去吸引玩家的注意。而在程序生成关卡中，游戏会逐渐变得重复起来。

不过这里也存在一个例外，即程序生成关卡比手工制作关卡更有趣的情况。即当我们在谈论环境是围绕着影响玩家在游戏中的看法进行创造的游戏的时候。通常情况下在战术和策略游戏中，线性环境总是会导致重复游戏。在《幽浮2》中，Firaxis便是基于AI程序生成环境障碍以及源自基本模板的细节去设计每个关卡的。

基于这种方法玩家将根据自己所面对的关卡设计而拥有一些不断改变的战术和想法。还有一个例子便是战术游戏《隐形公司》：在这里潜行游戏会受到环境生成内容的影响。

结论

现在我敢保证你肯定会好奇我喜欢怎样的程序生成游戏，我想说的是这样的游戏的确并不多。除了《洞穴探险》和《以撒的结合》外，我还喜欢Soldak Entertainment的ARPG游戏，它们的设计都具有复杂性和重玩价值。

在我写下这篇文章的时候最吸引我的一款游戏则是《No Man’s Sky》。这款游戏已经经过了很长的开发时间，而开发者也对此作出了巨大的承诺。

从行星到太空，这款游戏提供给了玩家一个程序生成的宇宙进行探险。

如果说有哪款游戏真正表现出了程序生成内容的优点或缺点，那定是《No Man’s Sky》。尽管我现在不怎么想玩这款游戏，但我仍会观察着它是如何带着程序生成设计不断发展的。

（本文为游戏邦/gamerboom.com编译，拒绝任何不保留版权的转发，如需转载请联系：游戏邦）

3 Failings of Procedurally Generated Game Design

by Josh Bycer

A major evolution of game development over the years would have to be the use of procedural generation. Instead of having a linear experience that is one-and-done, you can create something that always keeps the player guessing. Rogue-likes, survival and simulator games have been using procedural generation to extend their replayability. When it works, you have a game with almost unlimited replayability. However, as with all elements of game design, it’s not perfect and can hurt as much as help a game if not properly balanced.

What is Procedural Generation?

Procedural generation at its basic form is an evolution of random generation in games. Random generation is when a game does not present content in the same order every time. It can be as simple as what items drop during combat, to the order of level environments and more.

Random design is based on hard values with a degree of malleability; such as opening a treasure chest and getting one of X items. Given time, a player can see all the random elements of a game, as they are preset.

Procedural generation takes things a step further. Instead of just giving the game engine or AI a list of things to choose from, the designers create an algorithm or set of instructions for the AI to use. Here, the AI will create something from this set of instructions while using random and preset values. The algorithm is designed to give the AI’s creation some structure, and more importantly, make it something that the player can actually beat.

The most famous example and one I talked about before would be from Spelunky. Spelunky featured procedural generation to create a new level every time for the player. Designer Derek Yu talked about the creation and rules of the algorithm before.

Here, the game took the preset rules implemented by the designer, preset environments and obstacles, and creates something entirely original with them.

The key point of procedural generation is that the designer never 100% knows what the AI will create. They’ll have a pretty good idea because they designed the algorithm, but never a perfect one.

Many games feature examples of both random and procedural generation, with different degrees of both. In Diablo 3, area design and structure was kept relatively the same (not counting the rifts), but featured random events and placement of elite enemies. Loot generation was procedurally built, with the AI working off of the loot tables devised to create unique magic and higher quality gear. Again, the loot tables define the ranges and possible variables for the AI to use when creating the gear.

The Binding of Isaac used preset rooms as the foundation for its procedural engine, which stitched the levels together using the already made rooms.

With all that said, I know the design of procedural generation can add a lot to game development, but it does have several major limitations that no game can escape.

1: Limited Foundation

In order for procedural generation to work, the player’s own interaction with the game must be kept basic. The reason is that the algorithm has to be designed to work around every mechanic and system the player has access to. The more mechanics or abilities in the player’s toolkit, the harder it will be to create something that makes use of all of it.

This is why many games with procedural generation build the complexity horizontally in terms of items and not on the player mechanics themselves. In Don’t Starve for example, the player’s means of interacting are by using items which is a linear process every time. The procedural play comes in with how the world is built using biomes and filled with enemies and interactables.

Because of the limited interaction, it does actually put a limit on replayability with a procedural design.

Since the player knows their exact limits and abilities, they have a pretty good idea as to how to play the game no matter what the AI throws at them. Of course, this will still take many hours of play and a skilled player, but there really isn’t “unlimited replayability.”

2: Can’t Process Uniqueness

The more abstract the move set of the player is, the harder (to the point of impossible) it is to create an algorithm around it. Spelunky worked so well, because the rules and design of a platformer are easy (relatively speaking) to process by an AI. Once you know the max jump length and height, you can give those restrictions to the AI to build the levels around.

However, when we start talking about unique mechanics, this is when any attempt at procedural design falls apart. You can’t program an AI to make interesting levels using the time mechanics of Braid for instance, or the portals of Portal 1 and 2.

An AI at this point cannot be programmed around lateral thinking like that. Someday this could change when we have emergent AI, but that’s still a long time coming for game development. Going back to point one, this is why so many procedurally generated games’ complexity are built into the environment and not on the player’s interaction.

In a game like Rimworld, Clockwork Empires or Dwarf Fortress, you give generalized orders to your people who will then perform them based on the operation devised by the developer. Your means of interaction are locked to these orders.

Instead of adding in brand new systems or design, the developers can extrapolate out the base systems with new things to interact with: Giving more depth to the game, while not adding more control or systems to the player.

Minecraft is the other major example. The developers have added in tons of content from alpha to release, but not much that changes how the player interacts with things. The majority of content in the game is built off of the base foundation; keeping the player’s control the same throughout the experience.

You can tell me that your game has limitless environments and levels, but if I’m just doing the exact same thing every time, then all that is just window dressing.

3: Mass-Produced vs. Handcrafted

Another point and the biggie for a lot of people who don’t like procedurally generated content is the fact that procedurally generated content is always lacking behind handcrafted content.

A discerning eye can easily spot a procedurally generated level compared to a handcrafted one. Handcrafted levels by their very nature have a personality to them. There’s a reason why we still remember Mario world 1-1 after all this time. Handcrafted design allows the designer to create an experience exactly how they want it and test the player.

Anyone who has played games with procedurally generated levels can see how they’re usually filled with areas of dead space where nothing happens. In less quality games, the procedural engine can screw up and place objects in weird arrangements. You can have corridors that don’t lead anywhere, enemies stuck in walls, and the worst: A level that is literally unsolvable.

This is a huge far cry from games like the Souls series: Where Every. Single. Piece of environment was meticulously placed and designed. Returning to that point on personality, the Souls series has that in spades and is a crowning example of the advantage of handcrafted design.

Looking at how most games handle procedurally generated levels, despite the variety of possible designs, there’s always this sameness that you can see. In handcrafted levels its okay, because you have the personality and uniqueness of the design to keep you engaged. In procedurally generated levels, it starts to become very repetitive to play through.

There is one major exception to this and when procedurally generated levels are better than handcrafted. That is when we’re talking about games where the environment is built around affecting the player’s options in-game. Typically seen in tactical and strategy games, linear environments can lead to repetitive play. In XCOM 2, Firaxis designed each level to be built out of the AI procedurally generating environmental obstacles and details from basic templates.

This way, the player’s available options and tactics were always changing based on the level design they got; a huge improvement from XCOM EU and having preset maps. Another example would be the tactical game Invisible Inc: Where the stealth play was impacted by the environment generation.

The Sky’s the Limit:

At this point, I’m sure you’re wondering what procedurally-generated games I enjoy, and there aren’t many. Besides Spelunky and the Binding of Isaac, I’ve liked the ARPGs from Soldak Entertainment for the complexity and replayability of their designs.

The big game that’s coming out by the time of this post would be No Man’s Sky. The game has been in development for a long time and the developers have been promising big things with the game.

From planet-side to space, the game will supposedly give a procedurally generated universe to explore.

If there’s any game that will either truly show the merits of procedurally-generated content or fail it the most, it will be No Man’s Sky. And while I don’t have much interest in playing it as of right now, I’m still keeping an eye on it to see where things will go with procedurally generated design.(source:gamasutra)