Some of the biggest games now in development have begun embracing procedural generation. Last month’s E3 played host to Frontier’s space sim Elite: Dangerous and Hello Games’ No Man’s Sky, surprising attendees with their ambitious plans and enormous universes.
In fact Hello Games’ Sean Murray said his studio’s game, created by a small team, would take place in an entirely procedural universe that was “infinite” in size.
“Even we don’t know what’s out there,” he claimed while on stage.
The technique is used to automatically build levels and environments based on a set of rules and a plethora of assets, but can also be used to randomly generate items, fauna and other gameplay elements. Despite the random factor, however, Frontier CEO David Braben says the game itself should not feel completely randomised.
“If it does feel random, then it is bad procedural generation,” he states. “Imagine creating the 3D texture of tree bark up close – it should look just like the bark of a real tree.
“How it was created shouldn’t matter – whether from some complex procedural technique mimicking the way the bark formed in real life, or an artist, carefully sculpting the peaks and troughs to match real life.
“If the results of both approaches are visually indistinguishable, then the first method should be better, as you can then create as many differently curved tree trunks as you like, whereas an artist would have to, largely, start again. It is all about rules – and how to express them so that a procedural system can conform to them.”
During the last few years in particular, procedural generation has proved immensely popular among indie developers, and is also being used in part by the larger, triple-A studios for certain aspects of their titles.
But why is it such a hit with some small developers over more controlled experiences?
“While big developers can afford to build specialised level design tools, this often isn’t feasible for indies, so procedural generation can be an interesting option as an alternative to time-consuming manually crafted levels,” says Serious Brew developer Maarten Brouwer, an expert in procedural generation following his work on Cargo Commander.
“It also offers the possibility of seemingly unlimited playtime based on a limited amount of assets.”
Kitfox Games creative director Tanya Short, who has worked on games including turn-based RPG Shattered Planet, agrees that procedural generation is a smart business move for many indies, who are to some degree competing with big publishers. She adds that the technique offers a novelty and design aspect that many are keen to explore further, in turn generating fresh experiences.
“In short, we like new things. We like puzzles and problem solving and even a bit of risky adventure – and what’s more of an interesting puzzle than procedural generation?,” she says.
“It’s a puzzle within a puzzle, with a new challenge revealing itself with every playthrough. There’s a reason Rogue and Dwarf Fortress are 20-year projects. Procedural generation is a black hole of design, ready to eat your life. But in a good way.”
Short paints a picture of the process of procedural generation as one that is actually tightly managed. She explains that the team first hand-builds an example level in the editor – in her case Unity – and then analyses the various elements that make each level fun.
After this, developers can break down the design into rules, deciding how many tiles wide a room and corridor should be at minimum and maximum levels, how treasure and enemies should be placed and even what kind of decorative elements need to be placed and how, among other considerations.
“You usually think of procedural generation as writing on a blank page – drawing lines on graph paper,” she says.
“But really, it’s more like taming this wild infinite possibility where literally anything can happen, and limiting it to only a certain subsection of things that can happen.”
But when creating such a set of rules, how can developers stop a game that can essentially create levels itself from becoming boring and repetitive for players? Is there an issue that a lack of control can, at least to some users, result in an unintentionally monotonous experience?
Short recommends that to avoid a set of rules making a game “bland and samey”, as the algorithm will do exactly what a developer tells it, devs should create some exceptions to a rule.
Brouwer says part of the fun in procedural generation lies in the lack of control, and admits he often gets tired of the carefully designed events of more traditionally developed games.
“It can be much more exciting to play something that I know is unique for me, and of which the possibilities and outcomes aren’t fixed at all,” he explains. “As a designer, it’s the trick to find out what parts to control and what parts to let go, without levels getting too hard (or impossible) or too boring.
“The game mechanics matter a lot; it’s much easier if they are fun regardless of the context or environment. In that regard, game-influencing procedural generation is not something to ‘tack on’ or to simply replace manually designed content; the game really should be built around it.”
One of the key benefits touted for procedural generation is the cutting of costs, even for triple-A games. As with Hello Games, Sean Murray’s small team has been able to create what, on the face of it at least, looks like an enormous universe. But it can also be used to take care of other elements of development.
Kevin Meredith of Interactive Data Visualization, the company behind vegetation modeling and rendering tool SpeedTree, believes such tools can dramatically reduce the time required to model and place assets.
“This lets artists spend more time on other parts of the game, as well as get the game done more quickly and at higher quality,” he says. “So not only are there time and money savings in procedural development but also the increased revenue that results from a better game launched sooner.”
Brouwer feels the technique is already widely used to minimise certain level editing tasks, such as the aforementioned terrain vegetation and dirt decals, as well as enemy spawns and loot drops.
He warns though that procedural generation should not be always be the go-to to simply replace manually crafted content or create 100 hours worth of gameplay.
“Usually it takes a lot of time and effort to work, and requires changes in the entire game design to work well,” he explains.
“When players have encountered too many games with cheap implementations, they will stop seeing procedural generation as an interesting quality and start equating it with bland repetitiveness. But as a core part of a game it can not only lower costs, but make it possible to create really new experiences.”
But Braben believes procedural generation will not cut costs, at least in the bigger, triple-A titles. In fact, he says it could result in the opposite.
“Ironically there is a danger that it could cause an increase in costs, for triple-A games at least, over time because frequently such game development is constrained by limits; system memory, disc space, render performance, network bandwidth and so on, and the best game is the one that makes the best use of those limits,” says Braben.
“Procedural generation enables a great deal more content to be included in the same memory footprint and bus bandwidth, hence requiring more art time, and so greater cost.”
Whether it can truly cut costs, and that seems dependent on scale and use-case, there’s an exciting future ahead for procedural generation in games. And with more indie developers getting to grips with the tools and techniques behind it, who knows what implementations we might see in the next few years?