As the industry innovates and games improve, the challenges that AI programmers face also evolve significantly. Issues like scaling up content creation, modernising gameplay for casual audiences, dealing with procedural levels, managing large scale emergent AI, leveraging and improving existing codebases, etc.
With this year's Game/AI Conference preparations in full swing, we thought it'd be a great opportunity to pick the brains of some of our amazing speakers to see the major challenges they've been dealing with on their projects. We asked each speaker the following question:
“What were the biggest challenges you faced
while working on the AI for your project?”
The replies we got cover games including Rocksmith, Crysis 3, Warframe, EverQuest Next, and Creatures 3. Read on for more...
Daniel Brewer on Warframe
“One of biggest challenges we faced with the AI in Warframe is dealing with the procedural levels. We don’t know ahead of time what the level layout is going to be and so staging engaging encounters can be difficult. Getting the pacing right on the myriad different levels is really tough for the AI-Director. The NPCs also have to be very systemic so that they will behave appropriately throughout all the environments. We can’t always rely on markups for specific enemies.
The other big challenge is dealing with players of wildly varying capabilities. Players gain more power as they progress and level up their characters and change the load-out of mods on their gear. This is awesome for players as they can really push how powerful their characters are, but there is a huge range of power-levels for different weapons and mod combinations - sometimes a couple orders of magnitude! This makes balancing the game and boss-encounters extremely difficult and a constant challenge for us.”
Julien Hamaide on Creatures 3
“As we had to work with an existing AI, the biggest challenge was the introduction of modern gameplay without loosing the very nature of Creatures AI. As we discovered some hidden features, we had to adapt the game design and some technical solutions to keep the features that made the series famous. As the AI is also highly technical, game designers had issues understanding the underlying limitations (e.g. with genetic code).”
Nicholas Bonardi on Rocksmith
“Rocksmith’s Session Mode uses AI instruments to listen and follow along with the player in real time. In Session Mode, the Conductor is additional layer of AI that applies the mechanics of music theory against the instrument content in real time to make sure everyone plays harmoniously.
Creating content for instruments that would work in any permutation of the Conductor system was a tremendous challenge. The problem was more than just hitting the right notes at the right time. These instruments needed style, personality, and most importantly, to sound human. Finding the balance between personality and malleability was taken on an instrument by instrument basis.”
Stéphane Bura on Everquest Next
“I'd say that all the challenges were linked to the same issue: the holistic nature of the design (gameplay/experience/tech/architecture) process, as the storytelling AI touches everything.”
A lot has been written about the emergent AI for Sony's ambitious MMORPG. If you'd like to find out more about the solutions to these challenges, join Stéphane at the conference!
Francesco Roccucci on Crysis 3
“Crysis 3 has been a really ambitious project, especially comparing the scope and the time we have been assigned to ultimate the game. In that given timeframe we had to improve the systems used during Crysis 2 to increase the quality of the behaviors while keeping them easy to maintain and to extend.
Another big challenge has been the creation of Human AI agents that can behave as much as possible as real humans without ending up into situation that may break the immersive experience of the players into the game world.”
Ben Sunshine-Hill on Havok AI
“The biggest challenge of developing Havok AI's traversal analysis system was making it robust. Floating point precision becomes a major concern when implementing the primitive geometric queries underlying such a system, and with the huge variety of level data being used with Havok AI, any weakness in those queries would be nearly certain to cause a problem for someone.
At the same time, I had to be careful not to sacrifice performance. On some platforms, adding a single bit of precision to a floating point operation can nearly double the cost of that operation, so it was important to be very aware of how precision issues were likely to propagate through the system.”
Tim Gosling and Piotr Andruszkiewicz on Total War: Rome 2
Tim: “First, the TW Campaign AI evolves from product to product but with Rome II it required a significant overhaul to several key systems in order to manage new features and deal with the larger scope involved.
Second, maintaining and improving the AI efficacy and run time performance while doing this within the product time line was challenging.”
Piotr: “One of the biggest challenges we encountered was achieving consistent performance without sacrificing AI quality. There are over 100 factions in TW:R2 and each of them needs to make a substantial number of complex decisions every turn.
Most of those decision involve pathfinding of some sort which can be very time consuming so we had to use a number of schemes to reduce the burden - we cache as many query results as possible, use approximate pathfinding solutions whenever possible and use pre-computed influence maps for threat detection and target filtering.”
Join us on July 8th and 9th for many more insights into these games and solutions to these challenges. Tickets for the conference are still available, including options at €49 for students or €99 for indies.