You can download the mp3 of the podcast, play it in the embedded player below, or you can follow us on iTunes.
Some background info on the guests:
- Tom Betts (@nullpointer) has lectured in the arts, made In Ruins (cool 3D roguelike-ish procedural level explorer) and is currently working with Big Robot on the procedural stealth game Sir, You Are Being Hunted (which is being Kickstartered) whilst also doing a PhD on the sublime in video games
- Michael Cook (@mtrc) is doing a PhD at Imperial College London's Computational Creativity Group and has been working on an AI game developer called ANGELINA
- Ian Horswill is Associate Professor of Computer Science at Northwestern University with interests in AI, PCG and interactive narrative
- Leif Foged is Team Leader of the Creative Arts and Technology Studio at Northwestern Univeristy and works with Ian Horswill, and created procedural dungeon crawler Trial by Spire
Topics discussed in this episode include:
- Constraint propogation as a procedural tool
- Expanding relevance of academic game design study
- Freedom of study in academia
- Obstructions to good academic research
- Communicating academic research to a wider game development audience
- Producing real applicable results from academia
- How advances in AI research are failing to make it into the games industry
- Lack of innovation in the mainstream to take advantage of academic research
- Open source in academia and willingness of academics to release their code
- Availability of academic papers
- The need for people to contribute more to the Procedural Generation Wiki - you heard me, listeners!
- Potentials for academics and indies to work together, or independents and amateurs to access academic work
- The quest for procedural narratives
- The difficulty/intimidation of approaching procedural generation for those uninitiated
- Forums for discussing procedural generation and design
- Events in academia on game design, such as Imperial College London's GaME
Interesting links and academic papers:
- Fast Procedural Level Population with Playability Constraints (Ian and Leif's paper) - and here is the relevant code, which can expand to other games easily
- A Procedural Procedural Level Generator Generator
- Incorporating Required Structure into Tiles (procedural map generator study)
- Search-Based Procedural Content Generation
- Artificial Intelligence and Interactive Digital Entertainment 2012
- Toward Supporting Stories with Procedurally Generated Game Worlds
- Electron Dance series on The Academics Are Coming
- Procedural Content Generation Wiki
- AIGameDev.com
- StackOverflow
- Procedural Content Generation on Google Groups
- Foundation of Digital Games conference
- Digital Games Research Association
- PCG Mendeley Group
- 'TED talk' that Andrew mentions on show don't tell programming (and some further links)
Join us next week for discussion of game jams and jamming (strategy games will come the week after).
Thanks again guys! Found this very interesting, and will engulf myself in some papers now...
ReplyDeleteI heard a talk from Luke Dicken a little while ago, and he had some very interesting things to say as well within the scope of AI, games and use of procedural techniques:
http://lukedicken.com/
Very interesting stuff. As someone in academia and roguelike design (though the two are unrelated; my doctorate is science policy/industrial policy) I've actually been considering doing something to combine the two 'on the side' as well as my PhD. A big part of what I'm doing is detailed generative art design (from the next release in a few days onwards, anyway), which is something I've never really seen in roguelikes, though obviously that's not the only randomized part. I think there's a lot to be said about roguelikes academically, and not just on game design but on investment and challenge/difficulty/reward in games, and I'm keen to try and contribute there in the near future (and to try to deal with some of the issues mentioned like the freedom of academic papers, etc; many of the (few) game design journals out there don't need academic logins, though obviously anything game-related on larger journals, sadly, does)...
ReplyDeleteHere's another relevant paper, for any interested:
ReplyDeletehttp://www.academia.edu/1629350/Procedural_Generation_of_Narrative_Puzzles_in_Adventure_Games_The_Puzzle-Dice_System
Great episode! I've been inspired more and more by the recent episodes to get stuck in and make something from ideas spun out of the wistful lamentations of the RR regulars, and this one had me in a whole new world of itchiness to finish the mowing and get back to the PC for more reading. Constraint propagation looks ideally matched to my own pet project, and more PCG Wiki talk and opportunities for collaboration seem to fuel my desire to just get in and do it. Hopefully others are similarly inspired and we have an even bigger 7DRL to delight over.
ReplyDeleteI've been reading the paper and the code for the constraint based level population. It's been a revelation, thank you very much for this!
ReplyDeleteI got interested in this kind of work via this article on using Answer Set Programming (ASP) to generate dungeons.
The "Fast Procedural Level Population with Playability Constraints" paper introduces intervals as specific domain types, would it be possible to do something similar with grids? The ASP dungeon system introduces enormous numbers of constraints to construct the grid which no doubt makes it so slow for realistically sized levels.
Fascinating episode, guys.
ReplyDeleteIt's good to know my architecture talk at the IRDC made an impression! In building design, the complexity that Darren mentions is as much a driver as it is a goal. We need to do a lot of those calculations anyway - even if by hand - in order to make sure that the building we're designing is going to fulfill certain criteria (like not falling down!).
Over the years we've developed software to perform most of the tedious analysis stuff for us - the next logical step is to get the computer to be able to react to the results of those analysis tasks and adjust the design of the building accordingly. The tricky part is not so much the complicated analysis bits (which computers are good at) as knowing how to improve the design based on that information (which computers are not good at). An experienced engineer can look at a stress map of a building and immediately know how to change the structure in order to make it more efficient, but a computer can't (yet!) because it's looking at the problem in an entirely different (and far more complex) way.
I would actually say that the aim is to reduce the complexity, not increase it! At the moment I'm developing a tool to procedurally generate Sports Stadium seating bowls. Good bowl design is all about spectator experience - how well you can see, how close you are to the action, how comfortable you are and so on. Now, you could try and measure that by doing loads of fancy-pants lighting and acoustic and view analysis, but it would take all day and generally we don't need to (at least at first) because we have guys who have been designing sports stadia for 30 years and have developed a pretty good feel for that sort of thing. A lot of what I'm trying to do is actually just talk to these people, try to extract some of this knowledge and experience and boil it down into heuristics that the computer can then understand and use. The constraint based stuff looks like it might be useful for that sort of thing; I may have to bother Ian and Leif about that...
Regarding the 'using a more complex method when a simple one would do' thing: I cannot count the number of times that I have been reading through an academic paper and thought: 'huh. Couldn't they have got the same result in this other, much simpler, more efficent way?' and then had to re-read the proceeding two pages a couple of times on the assumption that there must be a good reason behind it and that I was just being too dense to see it. While on the one hand I'm pleased I'm not quite as moronic as I thought I might be, I suspect this may be one major reason why there's not more implementation of academic research!
Great episode.
ReplyDeleteIn the defense of arcane formulas brimming with Greek letters: formal notation is necessary for many proofs. Conveying a proof by induction using prose rather than formal notation can take a considerable amount of additional paper. Still, it's something that can be used poorly & ruin readability.
Regarding access to subscription only journals: my university provides library patrons (including those not enrolled or employed by the school) with electronic access to journals. It does require a physical trip to said library & I don't know how common this service is at other institutions, but it may provide some with an additional route to otherwise inaccessible information.
This comment has been removed by a blog administrator.
ReplyDelete