As before, here’s a YouTube playthrough of the level:

And if you’d like to play it yourself, you can find it here.

There won’t be any more games for a while as I need to take two days’ holiday, finish my paper, present at GaME – all in the next week! There should be more games up around the end of the month though.

It’s not up on the Games page yet but I’ll be doing that in early July when I’m over AIIDE, GaME and ICCC 2012…

I mentioned that I’m writing a paper about the latest edition of ANGELINA, but I haven’t really explained what it is that ANGELINA can do that’s different from a few months ago. It’s been very changeable up until about a week ago, but I’m finally producing games and getting reasonable results. This post is about what new capabilities ANGELINA has gained over the last few months, so that the games make the most sense when I post them soon.

I’ve already spoken a little bit about my intentions with ANGELINA right now, but the gist is this – ANGELINA can now read a news article on The Guardian’s website, go out into the web and retrieve images, sound effects and music, and arrange them as part of its game design process to create a playable game with a (usually appropriate) visual and aural theme. That’s not to say it produces September 12th-level masterpieces of newsgaming, because that’s not necessarily the aim. The aim is simply to show that ANGELINA can produce a game with a defined theme, and look at some of the issues surrounding that.

The games are rougher and readier than the previous ones we released, because ANGELINA is taking over more control in the design process. That means taking a hit in quality in return for a stronger, more interesting AI process behind the whole thing, which I’m very happy to do. My hope is that this will develop into a system that can read and expand on a simple theme and create an appropriate game to suit – the sort of system that could have taken the recent Ludum Dare‘s theme of ‘Small World’ and actually gone out and made something meaningful on that theme.

As you’ll see when I post two of ANGELINA’s games next week, we are a long, long way off that. But not as far as we were back in January, which is the encouraging thing about research projects. Slow, steady progress, with bits of excitement in between.

Once the paper’s been done and submitted ANGELINA will continue to publish games based on current news articles. Hopefully over time you’ll see the quality (and the independence of ANGELINA) increase, and perhaps we’ll even be surprised from time to time at what is produced. I really hope so.

There’s a lot more to say about the current batch of games and the state of ANGELINA’s art right now, but that can wait until after the paper. The games should go up early next week.

This all came out of a need to generate good titles for the games ANGELINA’s making now. Since they’re all much more distinct than before (as you’ll see soon) I wanted the titles to stand out too. I tried lots of different things, including a system similar to that which named Revenge and After Squad, but nothing really clicked. Finally, I had an idea for a simple system that would give ANGELINA some really rudimentary punning abilities. Here’s some examples:

Sitting On A Powder Clegg
Starting word: Clegg (Nick Clegg is the Deputy Prime Minister of the UK)

Dangerous Koreaisons
Starting word: Korea

Back To Square Pun
Starting word: Pun

Dance Dance Retribution
Starting word: Retribution

They’re all very basic, and nothing you couldn’t come up with yourself (except perhaps the Korea one, which I thought was quite inventive) but they get the job done and it’s very flexible. It’s also simple to boot:

1. Use online rhyming dictionaries to look up a catalog of perfect or near-rhymes for the target word.
2. Sift through corpora of phrases, games, films and song titles to look for matches.
3. Substitute a rhyming match for the original word.

Of course, with a system this basic, you also get a lot of duds. ANGELINA won’t be coming out with winners all the time right now, but what I’m hoping is that future versions (post-AIIDE) might be able to choose an appropriate substitution where the meaning is understood by ANGELINA and explicitly selected. For instance, the Clegg pun has its root in the phrase Sitting On A Powder Keg, which implies a dangerous situation that might explode at any moment. If we can find a way for ANGELINA to unpick certain phrases like this, even if we restrict ourselves to very simple ones, we can give the system a much higher-level creative control. That could be quite cool indeed.

ANGELINA has other methods for choosing a title for the game too, which it uses from time to time, mostly when it can’t find a good match in the rhyming dictionaries. More on this, and the games themselves, soon!

Below is an image taken from the paper Automated Collage Generation – With Intent, written by my supervisor Simon Colton and a then-Masters student Anna Krzeczkowska (as well as contributions from Cambridge researchers Jad El-Hage and Stephen Clark). The collage was made by software that read a Guardian news article about the Afghanistan war, analysed the text for the most relevant words, and arranged Google Image results onto a canvas. This is a favourite example of the group because the images chosen are so apt – bombers, war graves, Afghani children.

It’s not the most elegant arrangement of images, but it’s powerful and entirely novel to the authors of the software. There was no way of telling what Guardian article the software would read (it chose whatever was being written about the day it ran) or the images retrieved. That, to me, is a really exciting bit of AI, producing something interesting to look at. I want some of that in ANGELINA.

Over the past week or two I’ve put into ANGELINA a number of tools for extraction of interesting media sources, including most prominently the same Guardian news sources used by Anna’s software. I’ve added a few tweaks here or there, and made the content retrieval richer using a few techniques. I’ve just run the software (as I’m writing this post!) and it’s chosen an article off the Guardian titled:

François Hollande on top but far right scores record result in French election

ANGELINA managed to deduce that President Sarkozy was a living person (although amusingly, the current techniques I’m using did not detect the same for Francois Hollande!) and even concluded a negative emotional response towards Sarkozy from the public. Here’s some images the software pulled back as a result, first generally about the news story:

That’s everyone’s favourite French extremist-with-a-PR-budget, Marine Le Pen. As a bonus, because ANGELINA knows Sarkozy is a person, it can go and get us some extra material, so it retrieved this too:

The question remains how to use this media most effectively (and some of the retrieval algorithms need more work to get them to bring back more interesting/important results) but the basics are there, and hopefully this will build into something worth publishing over the next few weeks. I’ll keep you posted as to how that goes.

I’m taking the briefest of breaks and then I’m back to work. With a bit of luck we’ll be submitting some brand new ideas to the Computational Intelligence in Games conference, whose deadline is at the end of this month. Of course I’ll keep you updated on what those ideas are going to be, exactly, as well as game examples as and when they become available. If you’re on Twitter, now would be a great time to follow @gamesbyangelina, as there’s a good chance I’ll be taking advantage of the system for the first time in the coming weeks, in order to get live, human feedback on things that ANGELINA might not be totally sure of herself (like colour schemes).

Thanks to everyone who has read, commented or gotten in touch with me over the past few months. Everyone’s contributions and suggestions improve the system and help us put out our research. Thank you so much!

EvoGames, the games workshop in the conference, had some great talks (and a couple of very interesting posters) and also let me meet up with some of my favourite researchers too. We saw talks about adaptive game opponents, automatic camera shot composition for cutscenes, and ANGELINA featured in there somewhere too. If you’re interested in seeing my slides (unfortunately without the nice videos I slipped in) you can get the PDF of the slides here.

Hello to everyone I met at EvoStar this year! I had a great time. Back to ANGELINA from Sunday!

For those interested in APACE – it’s really not very functional right now. It needs better tools for creating walkable areas and placing items, which I don’t have in place yet. It’s also entirely undocumented. It’s really there right now in case someone wants to peek at some example code or hack it into shape themselves. I will be adding to it in my spare time though as I like the idea.

ANGELINA on GitHub

I’ll be posting more about these elements as I’ve got concrete things to show, but one thing that’s quite fun to show off is some work I did on recolouring sprites based on a given colour palette. Little work exists on this problem that I could find (one paper looked at recolouring images based on the colourings of other images, but this is different as we only have a palette, rather than information about which colours should be dominant, etc.).

Being able to dynamically recolour sprites gives ANGELINA more flexibility in designing the visuals of the games it produces. It’ll also be a key step in adding flavour to a game design, by choosing appropriate palettes and automatically updating the sprites to reflect the new theme. With that in mind, below are a few different methods for recolouring the sprite from the New Scientist game. Let me know which method you think is best (obviously this is a very limited selection, but still worth considering). The two palettes used are Trick or Treat and An Enchanted Forest. Here’s the original sprite:

The first three methods I’m going to show take the original image and look at the relative distances between the colours in HSV space. They then attempt to find a mapping that minimises the change of these relative distances according to one of three metrics – Hue, Saturation or Variance.

Here’s ValuePriority-Hue

ValuePriority-Saturation

ValuePriority-Value (what a great naming scheme this was, eh?)

I also wrote some code that attempted to use simulated annealing, with the aim of producing a solution that minimised distance between colours in the Lab’76 space (it’s a more complex representation of colour, but one that apparently has a real basis in human perception). Here’s the results from that:

 Finally, my supervisor has done extensive work in art and visual tech himself, and he offered up another method. In this method, we map colours in the original image to whichever colour is closest in HSV space, and then set the Value component of the new pixel to its original setting. This changes the hue and saturation of the pixel but retains the relative visual detail. As you can see, it’s by far the most stable of the recolourings, but the loss of colour information detracts from the effect of the palette, and the effect isn’t as pronounced (this is partly because the method doesn’t guarantee use of all palette colours).

Let me know what you think of the various approaches! There are a few ways left to try (as you can see, some palettes might give us extra information, like the Enchanted Forest palette which attempts to ratio the palette colours). But we can’t always rely on such information being there.

]]> http://www.gamesbyangelina.org/?feed=rss2&p=129 6 http://www.gamesbyangelina.org/?p=121 http://www.gamesbyangelina.org/?p=121#comments Fri, 09 Mar 2012 16:46:43 +0000 Mike http://www.gamesbyangelina.org/?p=121 A lot of very mad things happened in the past 24 hours, most of which I’m trying to document on ANGELINA’s ‘In The Press’ page, but in short – Engadget, Kotaku, The Verge, New Scientist… the list goes on. Lots of people have been talking about ANGELINA, and I’ve been getting a huge amount of traffic over here. I just wanted to spend a few hundred words explaining why this matters to people like me, and why I’m so grateful when things like this happen.

First, to add some context, here’s the last thirty days of traffic to this site:

Normally, this site would get maybe ten hits a day, and most of those were residual from one feature on The Sunday Papers over at Rock, Paper, Shotgun, and another larger feature over on Develop Magazine’s website. Of those ten hits, most wouldn’t get past the front page. In the last few days, I’ve amassed 14,500 visitors – and those are just the ones clicking through the news stories, many more have read the articles. So it’s clear that journalists drove a lot of traffic here recently. But why does that matter?

As a scientist, I’m funded by public money to perform research that is ultimately offered up to everyone. The reality is that information tends to trickle down from academic conferences, to the mainstream industry, to indies and the public if you’re very lucky. Journalists have the ability to invert this process, bringing scientific topics directly to the public and engaging and inspiring them with things that are commonplace and really tangible – videogames.

Scientists aren’t encourage to engage with the public much at all. It’s not linked to promotion, assessment at any level, it’s barely linked to the performance of a department or university. This is a problem, because a lot of research only makes sense when people are able to see it, play with it and benefit from it. One of the reasons I run this blog is because I want people to be able to see how this project is working, how slow progress is sometimes and how fun the results can be to see. Projects like Looks Like Science and I’m A Scientist Get Me Out Of Here exist because we’re out of touch with how science works and what scientists do. This is particularly true of computer scientists (in fact, I’m not sure we were ever in touch with the public).

Science, tech and games journalists are the gatekeepers that provide us an opportunity to talk to people about cutting edge ideas, new techniques and experimental projects. And, as I think ANGELINA might have at least hinted at these past two days, we can make great headlines too. There are dozens of undiscovered games projects across the world, waiting for the right journalist to give them a shot. If you want me to throw a handful of ideas your way, email me – mike@gamesbyangelina.org – or come along to open engagement events like Imperial College’s Games and Media Event.

Thanks to everyone who covered ANGELINA these past few days (or those writing up now, which I’m super-excited about). Drop me a line whenever if you want to know how ANGELINA is getting along.