A small teaser to get you to read the entire article

The neural net got lucky this time.

Graphics

A game is game, and the gameplay is what makes most of the game. But, for me at least, graphics are also really important. I am not talking about the latest shaders with deferred lighting and HDR scattering and pimples in the faces of the characters. I am talking about style, about art. If the graphics in a game have a certain style that appeals to me, it doesn't need to look as realistic as a movie. I hope you all like the style of Mad Skills Motocross.

Riders and bikes 

The graphics are being made by the excellent David Ferriz. He has already delivered a lot of the assets for the game. Just look at this rider:

Nice, huh? All those combine into something like this:  

And it looks simply gorgeous in the game when all those parts are moving according to the physics. Well, you've seen it in the video above. I have implemented a nice system that enables me to place any image on any part of the physics objects with a simple XML file. This makes it very easy to skin the bikes and riders and I hope some of you will enjoy that as much as I do. There will be five different riders and five different bikes in the game, and I hope that the community will create even more.

Backgrounds 

Ferriz has also delivered two backgrounds so far, one of them can be seen in the videos and it looks like this: 

A static image showing one third of the background standing still does not do it justice, as you can see in the videos the parallax scrolling really gives the player a sense of depth.

Here is a teaser screenshot of the other background in action: 

Logo

Logos are good right? You've got to have logos. Below is the logo for Mad Skills Motocross overlayed on some tracks in the sand. What do you think? I quite like it.

Music

My good friend Stefan agreed to make the music for Mad Skills Motocross, and I couldn't be happier about that. He is an excellent musician and can make music that really fits this game: hard, fast, and retro. Just listen to this track, which will be played during a race:

Can you smell the gasoline?

The Neural Nets

Remember the previous diary entry? Or did you force it out of your mind because it gave you nightmares thinking about giant computer brains taking over the world?

Anyway, my brains are coming along nicely. In the videos in this article, the upper motocross is driven by a neural net. But it kind of sucks at driving right now because it was trained before I tweaked the physics once more. However, before the tweaks it was an ok opponent. The only training it got was me playing for 4 minutes, but 4 000 times over (I didn't actually play 4 000 times, I recorded 4 minutes of play and fed it to the network over and over). 

This proves that the concept works, and that I have constructued a brain capable of learning the task. I just need more data, lots more data. I got a brilliant idea:

1. Record the races of the beta testers to a file
2. Have them send the file to me
   
3. Train the neural networks with all that data 

This will give me lots of possibilities when training the networks. With that much data I can choose the recordings which resulted in excellent times and train an über-brain. I can choose the less good recordings to train a brain for the easier stages of the game. I can also train a brain that likes to do back-flips.

It's going to be great.

For the technically inclined the neural net I have had the most success with consists of three layers: input, hidden, and output. The input layer consists of 29 neurons with linear trigger functions, the hidden layer has 14 neurons with sigmoidal trigger functions, and the output layer has 4 neurons with sigmoidal trigger functions. The learning algorithm I use is called Resilient Backpropagation, or RPROP, and it outperforms anything else I have tried by a large margin.

Final video

In closing

That's all for now, please discuss and ask questions in the forums, and I will try to answer them as best I can. Check back October 1st for the next installment of (suspenseful music) Turborilla Development Diary, same channel, same time.

Take care
// Tobias
Turborilla Entertainment

These diary entries are here for you, the readers. So I would like to know what you think, what you would like to see in the future, and any ideas and questions you might have really.

I have created a forum topic just for this diary entry: discuss here

Get the developer diary in your favourite news reader