I have to pick a solution from the spectrum of solutions which ranges from the practical to the accurate. Most space games invariably choose the practical solutions, which is, typically, having an unchanging, logical and scaled-down representation of the universe. Scales aren’t represented properly; entities may have orbits, but these are fixed; and, finally, entities have a very logical, typically parent-child, relationship, where one would typically enforce planets to be associated with a star, stars with a galaxy, etc. Such models only represent the universe in its most general terms. I want to be able to show true scales, irregular celestial structures (rogue planets, binary/multiple star systems and so on) and a greater degree of dynamics (changing orbits, colliding entities), but I can accept a loss of precision as long as it looks believable.

Scene Representation

To represent the systems in graphics space, I’m experimenting with using a tree of irregular coordinate spaces, where each child node has to be a subdivision of its parent. Each node is then its own coordinate space which is a scaled up space of its parent. It has a defined boundary for each axis (a box) and can be freely positioned and oriented. Each celestial body would then occupy a coordinate space that is appropriate for it. The camera would also occupy a node in the tree, but at the scale where 1 mm can be accurately represented. Some rudimentary testing I did a while ago indicated that I could manage this requirement using a coordinate space that’s bounded to 10,000 units (equiv. to 10 km, if 1 unit = 1 m). To be on the safe side, I’ve bounded it to 1,000 units. The camera then asks the tree to move it to a node that has a boundary that contains its position and has the required scale. Obviously, there will be missing nodes in-between with our requirements (we have to divide several times if we are to go from a node with galaxy-wide boundary down to one that has boundaries acceptable for a camera). These are created dynamically in a grid-like manner and kept track of using a hash map (indexed by its integer division “position”) in the parent node. When rendering a node, we simply move upwards in the coordinate spaces (from the camera node) to the common node and then down to the celestial entity of interest.

Logical Representation

Each celestial entity is represented by a CelestialEntity object, which contains rendering details (models, etc), physical details and a set of references to child entities. This set is used to generate the local OrbitalPhysics object which determines the physical behavior of the system. A CelestialView provides the scene graph with the minimum amount of data needed to accurately render the universe for a specified camera and settings. Some of these settings relate to scale compression, and I’m experimenting with lerping to logarithmic scale to provide better overview of a solar system.

Simulation

As mentioned before, the OrbitalPhysics object determines how the celestial entities move. Early on, I tried to use PhysX to simulate orbits and had a lot of skepticism about whether it would work well. After scaling down all the values to make them tolerable to PhysX, it turned out this skepticism was warranted. As I recall, orbits decayed very fast and it had nowhere the stability I needed. I’m fairly sure PhysX uses a low-order integrator. So, I looked into how Celestia, which is the first program of the top of my head that performs scale-accurate rendering and animation of the solar system, handles these physics and as I understand it they use analytical solutions limited to our solar system (VSOP87) for determining positions. No help there, then. Then I found a good paper on solving N-Body problems using multi-step methods. I decided to go with the 9th order position-acceleration method which is the one he has used himself for his simulations, and am currently using leapfrog to get the initial steps. This had satisfactory performance.

For orbits, I sort all the orbital periods, evaluate celestial states within that period using appropriate time step size given by a division of the lowest orbital period, then reset to the optimal time step size for the second lowest orbital period and evaluate celestial states from the current time index, and so on. The entities will then have a collection of points in their orbits with equal or increasing distance in time between them compared to the previous, but never more than the optimal time step for each body within a single orbit.

Caveats

The scene representation has some issues with floating point precision which is exacerbated by the fact that matrices can be oriented freely, so there are occurrences where a node might not fit into a child node that’s located at the edge of its parent, but when transforming it into parent space, it fits into the parent’s boundary, so, logically, it must fit in the child and the whole thing barfs. I currently use some hacks (fuzzy boundaries, etc) to get around it, but I still want to figure out a better solution.

I am struggling somewhat with rendering accurate orbits, because I want to use greater time steps for the orbits and in some cases I’m getting wildly changing orbit data from frame to frame. I’m testing them on binary stars moving elliptically, and suspect it’s because of the very tight and steep gradients when they’re closest. The 9th order method handles it fine, but it seems like the leapfrog is giving very inaccurate values. So, when the stars are close and I have to recompute the orbit, it starts getting the first 8 values using leapfrog and the 9th order method doesn’t get accurate enough values to begin with. I might try some more advanced single-step methods instead of leapfrog for initialization, but I’m leaning towards trying to reuse some of the previous states instead which seems somewhat nontrivial.

Another issue is estimating orbital periods to determine optimal time step size for orbit computation. Currently, I’m simply assuming a circular orbit (using current velocity and acceleration),  calculating the period of that and adjusting it based on previous orbit data, but it only works as a very rough estimate.

Uploaded new version of GameKiller (original thread) which was initially designed to kill the process of the active window (hung games, etc). It was originally made by jalf, but I added some extra features (with his permission, ofc).

The following description is mostly ripped from his post with some additions:

What it does:

Kills the current foreground process (The active window (or fullscreen app), just as if you’d terminated it from the Task Manager.

Makes windowed games (or applications) cover the whole screen, so you can play it virtually in fullscreen and alt+tab without the game minimizing and hanging (and possibly crashing), and IRC on your secondary monitor.

It is also multi-monitor friendly, so your game will fill the monitor that has the greatest area of intersection with the game’s window.

Note: Ack. I noticed there actually was a program that provided this behaviour, however it seems to require configuration for each game/application you use it with, which makes it not as quick and easy to use, IMHO anyway. Oh, well.

What it doesn’t do:

Destroy processes winthin your Windows folder, or a few other select processes (Trillian, Mirc, Visual Studio, 3d Studio Max, Photoshop and one or two others). Didn’t want people to accidentally close programs they actually wanted to run.

How to use:

• Download the app here.
• Run it (you’ll get an indication it is running when the ogre icon is added to the systray).
• Windows Key + Backspace to kill the active process.
• Windows Key + Escape to exit GameKiller.
• Windows Key + Page Down to remove borders and reposition/resize foreground window to fit monitor, i.e. makes your windowed game fullscreen.
• Windows Key + Page Up to revert foreground window to previous settings.

Workarounds:

Some games exclusively capture the keyboard or run under administrator privileges, so I added the following shortcuts to cater to these:

• Windows Key + Shift + Page Down works like Windows Key + Page Down, except it works on the window the mouse is currently above, instead of the active window.
• Windows Key + Shift + Page Up works like Windows Key + Page Up, except it works on the window the mouse is currently above, instead of the active window.
• Windows Key + Numpad Plus to elevate application with administrator privileges (required for broken games that have to run as administrator).

Compatibility:

So far, I’ve tested it successfully on:

• Team Fortress 2
• Battlefield 2 (requires use of shift’ed shortcuts and elevation if you run battlefield 2 with administrator privileges)
• Civilization 4: Beyond the Sword (requires elevation)

However, by using ShellExView, I was able to locate the culprit: unlocker (an application intended to identify which processes are keeping locks on files).

After uninstalling the application and restarting explorer, everything was fine and dandy.

I do a bit of googling to research it and it seems to be a fairly common problem for Surftown users. My guess is that their IIS is overloaded, but maybe their support crew will enlighten me.

Fallouty

With all the buzz surrounding the fallout 3 game prior to launch, a bunch of people on the bethesda forums started posting fanmusic. I got caught up in it and started experimenting with organic sound, texture and tribal sounds. This is the result of that session. I did not post it.

Cinematic

Researching the Korg KARMA for purchase, I got swept up in some of the demonstration tracks. Especially one that was called cinematic or something (I can’t find a link to it anymore). This was my attempt at replicating the sound and motif (with the KARMA, which I, of course, ended up purchasing).

Future Scream

This must be one of the earliest tracks I did with the KARMA, which was pretty cool, but I could never figure out how to progress from it.

Echo

Me playing with echo filters and C64 SID-ish sound. I believe I used KARMA MW software to generate some of the arps, which is pretty cool, but poor integration with Cubase made me not really use it much.

Familiar Drum’n'Bass

Some tangerine dreams-like drum’n'bass music which I’ve probably subconsciously ripped off from someone.

Funky Drum’n'Bass with Electric Piano

I heard a dnb track from a friend called DemonSeed, which was pretty awesome (sorry, still no links, I’ll have to get better at researching, keeping stuff). This track was inspired by it and probably copies a bit of the arrangement, but with a slightly sinister streak.

Tingaling

Again, a not very original drum’n'bass composition filled with endless clichés. But listening to it again makes me think I definitely need to use more compressed drums in my songs.

Lush & Logon

Ambient snippets which I hope to someday use in a game in some form or other.

Green

This is a track I wrote for a group I used to be in called 7am, which never really took off. Project start-up misfires seem to be the bane of my life, but then I partake in so many.

Approach

I think this track got 3rd or 2nd place in some lan party competition or other.

Final Cut

This track got 1st place at some lan party competition.

Construction Zone

I made this song for a Portal map that whoiscraig was making. I tried to replicate some of the sound of Portal’s music, but mostly got carried away. He didn’t use it as far as I know, even though he said it sounded great.

I had problems with uploading pictures, getting the dreaded “Is its parent directory writable by the server?” error. This turned out to be a directory which did not have proper permissions; /srv/www, in fact.

So, I ran

sudo chown o+rx /srv/www

Which allowed me to upload the images, but now another problem. The images were not showing up. This turned out to be due to a missing symlink in /usr/share/wordpress to /srv/www/wp-uploads/blog.metapathy.com.

Right! So, I added the symlink by running

ln -s /srv/www/wp-uploads/blog.metapathy.com /usr/share/wordpress/wp-uploads

and that would be that, you’d think. But, no, they still weren’t working! Now apache was getting mad and forbidding me to access it, so I started looking at what I had done with my apache configuration (I had been following this guide). Turns out I needed to use Options +FollowSymLinks rather than merely Options FollowSymLinks.

And now it finally seems to be working properly.

Here’s what it’s supposed to look like: