A battery of Gauss cannons can protect your planet pretty well and erase a big incoming threat in no time! #devlog #gamedev #indiedev

Man, I just found out that the SF2 lib I use is not really under a CC license. I found an open replacement. Good news is that my ambient music sounds better, but all other SFX is garbage now. At least I have not created all of them yet. #devlog #gamedev #indiedev

Would you still love #Frutiverse even if it gets Worms? #indiegames #gamedev #indiedev

I finally decoupled rendering from the physical simulation. Now the renderer uses linear interpolation between two simulated states. Now I can run physics at a lower FPS like 50, and still have a nice smooth game running at 165+. #devlog #gamedev #indiedev

Sometimes it is nice just to chill on the planet and watch the sky. #devlog #gamedev #indiedev

Planetary defense needs more hands on deck, so guess who started building new ships in #Frutiverse! #devlog #gamedev #indiedev

And I put together a small server for collecting error logs. Hopefully, it will not cause any problems while doing that. #devlog #gamedev #indiedev

Finally, my SteamPipe is set. It was surprisingly painful to do it correctly for a Java game. #devlog #gamedev #indiedev

Another small visual upgrade. Space stations now use my new rendering pipeline based on a JSON definition, and the docking ports are more subtle. #ScreenshotSaturday #gamedev #indiedev

One thing I really hate about recording my game is that, due to the nature of my graphics style, a lot of the atmosphere is killed by video compression and artifacts. #devlog #gamedev #indiedev

Improved visuals for the Gauss cannon. Synced the flash more closely with the projectile, added a shooting sound and energy drain effect from #Frutiverse. #devlog #gamedev #indiedev

And finally, I have a proper Maven repository on my server. No more annoying manual publishing of my libraries. #devlog #gamedev #indiedev

It is time to bring out the big guns. Behold the mighty Gauss cannon, the core of planetary defense, capable of tearing apart big asteroids. #devlog #gamedev #indiedev

I just configured old ThinkCenter M920q as my server. It runs builds and I also can run there long evolution runs unbothered. And #Frutiverse runs on it also surprisingly well. #devlog #gamedev #indiedev

Nobody will see that in real life, because the asteroids are spawned far from anybody's eyes in #Frutiverse, but this is what it looks like. I wanted more progressive growth instead of them just popping up out of nowhere. #devlog #gamedev #indiedev

I continued with experiments. I wanted to create a network that would primarily coast and use only short bursts of boost to correct direction, and in the worst case use normal thrust. This seemed too complex goal for SAN, especially designing a scoring function that would provide correct guidance for evolution. In the last attempt I let the ship use only boost. It is able to fly through, but it is far from optimal. For this kind of behavior I should stick to the flow field. #devlog #gamedev #neuralnetworks

It is evolving, just backwards. I am trying to let ship fly through debris field, but for some reason it just flies backwards. It does not make any sense, because flying backwards is slower, and the ship is rewarded for speed, and all other points are the same regardless of the ship's rotation. I guess for this kind of steering I should stick to flow filed. #devlog #gamedev #neuralnetworks

This is starting to look really nice.I just changed the success criteria for splitting asteroids. Goal is that all new fragments are less than 50% of the size of the original. Pilot is now encouraged to slice through them and even produce multiple smaller ones, rather than just perfectly splitting it into equal halves. #devlog #gamedev #indiedev

Pilot received a new order. He has to intercept the asteroid and split it into smaller ones to protect planet. Even though he is punished for bumping into it, that is still the best way for him to stay in position against fast rotating asteroids. #devlog #gamedev #indiedev

I finished some experiments. Using bias as one input really improved learning duration and made fire control more precise, which kinda makes sense because the rest of the network controls ship propulsion. Distance from the debris surface is not really required, but without it the approach is somewhat choppy. I was able to cut down hidden nodes, network is therefore 8-3-4. I am happy with that, so it is time to move on to another behavior. #devlog #gamedev #neuralnetworks