5th Year Anniversary Update

This month I realised Biplanes Revival turns 5 years old…

When I was a child, I used to spend dozens of hours in BlueTooth BiPlanes. Naturally, I became bored with the AI and had to come up with imaginary challenges like “beat every bot without getting hit” and “beat every bot with 100% accuracy”. As one can guess, I ended up having a dream of playing this game against a living being, not against this stupid AI! (Actually, I played one session over Bluetooth, but the netcode was terrible and unstable)

Many years later, I was in the middle of my 2nd year of university, without a slighest idea of who I want to be, making my first game on C++. Pleased with a working multiplayer Pong clone (black-and-white shapes, Borland Graphics!), I decided to create something more colourful and entertaining. Given the fact I had zero experience in game design, it struck my mind that fulfilling a childhood dream by simply recreating that ancient arcade will benefit everyone!

It took just a couple of weeks to finish a working prototype, and another month or two of polishing. This is how it looked roughly a month after its first release:

Several months later, an experienced friend of mine made the game compatible with Linux and implemented matchmaking, making it accessible to most players behind NAT (although, omitting many technical details about networking, there are lots of corner cases when matchmaking won’t work).

Roughly a year after, right in the middle of the COVID-19 pandemic, we were training a neural network on gameplay snapshots in an attempt to create a clever AI. Back then, my partner didn’t have sufficient experience in this field; besides, it was pretty hard to squeeze singleplayer logic into the game designed to be multiplayer-only, so the plan was scrapped.

Another 4-ish years passed, and we’re back in modern times! That is, April, 2024. Me and my friend are now much more experienced in all our fields of interests, meaning I’m as ignorant about neural networks innerworkings as XioN is competent in them. Since last summer we’ve been rewriting the codebase and experimenting with several neural networks… to no avail (again!). Unfortunately, these experiments took too much time and didn’t go well for the time being, so an exhausted me decided to implement a “traditional” AI using a mix of techniques like Utility System, Context Steering and Finite State Machine.

Despite its straightforward design and overall simplicity, the resulting bot took me by surprise when he presented quite a challenge even for my experienced hands. As shown in these YouTube videos, he completely obliterates my other friend, who’s been a primary playtester and a formidable opponent for me since the very beginning (he also reworked a great deal of original sprites, upscaled game logo and created two custom soundpacks).

It’s not that I can’t beat my own AI, but it’s still much harder in comparison with the original Java game. If you’ve managed to beat this AI on maximum difficulty, congratulations: you’re at least on a par with two best pilots in Biplanes Revival known to this day!