The developer has not uploaded a game yet...

Update: This game is now available on Steam Early Access: https://store.steampowered.com/app/1102370/Dr_Derks_Mutant_Battlegrounds/

Dr. Derk’s Mutant Battlegrounds is an experiment in using deep neural networks in a game. As a player you train agents, “Derklings”, in a MOBA-style setting, and battle other players teams. The Derklings can be equipped with a number of upgrades, and the challenge is to figure out the best way to train and equip them to beat other teams.

Some things to look forward to:

  • A dead simple way to get started training your own agents with deep neural networks. Just load up the agents and hit train! 
  • Some technical details; the agents employ a RNN with over 3,000 parameters and 32 memory slots, and are trained with GA (Genetic Algorithm). Plenty enough for them to come up with all kinds of strange, funny and clever solutions!
  • A bunch of fun upgrades that all change what the Derklings can do and what they might learn.
  • Three different battlegrounds with global leaderboards; 1v1, 2v2 and 3v3.
  • Visualizations for those who wish to dig deeper into the brains of the Derklings and fully understand their behaviors.
  • A community of people fascinated by AI and neural networks.

Dr. Derk wishes you and your team the best of luck

Development log

Comments

Log in with itch.io to leave a comment.

I can't seem to find this information. Are all derklings created equal? There are visual differences, but are there differing traits such as HP, detection distance, turn radius, attack frequency, and such? 

I love the game! I've been interested in AI learning in games since Galapagos and Creatures. I even created a game which faked a neural network called Darwin's Curse.

Thank you for your effort!

Thank you! Yup they're all created equal, no stats differences, they just look a bit different to give them some personality.

cant enter

Hm ok, where do you get stuck? And which os/browser are you on?

nah its fine, i figured it out

I've played this for a few hours now. It's an interesting concept with a chance of providing a kind of gameplay I haven't seen before. Guiding an evolving population towards greatness by trying to cleverly select, breed and kill individuals is something I haven't seen before.

The game provides you with various tools to filter out and find individuals that have the properties you like. At first you just look around and select individuals that do stuff you prefer, but soon you realize how slow that procedure is and you quickly start to crave more powerful tools like "hmmm is there any way I can select all individuals that have at least 1 kill, but who are also not exhausted, and then breed all of them at once?". And the game provides those tools for you, so it's just about trying to make a clever guess about what would work, and trying that out.

The game is obviously not done yet, and it's a little confusing at times. For example, it took me a while before I understood that I could only select and breed BLUE pixlings, because I hadn't understood that the RED pixlings were from another team.

But it's still kinda captivating. I think feedback from beta-testers on a site such as this is exactly what it needs at this stage, to figure out how to refine it. I messaged the developer with a few suggestions and one of them was even implemented the same day!

I noticed it runs like 8x faster on Chrome than in Firefox (at least on Linux). I like how it easily deals with scale, having thousands of little critters at the same time, and with good framerate (in Chrome).

Thanks for a really good comment! And yes more feedback is exactly what I need :)