NeuroDriver
by sem23
NeuroDriver
Tags for NeuroDriver
Description
In NeuroDriver you are managing a team of drivers instead of driving yourself. While the game ships with some pre-trained drivers, you are also able to train your own elite drivers with a machine learning technique running in train mode. Nevertheless, you should be able to beat the game without training, which does not mean that improving drivers has no advantage.
Furthermore you can create your own drivers in options menu. The size of brain depends on its architecture: multi-layer perceptrons can have more neurons and hidden layers than the long-short term memory (LSTM), because LSTM has - due to its architecture - more hidden neurons and connections, which impacts on computation speed. While applying new brain values will reset your current drivers, you can change genetics parameters all the time without auto-reset.
The currently used machine learning technique is part of the neuro-evolutionary algorithm theory, which combines neural networks with evolutionary/genetic algorithms. Feel free to ask me or have a look at wikipedia, it's worth reading!
How to Play
# win money by driving arcade races, tournaments or special events
# you can change driving style which influences speed and fuel consumption
# you can improve your drivers by training (although the supplied ones are quite well)
# get all cars & upgrades
# achieve best rank in all races
# create new tracks with track editor
Developer Updates
New Version out now! All tracks updated plus an additional training track!
Changelog 1.4.0:
- replaced all tracks by tracks derived from original DTM series
- added derived version of famous GranTourismo™ ComplexString
- minor changes related to more realistic car speeds (visualization)
- fuel is now at 180 (still using 1 unit per second at “normal” drive style)
Comments
sgschult
Jun. 16, 2015
Very frustrating when you have perfect racers in training, but then they all crash into each other and can't complete a lap in race mode.
Oh yeah! I know what you mean... I think it's because they never learn to "overtake". I tried to implement a scenario for learning exactly this, but my first try was not successful, but next tests are running to see how it is possible to train this. But as theory says: everything you want to learn - you are able to, but it depends on the environment ;-) My first try: I started a single "hunted" car, with enabled collision and vision, slower than others and in some distance in front of them, then I started all the others and hoped that they would learn to overtake the "hunted" car... LOL... They didn't, they learned evevrything from oscillating from left to rigtht all over the track to driving minimum distance to the barrier, but did not learn to do something what we would call an overtake maneuver :-D
TheDiscoverer
Jun. 10, 2015
nice game idea,racing + idle never before so much tried...the game needs a short intro and not a 6 paged-tutorial,besides that good game;)
This encourages me in the idea to split up the game in two parts: the "nerdy" experimenting part of training, and the gaming part which could be something like Gran Tourismo's Bspec-Mode, where you manage your racing team. Then the user could decide what he wants to do, just playing an idle-racing / management game, or having a look into neuroscience. Of course you will be able to do both.
Jondean
Jun. 11, 2015
tried building a track, but their needs to be more track parts to complete it, unless you can rotate the track parts.
Sorry for the confuse... the editor is not explained yet. You can rotate the tiles by "right mouse button click" and the plus and minus are for resizing. I will add an in-editor tutorial for making this more clear. Thanks for playing and testing the editor in this state! (I know that it is really hard to build a track at moment!)
TheDiscoverer
Jun. 10, 2015
When i upgrade something of the car a description would be nice,not only pictures of anything we dont know
Thanks for input! This will be fixed next update!
Skull_Krusher
Jun. 15, 2015
What does power distribution do?
It is the ratio between acceleration and maximum speed. In current version (1.1.0) the value starts at 0.6 (60%) and its max value is 0.8 (80%). When tuning option comes back in game (planned in next version 1.2.0), you will be able to change it as you want, of course inside its limits.