NeuroEvolving Robotic Operatives

[grassh0pper]

Training can take a lot of time, especially during higher-order tasks; whereas visual examination of the units' behaviour is important in general, it is not necessary to do this every second. After having set up the course taking a glimpse every once in a while is sufficient most of the times, thus we can relieve the CPU from the visualisation load, shifting focus to the training itself.

This is why I'd like to suggest to introduce two new graphic modes along with a speed slider (assuming, none of them exists currently):

* Speed Slider: The speed slider sets the speed of time passing in the simulation. If set to "1" time passes as usual, if set to "20", simulation time passes 20 times faster than usual (if this is possible; probably depending on the CPU-performance)

* grahic mode "tactical view": graphical minimal output at low resolution: [green dot = own unit; red dot = enemy; blue line = wall/obstacle; dark grey line = pathay; light grey light = fire-trajectory]. This low visualisation modus allows to speed up the training while still allowing the CPU to catch up with visualisation as well as allowing the player/training to get an idea of his units' performance.

* grahic mode "off": no graphical visualisation of the units' movement. Possibly, only statistics are shown in this mode: fitness graph; brain complexity, etc. This modus would allow the training to be performed fastest at the fastest possible speed.

Interactions: If speed is set to <=1, the user can flip between all visualisation modi at free will. From speed >1 on, it depends on the simulation performance what graphic modes can still be displayed. With raising speed set by the user, from a certain avg. CPU-load on, the program automatically switches into the tactical mode, then into the off-mode.

Being new around here (but not entirely new to neuronal nets), I am not sure if this exists already or not.
Maybe I've just not yet found the right button to click?

[Martijn]

don't make the training to fast, or everybody can just make a good team in no time.

as Mr. Miyagi would say: good training need patience

[grassh0pper]

It shouldn't be the time one spends in front of the computer watching how they get a tiny little bit better which qualifies a good trainer, but the quality of the training scenarios and tasks he sets up, I think. So there's nothing wrong with a fast-forward button every once in a while.

Until now, though, I have to admit that there seems to be plenty of room for me to improve.
Still haven't quite figured out how to have the units avoid fire.

[Cryptical]

Well, soemthing that we should consider in the engine is support for multible CPU and GPU, to increase the performance and more complex levels like one with a building to navigate throu and something like a balcony to shoot up too or down from.
A real 3D enviroment, this would make it posible perhaps to train Brains that might or moight not work in the real world.

[grassh0pper]

I agree with you, that more complexity in creating different scenarios will attract more people to the game. Nevertheless, before we only begin to think about this first we need to solve the speed-issue which we have already on the low complexity level nowadays.


Asking ourselves where does fun come from in Nero, we can identify several levels:

(i) - pure selfsufficient gaming through setting up scenarios and training
(ii) - general competition and particular competition-events
(iii) - social exchange via forum about a hobby (nero)

Although in (i) the fun comes mostly through seeing your bots advancing, it is not the absolute amount of time invested, but it is the teaching effect you see and get as a reward for playing. This is why I would like to repeat my suggestion from this post: Introduce a speed-slider and increase the max-speed the game runs, so that we can enjoy the game more!

Keep in mind that the basis for enjoying (ii) and (iii) also strongly relies on a fascinating game as stated in (i). So there is no fun for (ii) and (iii) if there is no fun found in (i). (exaggerating a bit).


One tool to increase the max-speed of the simulation could be found in supporting the usage of multiple CPUs and GPU, indeed. But before approaching this area, personally, I'd rather introduce the speed-sider and different graphic-modi.