I (think I) can understand how the inputs may be structured for a neural network playing a game like chess or go where the next action of the player depends on inputs of a fixed size. However there are simpler games where the state visible to the player is of a variable size (i.e. in the case of uno, a simple card game, the player may have any number of cards in their hand, and the next move may depend on the cards that have already been played of which there is no set amount).
Are these types of games incompatible with a neural network? If not, what might the input layer look like? And if so, what are the alternatives?
You can implement everything you want with Neural Networks/Expert systems/AI, but what you have to consider if this is the better tool to do the job.
When you work with this kind of approach the problem in most of the cases is how to encode the input in a proper way to archive the best result when you evaluate your AI.
Yes, you can implement chess and uno with AI, but you have to think first on how you can code your inputs and on how you think when you play the game.
Doing this with AI it's much more difficult than use a traditional approach.
If you google implement Chess AI you will find a few implementations, some using electromagnetic field intensity, other one using functions to rank moves.
The question is what you want to archive, a game developed with an AI or to code a good chess game?.
Do you want to use the tool because it's cool or because you think that it can fit on the problem context
Related
I'm new to this whole neural network stuff, and for my first project I'd like to detect a car moving trough the frame. So far I found the cocodataset and YOLO object detection.
What I'm think I'm after is Object/Instance Segmentation, so I can define a region of Interest inside my view and then check for overlap with the Instance segment. As soon as a certain area is reached I can use that information as a trigger.
I guess as soon as I have the classified instance outline the geometrical stuff shouldn't be the issue, but for now I don't know if this is possible.
Has this been done, is there some example code (preferably in some "lazy" language as python, js). Is this currently possible in realtime at all? Do I have any chance to run this on a modern smartphone?
thank you so much for all the fast responses I truly appreciate it. I have another question I've been wondering about, so I am working mobile game and I want to keep the size as small as possible. So I wanted to know does using Unity particle effects, impact performance and game size? I was wondering, if i wanted to make a visual effect of breaking apart a GameObject when destroyed, should I use unity particle system to do so? or should I make an animation of that game object breaking and use that instead?
to summeize should I be making animation for things such as rain effects and confetti and fire effect instead of making them with the particle system? is there any drawback to using one over the other?
As TEEBQNE said it is your choice between animations and the particle system depending on what you need. If your team is small or you are a single developer it is best to opt for the particle system for the flexibility of changes and quick execution.
Optimising the particle system is not hard, just make sure you don't overdo the numbers and avoid some common mistakes. There is a list of the best practices for optimization on mobile games found here
You can also find some pretty good looking particles premade in the unity asset store, which are already optimized well.
I have been building a game for VR using Unity3d. It has only low poly models and the file size is less then 40 mb still the game lags when played on mobile.. Please suggest how to improve the performance..
Thank you in advance..
In order to improve performance in VR for mobile you have to optimize everything as best as you can, you should keep some of these variables in mind:
Graphics Side
- Number of polygons in the scene
- How many source of lighting do you have
Programming Side
- How much work is taking your code, is doing it efficiently?
The programming part can include problems within the physics system, also some logic problems that can probably decrease the overall performance because of higher computation.
My advice is to learn about the Profiler that unity offers, actually you can observe how much work is taking your code and where exactly it is your bottle-neck. This video also can be useful.
Of course a solution could be implement your code following design standards, like design patterns and software architecture (depending on the size of the project).
I hope it can be useful for you!
What I found from developing and launching a vr game is some of the issues below
Number of polygons is usually your first to check even though your models are low poly. For example, I looked at Synty models in the unity store and some of them were over 1k for a bag and 7k for a character model. This seriously reduce the amount of objects you can if you want to target a max of 50000 per eyes.
With some models, you can use blender and the decimate tool to reduce the polygon count pretty easily. From there I would use LOD's to reduce their count further based on distance.
Use occlusion culling (pro version only)
Set your camera distance to maybe a 100 instead of the default
Use mobile shaders and careful using some of the standard shaders as they are expensive. Also transparent shaders will becomes expensive cause overdraw.
Batch your textures and make them static if possible
Don't use dynamic shadows on lighting but instead bake your lights
Try to avoid using physics as this becomes expensive and instead raycast to trigger events or shooting weapons.
Run profiler often and check for any bottlenecks (pro version only)
Reduce the count of Particles effects and their values
Character bones can also cause issue so remove as many as possible
There is also your code to look at as mentioned by Manujamming
Set quality setting to low in the inspector to gain best performance.
Could you provide a screenshot of your game scene?
I hope this makes sense.
Best of luck!
I have been digging around trying to find a way to show a game board of sorts.
It is basically a square board with a round hole in the middle, I am able to render the scnBox and the scnTube, but I would like the area where the scnTube sits in the box for the box to be transparent and see through the game board, but can't seem to find anywhere that has an example. Any help would be much appreciated. I am hoping that I am just missing something very simple, but this is my first time using scene kit.
Thank you.
Before Unreal Engine 4, (UDK and prior) Epic's modelling space was subtraction - a filled block was your game world and its extents. From inside this block you took chunks out to create space for players to run around in, and shoot each other. All's fair in love and war.
I'm telling you this because it's a good example of how contrived 3D modelling is compared to real world scenarios, and should (hopefully) put you at sufficient unease to digest what follows.
This approach of carving out of a finite block is still in Unreal Engine 4 and popular with older users, but it now defaults to an open, infinite world into which things are added. Most new users gravitate towards building into an infinite space of nothingness rather than carving space out of a solid, finite block.
Everything about 3D modelling is virtual, and virtually impossible to relate to the real world. Instead of thinking in terms of how things could be done if objects were real and literal, you need to think in terms of the limitations (and there are many) of geometry definitions as used in most 3D modelling and game engines.
The programming equivalent of this mental gymnastics is going from the concept of classes and objects to their realities within languages and frameworks. On the one hand the ideas and their ideals are wonderful, and on the other the realities are a bleak reminder that programming languages haven't really progressed very far, at all.
3D modelling is exactly like this. It's not much further along than it was decades ago, and is still using archaic ways to solve many of its original problems.
Cutting a nice, clean, efficient round hole in a cube is one such original problem.
A very simple shape is being intersected and cut by a shape with the potential for infinite complexity. What should happen? Should the simple become complex or the complex become simple? How to make the most graceful transition between the two?
That's the problem you're facing: a cube is a simple geometric shape, easily defined by minimal line segments. A cylinder introduces infinite possibilities for line segmentation around its circumference.
So somewhere along the lines of development, the architects of 3D modelling had to come up with a way to make these contrasting line complexities play well together for lightweight presentation on limited hardware. Their solution, in most cases, is a hybrid and a disaster of user operability, but masterful in its geometric efficiency: Polygon modelling, UVW unwrapping and subdivision!
All of which means that if you want to achieve this in the best way possible, with today's tools, for the purposes of Scene Kit, I suggest polygon modelling this board in Maya for 4 reasons.
It's got a 30 day free trial.
It works on a Mac
It's polygon modelling tools are second only to 3ds Max
It's easier to learn (for a complete newcomer) than MODO, and miles easier to learn than Blender.
MODO is interesting if you're already skilled in Polygon modelling, but it's so utterly discombobulating if you don't have that prior experience that I'd recommend using just about anything else first. Except Blender. Blender is free, but don't be tempted. It will cost you more in learning time than buying a copy of every other professional 3D app.
In MODO's favour, and the reason I mention it, it does export nicely for Scene Kit. I know that for a fact, but am not yet sure how well Maya exports for Scene Kit.
Which is the next problem you're going to come up against. All COLLADA files are not born equal.
New Maya does have Unity and Unreal export presets, so I presume it's possible to calibrate its COLLADA exports to match the demands of Scene Kit perfectly, just haven't yet needed to do it. This will (very likely) involve trial and error to get the settings right. It would be nice if Apple would tell us exactly how to configure export from all major 3D apps for Scene Kit, but instead they're giving us the half baked Model I/O, so we can double the effort of importing artwork.
All context aside (which has largely been to demonstrate that 3D is no simpler nor more refined than using an IDE and frameworks like Xcode and Cocoa), here's the meat and potatoes:
A video on one aspect of what's best to make holes, and starts out as you are, with a cube and a cylinder:
https://www.youtube.com/watch?v=zaEv5rio8bk
But it does presume a certain amount of Maya familiarity, some of which you can gain from this rather slow and ponderous examination of two other ways to make holes in cubes:
https://www.youtube.com/watch?v=lvMfoH5Ikrc
Yes, if you're counting, that's 3 ways to make holes. Actually four, because the first video starts with the boolean operation you might have been expecting to be how this could/should be done. In some parallel future we'll have well working boolean geometry operations. We're not there, yet.
Hopefully that same parallel future will offer us a programming language, frameworks and terminology that's not confusing and maintains metaphors long enough to make teaching easy and usage elegant and simple.
I dont know about that long answer but this can be achieved with Boolean Subtraction. You create a cube and a cylinder. You subtract the cylinder from the cube. In 3ds Max this is under compound objects-modifiers-boolean subtraction. I guess Maya has a similair function somwhere in the menus.
I'm working on a 2D game (kind of like a top down space shooter) for the iPhone using an engine very similar to cocos2d (not exactly though) on OpenGL ES. I'm trying to figure out how I'm going to do collision detection.
All the ships for my game are images, and the game will load the image as a texture onto the screen. I've got very very simple detection going already that basically just takes the rectangles of the images and checks to see if those collide and can do that just fine.
But, of course the ship isn't perfectly taking up the entire rectangle so there is whitespace in there. So my question is how am I supposed to account for that whitespace? Do I have to have the matrices of the ships stored? Or is there another way? I've also heard of possibly using the Chipmunk physics engine for collision detection? How would that work?
(1) regarding Chipmunk, the short answer is yes you should immediately download chipmunk, donate something to the bloke, and start learning about it.
Working with that for a day or so will basically answer all the questions you have. If you want to work with physics games you're going to need to get in to it.
(2) you ask about using an approximation ("just" a rectangle) instead of something more accurately shaped like your spaceships. In fact, you'll be perhaps amazed to learn, that is precisely how it is usually done in all your famous big-name games you've played since we were all kids! Indeed sometimes you might use little more than A DOT (!) to detect collisions.
What you'd probably do in production is try a more complicated model, and play with it for a few hours and see, is it actually any better to play with than your simple dot or rectangle model.
If you do want to make a more complicated model -- just make one! Build it up from three or four rectangles using your current system. Try them "all against each other", and have "one big one to check first" to see if it is even anywhere near each other (sort of a simple spatial hashing).
You will find that when you do it with Chipmunk, which as you now know you have to immediately begin after reading this message, you just build it up the same tedious way. It's not a magic bullet. But if you were going to use a "more complicated model" yes it is better to go with something standard, chipmunk, to do the work in - it will get done quicker and better. There is heaps to learn and you should hop to it!
(3) Unity is not just for 3D Finally if you want to do it the smart-ass grown up way, you'd have to use Unity3D which will let you access the very metal, the Nvidia physics on the chipset. Note that unity works perfectly for 2D games also - you just click one button in unity to use a 2D projection (many brand-name ifone 2D games are done exactly like that).
If you use that approach, you can (if you want) have "absolutely exact" physics, with every nook and cranny of your model modelled.
What is the downside to doing this? Ah hah ... well the thing is, you need superb actual 3D models of all the stuff in your game! (Like you see them building in the "how we made the movie" special features that come with your favourite Pixar blu-ray.) To do that you need things like autodesk, maya and the like. you would quite likely buy some models ready-made from a digital prop shop (no need to build "a chair" as it has been done 1000 times already and you can buy one for ten dollars).
(Unity3D is completely free to use for a few months while you see if it can make you money.)
Incidentally on the Chipmunk front --- you can just use Corona which is ridiculously easy to use and has chipmunk-like physics completely built in with zero effort on your part! You could have the whole game done in less time than it took to write this email. You could be selling your game already and thinking up the next one. Or, you could use "Cocos" which indeed has a chipmunk-like physics library built-in .. personally (just me) I do not like and won't touch cocos - but of course many games use it.
(It seems pointless, to me, using cocos which is a "for idiots" product, when you can just go ahead and use Corona, which is a "for idiots" product but stupendously easier to use, 1000x more solid, and probably literally 10x faster to finish your product and start making money.)
Noel Summary:
So in some sense using Unity3D (and hence, the actual nvidia physics on your computer's chips) is the ultimate solution if you want detailed nook-and-cranny collisions. Going down one step, Chipmunk is exactly, precisely what you should be using on the ifone/ipad for 2D physics -- it is precisely what is used in all the famous games we know so well. You have a bit of learning to do so hop to it - it's superfun. Finally go right ahead and just make your current model more complicated if you wish - roll your own by adding more rectangles!
And the fourth point is, be sure to remember that in games, astonishingly, you can often get away with remarkably simple physics (often SIMPLER!! than one rectangle - just a damn point - ie, simply measuring the distance between centers!) Fifthly after going to all the effort of testing more detailed physics, you would play test one against each other, and find out what is the simplest physics you can get away with.