I am trying to build a snake game using artificial neural network. I want to know what are the possible inputs. I am planning to input the direction of the apple relative to snake head? But what will be this value? Like it the apple is 90 degree right of snake? can i give it value 0.25 and when the apple is 180 relative to snake head, can i give input as 0.5 and values in between based on this?
I am a beginner only want to use simple backpropagation algorithm not the q-learning algorithm.
Is it necessary to give distance to apple as an input?
I think you should consider this for training:
distance and angle from head to apple
what is on the left from his head
what is on the front of his head
what is on the right from his head
Also you can take this as reference- Youtube
Related
I am making a game in Unity that involves creatures whose animations are determined by physics. For example, to move a limb of a creature, I can apply forces to the rigidbody it's associated with. I know how to apply forces programmatically through a script to create movements, but I'd like to create more complex and organic movements and thought that I might be able to use a neural network to do this.
I'd like each of the creatures to have a distinct way of moving in the world. I'd like to first puppeteer the creatures manually using my hand (with a Leap Motion controller), and have a neural network generate new movements based on the training I did with my hand.
More concretely, my manual puppeteering setup will apply forces to the rigidbodies of the creature as I move my hand. So if I lift my finger up, the system would apply a series of upward forces to the limb that is mapped to my finger. As I am puppeteering the creature, the NN receives Vector3 forces for each of the rigidbodies. In a way this is the same task as generating a new text based on a corpus of texts, but in this case my input is forces rather than strings.
Based on that training set, is it possible for the NN to generate movements for the characters (forces to be applied to the limbs) to mimic the movements I did with my hand?
I don't have that much experience with neural networks, but am eager to learn, specifically for this project. It would be great to know about similar projects that were done in Unity, or relevant libraries I could use that would simplify the implementation. Also, please let me know if there is anything I can clarify!
Not really an answer but would not fit for comments
I'm not sure the strategy you want to apply to train your model is the right one.
I would go for reinforcement learning methods (you can check this question for more infos about it) using, for example, the distance traveled by the center of mass of the creature on the x-axis as a fitness. If this leads to weird behaviours (like this well known robot) you could, for example, think of strategies like penalizing your individuals given the distance traveled on y and z axis (still by the CoM) to try having guys that keep there CoM on the same plane.
Without knowing exactly what you want to achieve this is hard to give you more advices. Although, if you are not looking only for neural network based techniques, there is this really great paper you might want to have a look at (here is the video of their results).
I'm trying to understand how belt pulley from Mathworks work. I have this simple sample:
In this case. I'm giving the velocity of a. And there is when appear my first question:
why the magnitude of velocity of b, is the same of a?
According to the theory of belt drive, the ratio between both velocities is this:
Va/Vb = 1 - (Ta - Tb)/EA
But the "Belt Pulley" description of Mathworks (http://es.mathworks.com/help/physmod/sdl/ref/beltpulley.html#zmw57dd0e1842), said that both will have the same magnitude but different direction.
Anyway, I could assume that they are not consider the elastic of the band, but I don't have any clue of how they get the angular velocity of the pulley to get then the relative velocity. At some point they need to know the elastic of the material for get the relative movement between the belt and the pulley.
Can anyone help me?
I think the answer to the first part of your question lies in the block assumptions:
Assumptions and Limitations
The model does not account for compliance along the length of the belt.
Both belt ends maintain adequate tension throughout the simulation.
To get the angular velocity of the pulley, use an Ideal Rotational Motion Sensor block. Connect the R port to the rotational free end of the pulley and the C port to a a Mechanical Rotational Reference block.
The motion between the belt and pulley is described by those equations (from the documentation page you mentioned in the question, where all the parameters are detailed):
I'm trying to use GA to train an ANN whose job is to move a bar vertically so that it makes a ball bounce without hitting the wall behind the bar, in other words, a single bar pong.
I'm going to ask it directly because i think to know what the problem is.
The game window is 200x200 pixels, so i created 40000 input neurons.
The obvious doubt is: can GA handle chromosomes of 40000(input)*10(hidden)*2 elements(genes)?
Since i think the answer is no(i implemented this solution and doesn't seem to work), the solution seems simple, i feed the NN with only 4 parameters which are the coordinates x,y of bar and ball, nailed it.
Nice solution, but the problem is: how can i apply such a solution in a game like supermario where the number of enemies in the screen is not fixed? Surely i cannot create a NN with dynamic numbers of inputs.
I hope you can help me.
You have to use features to represent your state. For example, you can divide the screen in tiles and assign a value according to a function that takes into account the enemy (e.g., a boolean if the enemy is in the tile or the distance to the closest enemy).
You can still use pixels but you might need to preprocess them in order to reduce their size (e.g., use a recurrent NN).
Btw, a NN might not be able to handle 200x200 pixels, but it was able to learn to play Atari games using a representation of the state by preprocessed pixels of size 84x84x4 (see this paper).
I have a small remote controlled car going on the room floor. For simplicity let us assume it is moving along say x-axis. Now, the floor seems flat but there are very minute uneven bumps in every surface. So whenever the car is not exactly flat (as it was at starting position) or in other words whenever the car has even slightest of tilt then,
Total Acceleration obtained from accelerometer = Linear Acceleration + Acceleration due to tilt
My question is how to remove the acceleration due to tilt so that I get only linear acceleration? Can I somehow use gyroscope to do that?
I have implemented sensor fusion for the Shimmer platform based on this manuscript, it's basically a tutorial:
Direction Cosine Matrix IMU: Theory
This manuscript pretty much answers your question.
These have also been a big help:
An introduction to inertial navigation
An Introduction to the Kalman Filter
Pedestrian Localisation for Indoor Environments
Combine Gyroscope and Accelerometer Data
Just promise me you won't try double integrating the linear acceleration because it won't work and I suspect that it is what you are trying to do.
I think it would be fun to model a top view of a train following a track, traversing switches and so on, using a physics library like Box2D. What joints and motors would I need to make this work?
I'm curious about how to implement the forces needed to make the car follow a spline track so it can bump into other train cars, pedestrians, DeLoreans etc. Just saying "the car is now at spline(t)" for each time step would create excessive forces in the physics engine. If I understand correctly, you have to stick the car onto the track with one force, constrain its angle to tend towards parallel with the track with another (or stick the front and back of the car to the track with two forces), and create another force to propel the train forward. I'm looking for some details on how to accomplish these things.
I believe it would be easier without "real" physics, like the ball movement of games such as Luxor or Tumble Bugs. Meaning: let the train follow a spline which is defined by the tracks.
Using phyiscs is probably overkill to make a train follow a track and could lead to all kinds of undesired side-effects, including jerky motion, train derailing, train getting stuck on junctions, etc.
You could still join the individual wagons together using physic joints, however. Just make sure that only the locomotive gets acceleration forces, the rest of the train just follows or is pushed but stays on the spline.
Why are you worried about keeping it "on the tracks"? Where is it going to go? Gravity should keep it down, object intersection should keep it up, and so the only directions you need to worry about are forward and backwards. That's where a motor comes in, and you're done. The rest is decorations.
In response to edit of problem:
Siderails. And have the train long enough / rigid enough compared to its width that you can navigate crossings (make them closer to right angles to minimize the crossing problems.
A top-down view (i.e. seeing the train from the sky) doesn't really require a 2d physics engine - if I understand you correctly. In fact, it seems like it wouldn't really help with the problem (if you want a train simulation), but then maybe you just wanna try it out for fun. :)
However, what about putting something like a slider joint on the train and the cars, and a motor on the locomotive. The slider joint might need some special implementation; you probably want to run the train along a spline and not a segment of straight lines, right?
Some sort of ball joint would connect the cars together.
The implementation is not so toughand I was able to prototype something in a few hours that does the basic job. It will require a lot of work to make it run smoothly, but it's essentially just "siderails."
Being top-down you obviously first must turn off gravity in Box2D. Second, build a train. Treat train wheels like car wheels and it'll suddenly get a lot more simple. For tracks you have a few choices:
Create your own game object (not in the box2D world) that is a simple line the train will then "follow" (you can use motors on train wheels to "steer" towards the line). Then just overlay the line with some nice wide "rail" graphics and you have a nicely faked system. Tell the wheels to turn off if it strays too far from the line and presto, you have a derailment.
Create actual physical rails - outside rails (like siderails) that the trains "wheels" will bump into. They will have to have gentle curves in this instance, which could be very difficult given the limited resources you have (simulating a nice slow curve out of boxes in Box2D is rough on the processor)
Then just let your train go!