Page 4 of 4 FirstFirst ... 234
Results 61 to 66 of 66
  1. #61
    Member BluePrintRandom's Avatar
    Join Date
    Jul 2008
    Location
    NoCal Usa
    Posts
    18,423
    bge could be used in conjunction with bpy or logic node addon or hive to write nodal logic using genetic algorithms or reinforcement learning using examples.

    pybullet is nice, we could use a featherstone solver /bt_multibody
    Tristan will likely add bindings if we ask after he finishes up what he is working on.

    training data =many diff style working enemy ai w/ logic nodes
    etc.

    just because other programs exist does not mean bge could not do it*


    so write ai. test ai.

    wrtite new ai,test new ai etc.
    Last edited by BluePrintRandom; 14-Dec-17 at 16:10.
    Break it and remake it - Wrectified
    If you cut off a head, the hydra grows back two.
    "headless upbge"



  2. #62
    Originally Posted by BluePrintRandom View Post
    bge could be used in conjunction with bpy or logic node addon or hive to write nodal logic using genetic algorithms or reinforcement learning using examples.

    pybullet is nice, we could use a featherstone solver /bt_multibody
    Tristan will likely add bindings if we ask after he finishes up what he is working on.

    training data =many diff style working enemy ai w/ logic nodes
    etc.

    just because other programs exist does not mean bge could not do it*


    so write ai. test ai.

    wrtite new ai,test new ai etc.
    For game AI, why not. I doubt that it makes sense for cutting edge research.



  3. #63
    Member BluePrintRandom's Avatar
    Join Date
    Jul 2008
    Location
    NoCal Usa
    Posts
    18,423
    with realtime ladar, the realworld can become a sim.

    once simulation matches reality you have something suitible for research.

    remember my walking ragdoll rig?

    with a btMultiBody running that rig w/ neural trained logic or even py you could do amazing animations that were procedural and lifelike. like crowd simulations etc.
    Break it and remake it - Wrectified
    If you cut off a head, the hydra grows back two.
    "headless upbge"



  4. #64
    Originally Posted by BluePrintRandom View Post
    with realtime ladar, the realworld can become a sim.

    once simulation matches reality you have something suitible for research.

    remember my walking ragdoll rig?

    with a btMultiBody running that rig w/ neural trained logic or even py you could do amazing animations that were procedural and lifelike. like crowd simulations etc.
    For neural trained logic/lifelike animations, you don't need Blender or any kind of visualization. You can work directly with the animations and train a neural network on that as has already been done, but not with reinforcement learning. They used supervised learning, because it is a lot easier and less computational expensive, but still not trivial.



  5. #65
    i believe this is whats needed.
    This from theguardian.com the article is by colin stuart.
    The future of computing power – from DNA hard drives to quantum chips

    Tom Ran, of the Weizmann Institute of Science in Israel, works with computers made out of strands of DNA.
    "Working this way, we can get three trillion computers, working in parallel, in a space the size of a water droplet," he says. The 0s and 1s of conventional computers are replaced with the four DNA bases: A, C, G and T. Operations can be translated into strands of DNA using these bases, and the way the DNA strands interact with each other produces new strands which can be decoded as output values.



  6. #66
    Originally Posted by Lostscience View Post
    i believe this is whats needed.
    This from theguardian.com the article is by colin stuart.
    There are many techniques which are years or decades away and lots of them that are never going to be feasible. If the goal is to tackle a problem now, all of them are irrelevant. Before we see this kind of new hardware technologies, it is more likely that we are seeing significant improvements on the software side.
    It is not unusual in machine learning that the first goal is to solve a problem with an exorbitant amount of computational power, just to get the understanding whether it can be tackled with the existing tools. And if that is possible, it is being improved, e.g. by using new techniques which seems suitable for the given problem, until it becomes way more feasible and can actually be used for practical purposes.



Page 4 of 4 FirstFirst ... 234

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •