Simulation Based AI for Robotics - Why is this not Happenning?

  • 11 Replies
  • 3796 Views
*

Kevin Long

  • Roomba
  • *
  • 4
  • LeapSpecies.Space
    • LeapSpecies.Space
Here is an AI that is the antithesis of neural/deep/machine learning.  It is a simplistic brute force physical AI.  Specialized for robotics but will quickly surpass much of our AI for functionality in all branches of science as it produces it's own laboratory out of its operational duties. 

– – -the loop- – –
Simulation Based AI loops feedback between the real world operational area and the simulation of that environment.  Robotic sensors relay operational changes to update the terrain in the simulation copy  –  the AI directs the possible robotic actions within the simulation  – finds the best chain of events by standard iterative brute force – then actions are issued to the operational machine and teams – the outcome is measured against the simulation’s prediction of behaviors – discrepancies in prediction are explored by iterative brute force to refine the behaviors of the simulation – the physics/materials/chemical models are continually updated by the actions of a thousand eyes and machines. The birth of the Universe Machine. The Evolution Machine is the turning of the same iterative functions upon improving the system and the robotics.
– – -the loop- – –

Bold statements - if so, then why is this not already in place?

*

keghn

  • Trusty Member
  • *********
  • Terminator
  • *
  • 824
Re: Simulation Based AI for Robotics - Why is this not Happenning?
« Reply #1 on: August 30, 2017, 05:12:42 am »
Designing Robots of the Future with Hod Lipson. At about 10 minutes into the video:

https://www.youtube.com/watch?v=89kUdJtyXCw


 I think it is possible thou.

*

Kevin Long

  • Roomba
  • *
  • 4
  • LeapSpecies.Space
    • LeapSpecies.Space
Re: Simulation Based AI for Robotics - Why is this not Happenning?
« Reply #2 on: August 30, 2017, 06:14:52 am »
Outstanding!  Thank you kegn.  It is disturbing that he is pursuing self aware type physical self image experiments.  Instead of self image experiments the concept seems functional for operational use now?  Put some robots to transporting for instance but the AI coordinates all movements and improves the simulation to make the predictions match up with the actual outcomes.  The concept seems amazingly simple.  I think what is missing is a project for robots with a command and control AI (like a video game) and set it loose to update its simulations to the real world.  A universe machine is probably in place already?  Chinese?  Indian?  Too powerful to be overlooked for application.

*

infurl

  • Administrator
  • ***********
  • Eve
  • *
  • 1365
  • Humans will disappoint you.
    • Home Page
Re: Simulation Based AI for Robotics - Why is this not Happenning?
« Reply #3 on: August 30, 2017, 09:40:49 am »
Here is an AI that is the antithesis of neural/deep/machine learning.  It is a simplistic brute force physical AI...
Bold statements - if so, then why is this not already in place?

If you could calculate how many iterations would be required to get results and how long it would take, you would be able to answer your own question. Alternatively, try a simpler problem such as calculating how many combinations of chess moves there are.

*

ranch vermin

  • Not much time left.
  • Terminator
  • *********
  • 947
  • Its nearly time!
Re: Simulation Based AI for Robotics - Why is this not Happenning?
« Reply #4 on: August 30, 2017, 11:10:42 am »
Hahaha!! I think this is a case of too much pipe dreaming and not enough actual development.

We also understand the nature of alot of the atomic activity but you cant possibly simulate the universe unless ur tripping out in the computer either, because youve only only 5 gigahertz of instructions to get through even to make 1fps....   and theres a much a larger amount than that there.
And a fullish physical model probably would take a good 100 kilaherts easily,  so youd only get 50,000 atoms,  not enough for a 1/10000000000000 of a dna string.   so its a waste of time.
Besides, even if you were stupid enough to wait long enough for it to run one frame,   are u sure that evolution theory is even correct or anything interesting happens at all?
Theres things that are too big...       that guy thats writing that interesting text knowledge database here is doing something more feesable given the processor budget we actually use to make our stuff for real.

Not just crappy dumb unimplementable speculations.

Your better off with that q-learning, it runs this little dude here->  "https://www.youtube.com/watch?v=CVHd2_NUgIs&t=43s",     its the version of this that actually works by doing it with less horsepower/samples required.     Theres tutorials on the internet about it.
I think the guy in the video actually came up with it,  but i might be mistaken.

But i guess im being a bit satirical.

If you have a quantum computer you might be able to do it.
But its just easier said than actually done actually having one.

*

Art

  • At the end of the game, the King and Pawn go into the same box.
  • Moderator
  • **********************
  • Colossus
  • *
  • 5865
Re: Simulation Based AI for Robotics - Why is this not Happenning?
« Reply #5 on: August 30, 2017, 01:16:47 pm »
Nice article Ranch!
I found another posting about his Hexapod with a bit more details.

https://www.youtube.com/watch?v=xLxrb_P6N7o
In the world of AI, it's the thought that counts!

*

Kevin Long

  • Roomba
  • *
  • 4
  • LeapSpecies.Space
    • LeapSpecies.Space
Re: Simulation Based AI for Robotics - Why is this not Happenning?
« Reply #6 on: August 30, 2017, 06:39:29 pm »
Good points :)
"The universe machine"  is grandiose.  For application the sim AI only needs to update the physics etc. as needed for the operation.  That is far scaled down from the silly "universe machine".  And the evolution of the robotics in the stable is straightforward iterative approaches already proven but not so much in operational requirements.  I will continue to talk design because yes that is how design is done - it happens before implementation :)  So is this still pie in the sky?

*

LOCKSUIT

  • Emerged from nothing
  • Trusty Member
  • *******************
  • Prometheus
  • *
  • 4659
  • First it wiggles, then it is rewarded.
    • Main Project Thread
Re: Simulation Based AI for Robotics - Why is this not Happenning?
« Reply #7 on: August 30, 2017, 06:42:30 pm »
"that guy thats writing that interesting text knowledge database here is doing something more feesable given the processor budget we actually use to make our stuff for real."

( ͡º ͜ʖ ͡º)

who's that guy?

You know, I don't NEED to evolve a universe nor a body NOR an AI brain !   I already have my physics simulation universe, and I'm almost done a good amount of my HUMAN baby's body! Then after that the code comes. And no breeding nor cut-scene breaks for training/learning/improving actions.

Sorry but I had to say this...

If anyone is going to get that technical to update how good a physics simulator is, then it would take just a few weeks/months to hand-tune it by human eyesight. Just like my baby's body. And, the AI code.

Stop evolving little critters.
Emergent          https://openai.com/blog/

*

keghn

  • Trusty Member
  • *********
  • Terminator
  • *
  • 824
Re: Simulation Based AI for Robotics - Why is this not Happenning?
« Reply #8 on: August 30, 2017, 08:23:11 pm »
 @Kevin Long.
 I have been working on AI that has a internal simulator in it's head. Then to make thing more confusing,
you can have a robot AI trained and live out a lot of it's life in completely different simulator. Because not every
one can build a high  end expensive bot body. So yes it is possible for bot to live in a simulation and have a
second simulator running in its head. So where do you stand???????
« Last Edit: August 30, 2017, 09:26:28 pm by keghn »

*

Kevin Long

  • Roomba
  • *
  • 4
  • LeapSpecies.Space
    • LeapSpecies.Space
Re: Simulation Based AI for Robotics - Why is this not Happenning?
« Reply #9 on: August 31, 2017, 02:48:51 am »
@ kegn
Rephrasing what you said or combining your statements to catch up with you.  Sounds like you are developing the behavior commands of the robot by having it copy its terrain (simulation) and use that for making its decisions.  But the terrain is a simulation.  So the simulated robot is making a simulation of the simulation it is being tested in?   :)   
Seems like a great way toward operational robot development.

*

keghn

  • Trusty Member
  • *********
  • Terminator
  • *
  • 824
Re: Simulation Based AI for Robotics - Why is this not Happenning?
« Reply #10 on: August 31, 2017, 04:43:24 am »
 Ooooops. Real robots do not have internal 3d simulator, built into their brain, to model the world.
Only AGI and ASI do.  "Artificial General Intelligence' and "Artificial Super Intelligence".
 
 Yes, A virtual robot can be put into a simulation for training. And then down load into a perfectly matching
body when done.

*

LOCKSUIT

  • Emerged from nothing
  • Trusty Member
  • *******************
  • Prometheus
  • *
  • 4659
  • First it wiggles, then it is rewarded.
    • Main Project Thread
Re: Simulation Based AI for Robotics - Why is this not Happenning?
« Reply #11 on: August 31, 2017, 07:06:37 am »
Um no keghn you mean as the simulated baby lives and crawls around, it is, at the same time/life, daydreaming and testing a body in a inside internal another 3d sim. I can do that right now. As I sit here, I can see myself in a desert, on a plank, and if I fall hit spikes, and now I try to walk forward, very gently, trying to balance. I just did it. I have a internal simulator. My sim baby will end up having its own sim to daydream/dream with.
Emergent          https://openai.com/blog/

 


OpenAI Speech-to-Speech Reasoning Demo
by ivan.moony (AI News )
Today at 01:31:53 pm
Say good-bye to GPUs...
by MikeB (AI News )
March 23, 2024, 09:23:52 am
Google Bard report
by ivan.moony (AI News )
February 14, 2024, 04:42:23 pm
Elon Musk's xAI Grok Chatbot
by MikeB (AI News )
December 11, 2023, 06:26:33 am
Nvidia Hype
by 8pla.net (AI News )
December 06, 2023, 10:04:52 pm
How will the OpenAI CEO being Fired affect ChatGPT?
by 8pla.net (AI News )
December 06, 2023, 09:54:25 pm
Independent AI sovereignties
by WriterOfMinds (AI News )
November 08, 2023, 04:51:21 am
LLaMA2 Meta's chatbot released
by 8pla.net (AI News )
October 18, 2023, 11:41:21 pm

Users Online

305 Guests, 0 Users

Most Online Today: 346. Most Online Ever: 2369 (November 21, 2020, 04:08:13 pm)

Articles