[MagnusWootton]

Ok.  so i've rewritten this thing about a million times now.    now again, only some shaders to go,  then i suddenly just.... cant... code anything.

[ivan.moony]

What are you trying to do?

[MagnusWootton]

These little guys are supposed to be running along, not rolling on their backs.    But I can never get it done.

If these ones work,  the net is built to do the same thing in practical physics as well as virtual with no change in the code.

[LOCKSUIT]

Ok.  so i've rewritten this thing about a million times now.    now again, only some shaders to go,  then i suddenly just.... cant... code anything.

:D Those are fun yes. Google is going at their own again here, they got as far it seems as the robot dog does a moon walk xD, and the robot hand picks up the apple and puts it in the droor and closes it, among many other things shown:

https://twitter.com/_akhaliq/status/1669538223335387136

https://language-to-reward.github.io/


BTW this one is a good one to watch for the learnign how to get it to walk and avoid walls etc, really vivid colorful and shocking video:



With GPT-4 and all these AIs, we so got AGI coming before 2029. GPT-4 has been made to be talkative and instruction following, while music and video AIs and robot AIs are coming along far being able to do all sorts of tasks now. It's profound, I'll share the 200+ AIs from the last couple years one day soon to show AI progress picking up speed.

[WriterOfMinds]

If it were easy, everyone would do it. It's okay if it takes a while!

[MagnusWootton]

Thanks Writer of minds,  your being a good influence on me,  my attitude is deteriorating lately.

I really could get this done in a couple of days, but im in full burnt out mode now, and i just have been stuck on completing this thing for years now.

Its hard to explain to people how burnt out you really can get,  to stupendous amounts of it.

Thanks Locky!   Always good to hear from you,  if I get it done, could make me really rich!    Mine should learn very quickly, and it will be able to actually dodge the balls coming at it, even catching balls.    And!  not just in the computer.

[MagnusWootton]

Bit of a depressing update.

I've got alot more done, the neural network is in, and i've got it very correct, its nearly ready to walk across now, but the prob is my video card has started crashing under the training workload.   I train the net synapses, but then I lose the pointer to the video card afterwards.

Goes to show u should get nvidia cards not random no-brand cards cause they have high core counts on their sheets but when they consistently crash and burn when u expend enough power on them its useless having the number written there as high.

Ive also started on the computer vision system as well,   and thats going to integrate a second memory system type other than the perceptron for all the similarity matching im needing.

So, I'm not sure what to do,  I have a computer getting fixed at the shop ready to come back with my nvidia card on it,  nearly ready soon, but Ive started worrying about the fact maybe when the power supply died it took the motherboard and ram with it or something,  with a bit of luck the crowbar circuit worked and saved the rest of the computer, but who knows I might be without a computer completely!

My best bet is either living at my little bro's house a while and jump on his giant whopping 10,000 core gpu, where I've never been above 2000 ever, so itll be cool using that,  but thats only temporary, so I'm thinking about getting a second hand predator laptop (they pack good video cards) for $1000 and then ill be off and running again.

But I don't have $1000, ill have to see if I can convince my mum to lend it to me,  she should, but I don't know how long itll take for her to work out I need the fricken computer, before she ends up giving me the green light.

So,  no video card.   I guess i'll still keep coding,  but the demo to show looks like its not coming for the near future, like my last videos,  where I wanted to show you guys something cool but I couldnt quite make it.
So I'll see what happens,  but the video card on this comp I'm on now, the video card is burnt out, cheapo crap that doesnt even run that well.

[infurl]

If it's crashing but not actually destroying itself maybe you improve the reliability by improving the cooling.

You can even run everything without a case to maximize cooling.

https://www.cgdirector.com/turn-on-motherboard-without-case-switch/

[MagnusWootton]

yeh! definitely I think its over heating.


But just to show im fully serious about this thing, ive got my open concept filter to show->>>   (its my little uzi system,  very proud.)


//setup robot
  //1) VERTICES
  //2) INDICES
  //3) HINGES
  //4) TENDONS
  //5) ADJACENCY MESH
  //6) SETUP VERTEX ORDERS
 
//PPU
  //1. APPLY HINGE ACCELLERATIONS
  //2. COMPUTE TENDON LENGTHS
  //3. INTEGRATION
  //4. COMPUTE CENTRE OF MASS
  //5. COMPUTE GROUND COLLISION
  //6. DESTRETCH STICKS
  //7. DESTRETCH STICKS DAMPENER (doesnt work without this!!!)

 ///////////////////////////////////////////////////////////
 //
 //ROBO MOTOR SEARCH

 //1) ADVANCE MOTOR STRING
 //FOR ALL SIM EPOCHS
   //2) set scene input (hinge+simperms)
       //FOR ALL MEMBERS
          //FOR ALL FRAMES IN FRONT
             //3) get scene output
             //4) integrate
             //5) motivate
             //6) GET MAX (hinge + simperm)
             //7) UPDATE MOTORSTRING

//viz for now->

     //CORNERS    +tracker to this.
     //MATCH
     //DEVELOP3D
    // + scanner system.
    // get tra+rot for all points
    // form groups and get skeletons
    // u have to predict points accellerations you cant see. (coming out of view, and back in as long as it can keep it whole as it goes.)
     // - map to memory that will convert the photo units, to skeleton id and coordinates per point.
    // - motivation brute force for opponents, single behaviour chain -> n guessed motivations pick closest match.

///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
 // and u want it so its trainable and fully testable and programmble virtually without its body using practical record!!!


THATS THE WHOLE THING!!!  just missing some of the compviz and the 3d eye isnt working yet,
but all the motor is in,  JUST WAITING ON THE VIDEO CARD!!!!

This hasnt got the commanding system written here tho,  im working on that later but its basicly just swapping what mode its in and changing its target object, pretty much,  but if it doesnt do decisions itself as well its nowhere near as good,   you can only command something so much before its got to do the right thing by itself.

[MikeB]

If a video card is older than 5 years it helps to repaste it. Thermal paste dries up eventually and doen't work well... The Honeywell PTM7950/PTM7958 is one of the best (long lasting).

If you need a laptop with RAM, some Dell Precision & HP Zbooks have Xeon processors and support up to 128MB ECC RAM. Can't get a super modern GPU for under $1k though.

[MagnusWootton]

This is the one I'm looking at,   its a 2060,  i wish it was a 2070 but its still ok I bet,   so this could go in a robot body I think.
Its got just under 2000 cuda cores.

https://www.ebay.com.au/itm/185902568085?chn=ps&_ul=AU&_trkparms=ispr%3D1&amdata=enc%3A1rkVOWGAYQDe1mQ8TFXWQqw12&norover=1&mkevt=1&mkrid=705-139619-5960-0&mkcid=2&mkscid=101&itemid=185902568085&targetid=1598469862518&device=c&mktype=pla&googleloc=9070532&poi=&campaignid=19657035767&mkgroupid=143201283022&rlsatarget=pla-1598469862518&abcId=9305369&merchantid=9516932&gclid=Cj0KCQjwho-lBhC_ARIsAMpgMoenCAq-0yg1flly-0I20sePpnDhrKEktTrkA6Yg-IZFG8UgUWogvj0aAvyXEALw_wcB