my first "walker"

  • 23 Replies
  • 986 Views
*

goaty

  • Autobot
  • ******
  • 220
Re: my first "walker"
« Reply #15 on: May 01, 2019, 06:11:56 pm »
Ok ive got another update on the way.

That joke I said, about giving him two halves ill do,  because I think I get to increase its intelligence, by having a and b!

" (2^(DIMS/SPLITS)^TIMESTEPS)*SPLITS "    INSTEAD OF   " (2^DIMS)^TIMESTEPS "  looks like it gets an advantage by splitting itself.

This should change how hes walking.  Back soon, I just need a bit more of a break, then off I go!

*

goaty

  • Autobot
  • ******
  • 220
Re: my first "walker"
« Reply #16 on: May 02, 2019, 01:24:37 pm »
I did it,  and it does have increased "sight", wasn't much different of a style of walking except it took bigger strides. and in one setting at the beginning it jumped along instead of walked, with 1 brain on each leg, but it was uncoordinated how it walked.  I could upload it, but its not that exciting yet, I think I can get it better.

But I think it can actually work.    I think my problem is I haven't described enough context for it to predict properly,  and that means I have to go from a quick array to a more slower method of matching, which includes a bigger key.  At the moment the key is too small, and its confusing its predictions. (double matching)

I think its important to do this,  to A) get more bruteforced at the same cost.  and B) the robot has to have split mindedness to predict the environment anyway, so if I can get the robot predicting its other leg, its a good sign that im ready to add the environment to it.

Ive got my goal written down for the "guard starfish" im going to make,   ive had it for a while,  and I think that you can definitely make some funny things happen with modifying the goal, even adding a state machine to it!   

Ok that's all for now.

*

goaty

  • Autobot
  • ******
  • 220
Re: my first "walker"
« Reply #17 on: May 02, 2019, 03:47:01 pm »
Im using a knn memory to store the other leg positions as it goes,  its working, nice and quick and I get a bigger search depth,  but I am having hiccups still...  its worse when its on than when its off,  so I think I need to increase the context.

https://www.youtube.com/watch?v=9CrqBiKVGeQ&feature=youtu.be
« Last Edit: May 02, 2019, 08:28:16 pm by goaty »

*

LOCKSUIT

  • Emerged from nothing
  • Trusty Member
  • ****************
  • Admiral
  • *
  • 3127
  • First it wiggles, then it is rewarded.
Re: my first "walker"
« Reply #18 on: May 02, 2019, 08:49:56 pm »
I expected to see 2 glued together, like a 4-arm. Haha, it walks like a drunkard.
Emergent

*

goaty

  • Autobot
  • ******
  • 220
Re: my first "walker"
« Reply #19 on: May 02, 2019, 09:07:23 pm »
Its a separate brain/search each leg!   that would be confusing no?
I decided to split it down the middle, with 3 dimensions either side.

Its not predicting itself properly,   but its only a 32 bit key,  so im going to bust it up to 1024,  and see if it goes better.   

Its using similarity matching,  not exact matching,  exact matching I think now is only good for symbolic data like text.

*

goaty

  • Autobot
  • ******
  • 220
Re: my first "walker"
« Reply #20 on: May 02, 2019, 10:21:49 pm »
https://www.youtube.com/watch?v=nzjT-QJQctI&feature=youtu.be  <-HERE! GOT IT!   it gets to search twice as far as long as my prediction of the opposite side is correct,   and don't think your going to get away with a small key and without near matching.  it has to be fully fledged or it doesn't even work. (im doing things for real now, and all my baby theories are dieing.)

Its cheat sliding against the ground,  so Im going to fix the physics more now.

* REMEMBER BACK FIRE ROTATION -  you push your leg, and it pulls the rest of the body in the opposite direction.
* THE WEIGHT OF THE LIMBS, puts them in a permanent rotation with each other due to gravity.

- and then theres problems with contacting the floor,  like im saying whats happening to me.

[edit] actually, maybe it needs a bit more work...  but its almost as good as the whole bodied one. [/ed]
« Last Edit: May 02, 2019, 10:59:32 pm by goaty »

*

goaty

  • Autobot
  • ******
  • 220
Re: my first "walker"
« Reply #21 on: May 22, 2019, 05:39:33 am »
ive switched over to a tendon based system.   looks more complex but theres actually less code than the toothpick one,  which had all these plane tests, knees knocking inside out...   this looks better but his knees are working yet, and it seems to be pinned to the floor.

https://www.youtube.com/watch?v=RneW9SU1IZg

more updates soon...

*

goaty

  • Autobot
  • ******
  • 220
Re: my first "walker"
« Reply #22 on: May 22, 2019, 01:50:56 pm »
Tendon based robbers are indeed soft-body, it looks like a jelly fish.
Ok, a bit more work.   all the dims are pulsating when I toggle all the tendons, but for some reason when the search is on, one of the knees goes into remission.
That's just a side problem,   the main problem here is when the hip rotates it rotates the top instead of the leg, and the feet aren't getting traction or push up from the knee.   But its definitely going better,  im still confident this way is easier with less code, I just need to fix it,   ill leave it for now,   give me 2 days and this one will be jogging along nicely.   Im thinking maybe the top part of it needs to be heavier than the hip or the hip will rotate the body instead of the leg.

That means when they make bottom halfs of robots, they need a weight at the centre that connects the legs together,  if its true.

that's it, had enough for the day.

One thing im going to try, is if I make the "tendon cubal" longer, then itll give it more freedom of rotation.   long and skinny bellows go for more angular freedom.

https://www.youtube.com/watch?v=9A22rdkU24Y&feature=youtu.be

*

Korrelan

  • Trusty Member
  • ***********
  • Eve
  • *
  • 1379
  • Look into my eyes! WOAH!
    • YouTube
Re: my first "walker"
« Reply #23 on: May 22, 2019, 02:16:44 pm »
I think you are doing an excellent job.

Does the physics model include mass inertia?

 :)
« Last Edit: May 22, 2019, 03:45:21 pm by Korrelan »
It thunk... therefore it is!...    /    Project Page    /    KorrTecx Website