Introducing the STALKER! (project ended)

tnay · May 30, 2009, 10:44am

Thanks for pointing out, actually suppose to be Z in stalker. Hehe…

Innerbreed, I sent you the final transformation code. Lets work it out behind the scene. I have posted too much in this thread which none has work…

innerbreed · May 30, 2009, 8:39pm

great work. its really nice that so many of you are willing to help.

I have added the code and unfortunatly it knocks all the ZXY axis out.
I cannot get any control using both J-Sticks.

it may just need tweaking.

innerbreed · May 30, 2009, 9:58pm

OK i got control again. a rouge command was left unnoticed.

am having battery problems. i charge up both batts (1600 and a 2800) but im only getting about 1min of play out of them at a time intill the servos start to die.

so no further testing tonight but i have managed to see what the current code is doing and it doesnt seem to be responding in the way we had hoped.
i will post a vid or something tomrrow night.

it could just be having uinsufficient power.

would we need to change any of this?

[code];IKCoxaAngle
GOSUB GetBoogTan [IKFeetPosZ, IKFeetPosX]
IKCoxaAngle = TOINT((BoogTan*180.0) / 3.141592)

IF(IKCF < TOFLOAT(FemurLength+TibiaLength-30)) THEN
	IKSolution = POSITIVE ;Output POSITIVE IF the solution is possible
ELSE
	IF(IKCF < TOFLOAT(FemurLength+TibiaLength)) THEN
		IKSolutionWarning = POSITIVE ;Output POSITIVE IF the solution is NEARLY possible
	ELSE
		IKSolutionError = POSITIVE	 ;Output POSITIVE IF the solution is NOT possible
	ENDIF
ENDIF

return[/code]

innerbreed · May 30, 2009, 10:36pm

euclideanspace.com/maths/geo … /index.htm

and

uwgb.edu/dutchs/structge/SL1 … liqAxH.HTM

this is a great site that i have been browsing over for the last 24 hours.
here it explains more about Axis Vs Angle and Euler Angles and Rotation Points Around an Arbitrary Axis.

the math is quite well documented but understanding most of it is not my field unfortunatly.

i have tried to find anything on tilted axis but i cannot find it. someone else might have more luck.

it may help.

innerbreed · May 31, 2009, 12:15am

I have seen equations that rotate 3d data around a 3d (x,y,z) axis. I don’t understand that complex equations. Anyway,

function rotateZ(x, y, center, dAngle) { //note: here dAngle is in arc unit PI format var dy = y-center.y; var dx = x-center.x; var orgAngle = Math.atan2(dy, dx); var hypo = Math.sqrt(dy*dy+dx*dx); var newAngle = orgAngle+dAngle; var xx = hypo*Math.cos(newAngle)+center.x; var yy = hypo*Math.sin(newAngle)+center.y; var pt = {x:xx, y:yy}; return pt; }

Old Text i found:

*The same math equation can be modified easily to take (z,x) to rotate around Y axis, take (y,z) to rotate around X axis. But what if the axis is not straight Z, but has a little deviation? Then we can’t take only (x,y), we need to handle z.

Well, I said that, I dont understand this complex math. So, I try to find a work-around. I make it by several steps. Each step is a rotate around coordinate axis. First I rotate it around Z axis, to make the axis in the YZ plane. Then I rotate it around X axis to make the axis the same as Y axis. OK, we can easily do a rotation around Y now.

After this, I counter-rotate it by the same angle as above to go off the Y axis back to the YZ plane. Then I counter-rotate the angle to make the axis back to the original (x,y,z) vector. That is it. Take a pen and paper, it should not be difficult to find out during steps what angle should be rotated. *

Would this be a step in the right direction?

tnay · May 31, 2009, 3:19pm

Mind to share what was the problem before?

innerbreed:

would we need to change any of this?

[code];IKCoxaAngle
GOSUB GetBoogTan [IKFeetPosZ, IKFeetPosX]
IKCoxaAngle = TOINT((BoogTan*180.0) / 3.141592)

IF(IKCF < TOFLOAT(FemurLength+TibiaLength-30)) THEN
	IKSolution = POSITIVE ;Output POSITIVE IF the solution is possible
ELSE
	IF(IKCF < TOFLOAT(FemurLength+TibiaLength)) THEN
		IKSolutionWarning = POSITIVE ;Output POSITIVE IF the solution is NEARLY possible
	ELSE
		IKSolutionError = POSITIVE	 ;Output POSITIVE IF the solution is NOT possible
	ENDIF
ENDIF

return[/code]

The above code is to limit the IK calculation solution when it approach the reach limit. I am writing similar code for my quad as well. Without these code, the leg will go hair-wire! It’s ok to leave it alone

innerbreed:

innerbreed:

http://www.euclideanspace.com/maths/geometry/rotations/index.htm

and

uwgb.edu/dutchs/structge/SL1 … liqAxH.HTM

this is a great site that i have been browsing over for the last 24 hours.
here it explains more about Axis Vs Angle and Euler Angles and Rotation Points Around an Arbitrary Axis.

the math is quite well documented but understanding most of it is not my field unfortunatly.

i have tried to find anything on tilted axis but i cannot find it. someone else might have more luck.

it may help.

I have seen equations that rotate 3d data around a 3d (x,y,z) axis. I don’t understand that complex equations. Anyway,

Old Text i found:

To rotate data around x or y or z axis is easy. For example, we rotate (x,y,z) around Z axis, then we can ignore z value and take (x,y) in 2D. To rotate (x,y) in 2D for an arc angle delta, we just convert (x,y) data into radius and angle. The radius is the distance (x,y) to (0,0); that is Math.sqrt(xx+yy); The angle arc in PI format is Math.atan2(y,x); now rotate a delta angle. The radius is the same. While the angle should increases by delta. So, the new (x,y) is x=radiusMath.cos(angle+delta); y=radiusMath.sin(angle+delta).*

function rotateZ(x, y, center, dAngle) { //note: here dAngle is in arc unit PI format var dy = y-center.y; var dx = x-center.x; var orgAngle = Math.atan2(dy, dx); var hypo = Math.sqrt(dy*dy+dx*dx); var newAngle = orgAngle+dAngle; var xx = hypo*Math.cos(newAngle)+center.x; var yy = hypo*Math.sin(newAngle)+center.y; var pt = {x:xx, y:yy}; return pt; }

Old Text i found:

*The same math equation can be modified easily to take (z,x) to rotate around Y axis, take (y,z) to rotate around X axis. But what if the axis is not straight Z, but has a little deviation? Then we can’t take only (x,y), we need to handle z.

Well, I said that, I dont understand this complex math. So, I try to find a work-around. I make it by several steps. Each step is a rotate around coordinate axis. First I rotate it around Z axis, to make the axis in the YZ plane. Then I rotate it around X axis to make the axis the same as Y axis. OK, we can easily do a rotation around Y now.

After this, I counter-rotate it by the same angle as above to go off the Y axis back to the YZ plane. Then I counter-rotate the angle to make the axis back to the original (x,y,z) vector. That is it. Take a pen and paper, it should not be difficult to find out during steps what angle should be rotated. *

Would this be a step in the right direction?

These are really good reference. I didn’t go through all the line, but from your explaination, you have already got the concept. But the old test you found, the one which you refer to, is not correct for your application. Sorry to point out, but the code you wrote is incorrect

By the way, I have PMed you some new codes, just hours after my last post here.

I added negative(-) sign in front of both equation. I have shared the new IK model, in excel format.

zenta · May 31, 2009, 3:31pm

You are scratching to some of it…

I believe using a successive rotation about global cartesian axes is one way to go. Remeber the correct order of rotation: Y rotate followed by a Z rotate (I’m refering to the axis used in PEP and Xan’s code). I also sent you a PM I hope you don’t get confused

innerbreed · June 3, 2009, 5:02pm

Ok while trying to get the tilted coxa calculations corrected i have also hit
another problem. even with the tilted coxa calculations sorted i would still
need to sort out the COG. now im open to all answers “but” adding
motors/servos to redirect the COG while walking really isnt on my list.

Here you can see the idea i wish to impliment into my code.
youtube.com/watch?v=Q2YAZHRZ … re=related

i need to work out how to program this.
i think the intended plane would be Z & X and the body would have to move
into the triangle created by the tripod in each step. (as each leg lifts)

im not sure how to get these globe positions “on the fly” and add them to the code.
any help will be critical.

how is this done?

with this being said i would also need to get the timing right so that once
the COG has been transfered away from the fly leg i will need it to return
back when leg is returning to ground again.

any ideas?

km6vv · June 3, 2009, 5:42pm

That’s certainly a nice walking quadruped!

Are you saying that the body shifts around to move the COG as it walks? the gait looks the same as this one:

youtube.com/watch?v=lxIy3jYuQCo&feature=related

But I guess it does move it’s body around more. I like this video for explaining a quad gait; although I’m told it’s not very good?

youtube.com/watch?v=wt_k2aoF6l0

I can get a feel of moving the body 1/4 the travel as each leg moves in succession.

And now the notion of “Body IK” has been introduced. I guess I’m way behind!

Alan KM6VV

innerbreed:

Here you can see the idea i wish to impliment into my code.
youtube.com/watch?v=Q2YAZHRZ … re=related

i need to work out how to program this.
i think the intended plane would be Z & X and the body would have to move
into the triangle created by the tripod in each step. (as each leg lifts)

im not sure how to get these globe positions “on the fly” and add them to the code.
any help will be critical.

What you need to do is take this calculations and exclude the flying leg. This should place the body between the other legs which are on the ground. You need to build something around it which places the body between the 3 legs before you lift the leg but it should do the job.

Xan

with this being said i would also need to get the timing right so that once
the COG has been transfered away from the fly leg i will need it to return
back when leg is returning to ground again.

any ideas?

innerbreed · June 3, 2009, 5:59pm

hi alan.

well the first vid (robot) called StarLight using the same motion as a lizard and can only transverse in one direction. non-omnidirectional.

“Park” did a great job with his creation. it was one of his works from uni.

you may think the second vid does show the same kind of loco as the first but in slow motion. but if you study the gait steps you will also see that park (first vid) using a Ripple gait, were as the second uses Wave.
the body

J

km6vv · June 3, 2009, 6:17pm

Yes, you’re right. So Ripple is move legs on one side and then the other, while Wave alternates the sides? I’ll probably need to watch these vids over and over to get the full flavor of the differences!

I can appreciate that StarLight does have a lot more body “swagger” in his gait. But you say no translation?

Maybe we need to define or differentiate more on whether we are wanting to do sprawled (reptile/insect) or upright (mammalian) stances? The classic hexapod where we’re wanting to start from is of course sprawled. Dogs, Cats, Horses have an upright stance. and there appears to be considerable discussion on whether Big Dog and Little Dog have sprawled or upright stances. They do appear to have rotated-under-the-body legs.

Alan KM6VV

innerbreed · June 3, 2009, 6:17pm

ok i think i have a bisic idea.

TotalX var sword ;Total X distance between the center of the body and the feet TotalZ var sword ;Total Z distance between the center of the body and the feet TotalXBal var sword TotalZBal var sword BodyOffsetX var sbyte ;Input Offset betweeen the body and Coxa X BodyOffsetZ var sbyte ;Input Offset betweeen the body and Coxa Z PosX var sword ;Input position of the feet X PosZ var sword ;Input position of the feet Z TotalTransX var sword TotalTransZ var sword

innerbreed · June 3, 2009, 6:21pm

KM6VV:

Yes, you’re right. So Ripple is move legs on one side and then the other, while Wave alternates the sides? I’ll probably need to watch these vids over and over to get the full flavor of the differences!

I can appreciate that StarLight does have a lot more body “swagger” in his gait. But you say no translation?

Maybe we need to define or differentiate more on whether we are wanting to do sprawled (reptile/insect) or upright (mammalian) stances? The classic hexapod where we’re wanting to start from is of course sprawled. Dogs, Cats, Horses have an upright stance. and there appears to be considerable discussion on whether Big Dog and Little Dog have sprawled or upright stances. They do appear to have rotated-under-the-body legs.

Alan KM6VV

innerbreed:

hi alan.

well the first vid (robot) called StarLight using the same motion as a lizard and can only transverse in one direction. non-omnidirectional.

“Park” did a great job with his creation. it was one of his works from uni.

you may think the second vid does show the same kind of loco as the first but in slow motion. but if you study the gait steps you will also see that park (first vid) using a Ripple gait, were as the second uses Wave.
the body

J

have you been here yet?
you may remember some of this stuff from the group.
lynxmotion.net/phpbb/viewtop … c&start=15

zenta · June 3, 2009, 6:37pm

Hi I’m quoting myself again about the balance part:

I believe this method can’t be used to keep the COG inside one of the tripods.

Interesting topic though. I hope to put my mind into the quad challenge in the near future. But starting to study others work is a good thing. I hope you find a method for this. I remember I saw an excel sheet of a walking quad one time where the tripod and COG where calculated on the fly (on this forum).

Wait, that was tnay, again :

Pretty awesome, especially sheet #3.

km6vv · June 3, 2009, 6:50pm

Yes, I had been there briefly, but hadn’t made it back. These videos would be good to reference there as well.

Have you worked out any sort of “phase chart” for what you want to do with your body IK and leg IK? Might be interesting if something could be worked out in a spread sheet, that would generate a chart or graph.

EDIT:

Looks like KÃƒÂ¥re has just posted on an XLS that does just that!

Alan KM6VV

innerbreed · June 3, 2009, 7:58pm

tnay · June 4, 2009, 1:30pm

innerbreed:

Ok while trying to get the tilted coxa calculations corrected i have also hit
another problem. even with the tilted coxa calculations sorted i would still
need to sort out the COG. now im open to all answers “but” adding
motors/servos to redirect the COG while walking really isnt on my list.

Here you can see the idea i wish to impliment into my code.
youtube.com/watch?v=Q2YAZHRZ … re=related

i need to work out how to program this.
i think the intended plane would be Z & X and the body would have to move
into the triangle created by the tripod in each step. (as each leg lifts)

im not sure how to get these globe positions “on the fly” and add them to the code.
any help will be critical.

Xan From earier thread:

Hi Innerbreed,

V1.3 includes ZentaÃ¢â‚¬â„¢s balance calculations. One of the things it does it calculate the middle position between the legs and places the body there.

how is this done?

What you need to do is take this calculations and exclude the flying leg. This should place the body between the other legs which are on the ground. You need to build something around it which places the body between the 3 legs before you lift the leg but it should do the job.

Xan

with this being said i would also need to get the timing right so that once
the COG has been transfered away from the fly leg i will need it to return
back when leg is returning to ground again.

any ideas?

Have you got stalker walking already? Sorry, I can’t open the youtube link, some kind of error (or perhaps block by local server ) Are you still using the wave 18 steps? Anyway, I ran a sim on the code. The result doesn’t look good COG is out of the tripod some part of the cycle. Just a hunch, try this:

[code] IF (GaitType = 0) THEN ;Wave 18 steps
LRGaitLegNr = 10
RFGaitLegNr = 14
RRGaitLegNr = 1
LFGaitLegNr = 5

NrLiftedPos = 2
HalfLiftHeigth = NEGITIVE
TLDivFactor = 16
StepsInGait = 18
TitularGSpeed = 100
ENDIF
return[/code]

I don’t know how xan/zenta wrote this, but assuming this part of code is independent of the rest; it should at least work for forward gait.

xan · June 4, 2009, 2:14pm

These are the settings for the gait engine. You can find some aditional information about this on my Project page. Let me know if there are any more questions about this part. (or any part of the code )

Xan

innerbreed · June 4, 2009, 3:32pm

i havent tried the gait you supplied yet but i soon will.

Been looking at the GAIT section of the code:

[code];-----------------------------[GAIT]
Gait [GaitLegNr, GaitPosX, GaitPosY, GaitPosZ, GaitRotY]

;Check IF the Gait is in motion
GaitInMotion = ((ABS(TravelLengthX)>ps2Tolerance) | (ABS(TravelLengthZ)>ps2Tolerance) | (ABS(TravelRotationY)>ps2Tolerance) )

;Leg middle up position
;Gait in motion Gait NOT in motion, return to home position
IF (GaitInMotion & (NrLiftedPos=1 | NrLiftedPos=3) & GaitStep=GaitLegNr) | (GaitInMotion=NEGITIVE & GaitStep=GaitLegNr & ((ABS(GaitPosX)>2) | (ABS(GaitPosZ)>2) | (ABS(GaitRotY)>2))) THEN ;Up
GaitPosX = 0
GaitPosY = -LegLiftHeight
GaitPosZ = 0
GaitRotY = 0
ELSE

;Optional Half heigth Rear
IF ((NrLiftedPos=2 & GaitStep=GaitLegNr) | (NrLiftedPos=3 & (GaitStep=GaitLegNr-1 | GaitStep=GaitLegNr+(StepsInGait-1)))) & GaitInMotion THEN
  GaitPosX = -TravelLengthX/2
  GaitPosY = -LegLiftHeight/(HalfLiftHeigth+1)
  GaitPosZ = -TravelLengthZ/2
  GaitRotY = -TravelRotationY/2
ELSE
  
  ;Optional half heigth front
  IF (NrLiftedPos>=2) & (GaitStep=GaitLegNr+1 | GaitStep=GaitLegNr-(StepsInGait-1)) & GaitInMotion THEN
    GaitPosX = TravelLengthX/2
    GaitPosY = -LegLiftHeight/(HalfLiftHeigth+1)
    GaitPosZ = TravelLengthZ/2
    GaitRotY = TravelRotationY/2
  ELSE  	  

  	;Leg front down position
  	IF (GaitStep=GaitLegNr+NrLiftedPos | GaitStep=GaitLegNr-(StepsInGait-NrLiftedPos)) & GaitPosY<0 THEN
      GaitPosX = TravelLengthX/2
      GaitPosY = 0
      GaitPosZ = TravelLengthZ/2
      GaitRotY = TravelRotationY/2

  	;Move body forward      
  	ELSE
      GaitPosX = GaitPosX - (TravelLengthX/TLDivFactor)
      GaitPosY = 0
      GaitPosZ = GaitPosZ - (TravelLengthZ/TLDivFactor)
      GaitRotY = GaitRotY - (TravelRotationY/TLDivFactor)
    ENDIF
  ENDIF
ENDIF

ENDIF

;Advance to the next step
IF LastLeg THEN ;The last leg in this step
GaitStep = GaitStep+1
IF GaitStep>StepsInGait THEN
GaitStep = 1
ENDIF
ENDIF

return[/code]

by adding another section to this where the body moves forward

i think i can work something out. the bit id like to add will allow the body to move 'not only forward but also to the oposite side of the fly leg using the X & Z axis, like a diagonal move rather than just forward.

eg. If the front left leg lifts then the body moves forward to the front right leg by say, +1 and also to the side say +1.5 moving the body away from the fly leg but still walking in a forward direction. get me? iv confused my self now.

would this work? can it be done?

tnay · June 5, 2009, 5:19am

I remember first saw it in your thread (Xan’s Phoenix Code). Went back to that thread but couldn’t find the wave 18 steps, but the few examples are good to give me a head start.

I have not look into those codes, though it’s there… hehe… Since innerbreed is using the same code as phoenix, except the middle leg, so I guess the code should work

The idea is good, like sine-walking robot? The GaitPosX,Y,Z and GaitRotY is picked up by another subroutine, need to look into that first before making any changes. Just a quick look, the Gait function does not include any variable that mention which leg is on the fly. You need to call another variable (or existing one) and put it into the equation so that it know which side (left or right) to move.