Testing Newton for Industrial Robtics

by **DarthPhysics** » Fri Jan 18, 2008 4:06 pm

Heya all

I just want to show you what a noob in newton with the help of a MASTER (Hi Dave :lol:

) can do.
http://fr.youtube.com/watch?v=4XGqG6EI9ek
Well , it's the start of my project and it will take alot time before i can
do what i really want :wink:

But now i will need some info about the moving/rotating of a dynamic object without to use force/torque etc.
As i know i cant directly set position or rotation, because the physic is disable during this process ,but in this case it's very critical.
A real Robot have to stay on path , and as i have see in all my try , is that
the robot move his end-effector (the red sphere in middle the tool) like as all axis are sustended with eleastic :shock:

when i use force , torque , etc ...
The robot stay on a stable static base , the good thing will be that the end effector can have the same property but in the case of this object move,
all other axis follow this object , and that most stable as possible.
Is anything existing or planned in future version to do this ??

Cya all

PS : sorry for my English , i'm French :wink:

by **Julio Jerez** » Fri Jan 18, 2008 4:25 pm

Master K00m show me the video.
He also mention you are having some problem with the joint stiffness.
But I see in you third video you have much better result.
I mentioned to him that there is a Newton version with much strict tolerances that I send to some people doing work in robotic and academic simulations .

It does not make the problem go away because 100% but it is much stiffer. about 100 times stronger,
I had not send this special build to him yet but I will this weekend.

There will be included wit the next version so people will not have to request special build for realistic simulation, so the engine will be configurable for both games and real time simulation.

Very nice job

by **DarthPhysics** » Fri Jan 18, 2008 5:01 pm

Yes , on this video we dont see any problem with the joints.
It's just because i dont do "forcing" at axis limit :wink:

This one was to show that the TCP 'the red ball' can be controlled with
forces.
But if you have see a robot work for real , then you will see my actual problem

As i explain to Dave (K00m) , what is needed here is a fixed point or a object in place of the red ball , and when this point move or orient , the rest of the robot follow.
if you prefer , a robot have 2 fixed point , the base and the end effector , the only difference is that the end effector can move and orient.
The day i go program a robot in a company and the result are like what you see on this video , i'm fired

And yea , it will be very good to have stiffer 100 time stronger

I just hope oxNewton is compatible with your new dll :shock:

Cya.

by **Julio Jerez** » Fri Jan 18, 2008 5:11 pm

DarthPhysics wrote:But if you have see a robot work for real , then you will see my actual problem
As i explain to Dave (K00m) , what is needed here is a fixed point or a object in place of the red ball , and when this point move or orient , the rest of the robot follow.
if you prefer , a robot have 2 fixed point , the base and the end effector , the only difference is that the end effector can move and orient.
The day i go program a robot in a company and the result are like what you see on this video , i'm fired

edit:
For a path planning for the robot applying such big displacement is what make it the elastic behavior.
the displacement create too strong side forces on joints.
A better way would be taking the trajectory and decompose it into small displacement, alligned to the free DOF of the robot (the motors).

That way the robot will not feel strong stress on the constraints and only the free DOF will apply the strong torques.
I am surprised that it even work at all when you apply mouse displace directly to the end effector.

I can put a small demo using this technique in the SDK, and you can get it form Master K00m

Basically the promblem can be stated as:
you want to move the red ball and the robot have to keep up with it. is this coprrect?

By the powaaaaaaaaaaaaaaa of teh darksite.

by **DarthPhysics** » Fri Jan 18, 2008 5:27 pm

Basically the promblem can be stated as:
you want to move the red ball and the robot have to keep up with it. is this coprrect?

Is exactly that

move and orient the red ball , and the rest follow ... (just the robot base stay fixed)

I have try the Force from the Dark Side but it dont help :wink:

Nah !! , padawan i am ....

Cya.

by **Julio Jerez** » Fri Jan 18, 2008 5:32 pm

I got you cover, give some time to put it together.

DarthPhysics wrote:I have try the Force from the Dark Side but it dont help

That's because you had been trying the jedi site, you need to try the secret powaaaa of darkside, did you know that once there was a dark master master that was so poerwful he could make robots?

by **DarthPhysics** » Fri Jan 18, 2008 6:01 pm

Before i forget .
I have start this project for control a real robot.
It's my first proto and it move very well

The only i need now is that newton compute each axis angles :wink:

http://img138.imageshack.us/my.php?image=imag0326rh5.jpg

by **DarthPhysics** » Sat Jan 19, 2008 1:03 pm

Today a new video to show how the robot try to orient his tool

http://fr.youtube.com/watch?v=2vqii4AlwDg

You will see that the use of Torque is the wrong way , he he ....
The robot just never stop his movement , just because he never reach
the good orientation (Red ball Look at Green Ball).
I just added the Green ball that you can see where the TCP have to orient.
So , i think that is now all done arround all force type in this project and
as you see , hmmm , it's not what we have to use for the TCP :wink:

Julio you say :

A better way would be taking the trajectory and decompose it into small displacement

I'm ok with this , is what i have planned to use for Linear interpolation movement.
But Robot can move point to point in 'Joint' mode , and then he just look that he reach his final position faster as possible without look on the path.
This 'dangerous' way is just used in real robot for move fast in free space.

The force be with you

by **Julio Jerez** » Sat Jan 19, 2008 5:47 pm

can you tell me how are to appliyin the for, of the displacement?
are you addiing just a force to the end effector in order to steer it to the target?

by **DarthPhysics** » Sat Jan 19, 2008 6:50 pm

Hi

I apply x,y,z Force for positionning directly to the TCP (red ball).
For the orientation of the gripper , the pitch,turn,roll torque are applyed to
the 6th Axis from the robot (the part before the gripper).
I dont use any other force in the robot arm , and i dont want do like this,
because in this case it will be the same sh*t than use a real inverse kinematiks system with all his limitation.
Your physics engine work alot better than every programmed IK i have see.
And because of this , i never want see jacobian , CDD or other IK again :wink:

Now is just the positionning and orientation to do , but with no force in use.
Ho !! , i forget this , on the video , the gravity is set to zero

If i add gravity , and i will because the robot have to take and place objec that not have to fly alone , the robot become very instable and the force for play with are very huge.
Other problem if i add gravity and use force for the TCP , every robot model have another mass , then for each robot it will be another amount of force to use.
And it will be very hard to found how much.
So , now you know all

Regards.

by **DarthPhysics** » Sun Jan 20, 2008 1:10 pm

Hi

I just got an idea ....
As i see , alot peep's are Ragdoll fanatic (Just look how much k00m ragdoll demo exist :wink:

)
As i know from some 3D soft (3DS Max , maya, etc) , a normal 'ragdoll' reference point is the hip.
Now , how do you move the hand to a precise position and orientation ?
I just realise that a Robot work like a Ragdoll !! :shock:

It's just another 'form' ... (We are all robots :wink:

)
Then if you can use NGD for give any position to the hands , foot etc , and that when the hip stay in place , then it is possible to move a robot end effector too , with the same technic ( i hope ).
What do you think about this ???

Cya

by **Epimetheus** » Sun Jan 20, 2008 6:16 pm

If i've gotten this right your thinking of IK (inverse kinematic) which really isn't the same as ragdolls, a ragdoll is just a series of bodies connect via joints (or sometimes motorized joints)

IK just does the calculations for the joints inbetween two fixed points (or one fixed and one controlled point if you get what i mean)

by **DarthPhysics** » Sun Jan 20, 2008 6:37 pm

Hi ,
I know how IK work , and is because i know how it work and what are the limitation that i have trashed 3 month of intense programming.
A pure physics engine work alot better , but ....
At the moment it seem be not possible to have 1 fixed point + 1 controlled point in NGD.
Or Julio have found other solution :wink:

For the Ragdoll story , i dont have use any in NGD , what i have say was from observation from some 3D software ...
After looking little more i see they use Jacobian and CCD IK algorithm for
control the diverse part from a ragdoll body.
Now if you try a very complicated assembly with those algorithm's , you will just stuck in no solution or in a full aberration position.
This point make me trash every IK thing i have see before.
In a Physics engine it give everytime a solution

It's just more or less tricky when they is a assembly with more than 6DOF.(that is then just some logic stuff to add)
But it's not a problem for me to resolve those problem if i can control the end of the chain like any other object.

Cya.

by **DarthPhysics** » Wed Jan 23, 2008 9:49 am

Heya all

Today , i've make some test to imitate a magnet.
on this video :
http://fr.youtube.com/watch?v=vbfrDRMUulg
you will see Spheres arround the robot end effector.
They are other sphere that work like magnet , the spheres arround the
robot TCP have to 'glue' to the magnet(direction) spheres.
I use force on each spheres to move the Spheres arround the TCP but it's
diffinitely not the good solution :shock:

The Robot orient well now , but i cant stop the 'glitch' orienting or it dont orient !

@Julio : i really need a control point

Cya

by **Julio Jerez** » Wed Jan 23, 2008 11:43 am

give a few days, I will help you out I promise.

but I am in a tight spot now.

Testing Newton for Industrial Robtics

Testing Newton for Industrial Robtics

Who is online