How would you design a real time application for cnc control in c / c++ ???
04 Dec 2020 10:11 - 04 Dec 2020 12:34 #191018
by Grotius
Replied by Grotius on topic How would you design a real time application for cnc control in c / c++ ???
Hi,
A little update.
After i did making the iso. Finally the stepcolors are detailed. This is nice !
To save pc memory it now only load's the stepfile surfaces.
To get a more realistic machine movement,
I just experimented with updating the screen preview based on the "Position Feedback Value's"
from the Hal stepgenerator. My motors are connected and the output is nice.
Video that show's updating the machine movements on feedback position.
video
This is a example halfile used by above video.
I am thinking to integrate a stand alone motion planner now. This can be a hard job to do.
@Arciera,
This looks like what's also used by MoveIt! in the ROS framework. Not quite sure as it's been a while.
Yes.
Have you walked trough the source code already? Is it hard to understand for you? Or do you think something else?
Please let me know so i can improve it.
A video of the steppermotors running.
A little update.
After i did making the iso. Finally the stepcolors are detailed. This is nice !
To save pc memory it now only load's the stepfile surfaces.
To get a more realistic machine movement,
I just experimented with updating the screen preview based on the "Position Feedback Value's"
from the Hal stepgenerator. My motors are connected and the output is nice.
Video that show's updating the machine movements on feedback position.
video
This is a example halfile used by above video.
Warning: Spoiler!
I am thinking to integrate a stand alone motion planner now. This can be a hard job to do.
@Arciera,
This looks like what's also used by MoveIt! in the ROS framework. Not quite sure as it's been a while.
Yes.
Have you walked trough the source code already? Is it hard to understand for you? Or do you think something else?
Please let me know so i can improve it.
A video of the steppermotors running.
Attachments:
Please Log in or Create an account to join the conversation.
14 Dec 2020 23:57 - 15 Dec 2020 00:08 #191999
by Grotius
Replied by Grotius on topic How would you design a real time application for cnc control in c / c++ ???
Hi Arciera,
The motors are handled by trough hal.
I haven't done much, as I'm trying to reverse engineer my robots teach in pendant to see if I could some how use that together with linuxCNC.
Interesting !
I had some problems to solve. The Qt program was not able to rotate the motors trough my wishes. It was doing a quite good
job. But i was not satisfied ! The solution is to bring the Ik and Fk outside of the QT app as stand alone c++ component.
In the meantime i made a linuxcnc kinematics component for Ik and Fk based on KDL kinematics.
The problem in this case is that realtime code is in C and the KDL library is in C++. So downgrading KDL to c is an option.
But that would be silly to do.
So i made a tiny C++ program that act's as a realtime component. The output is very promising. When performing IK and FK the
cyclus time is < 100 us. In the QT program the position command update time was 10 times a second. So a very big improvement.
So if the Stepgen was updated 10 times a second you will hear the motors change speed 10 times a second. If the update is for example 1000 times a second, you will hear a perfect motor speed increase and decrease sound.
You could use this code too. The code is written as basic as possible to be more readable for users.
Later on adding the xml config file input etc. is easy to do. For now the kdl chain setup is written in the code.
Users can choose to do "Perform_IK" or "Perform_FK"
You see the chrono timer is doing a cycle of 61 nanoseconds. This is really fast.
I will add a sort of synchronisation pin to stay in line with the servo-threat and avoid making waste calculations.
SourceCode
The motors are handled by trough hal.
I haven't done much, as I'm trying to reverse engineer my robots teach in pendant to see if I could some how use that together with linuxCNC.
Interesting !
I had some problems to solve. The Qt program was not able to rotate the motors trough my wishes. It was doing a quite good
job. But i was not satisfied ! The solution is to bring the Ik and Fk outside of the QT app as stand alone c++ component.
In the meantime i made a linuxcnc kinematics component for Ik and Fk based on KDL kinematics.
The problem in this case is that realtime code is in C and the KDL library is in C++. So downgrading KDL to c is an option.
But that would be silly to do.
So i made a tiny C++ program that act's as a realtime component. The output is very promising. When performing IK and FK the
cyclus time is < 100 us. In the QT program the position command update time was 10 times a second. So a very big improvement.
So if the Stepgen was updated 10 times a second you will hear the motors change speed 10 times a second. If the update is for example 1000 times a second, you will hear a perfect motor speed increase and decrease sound.
You could use this code too. The code is written as basic as possible to be more readable for users.
Later on adding the xml config file input etc. is easy to do. For now the kdl chain setup is written in the code.
Users can choose to do "Perform_IK" or "Perform_FK"
You see the chrono timer is doing a cycle of 61 nanoseconds. This is really fast.
I will add a sort of synchronisation pin to stay in line with the servo-threat and avoid making waste calculations.
SourceCode
Attachments:
Please Log in or Create an account to join the conversation.
18 Dec 2020 18:57 - 18 Dec 2020 19:16 #192364
by Grotius
Replied by Grotius on topic How would you design a real time application for cnc control in c / c++ ???
@Grijalvap,
I Think what rodw is traying to indicate is that you can replace the void tpCalculateTrapezoidalAccel funtion with the one you write in order to test and implement.
Ahh. I didn't mention that.
If you ask me. I am not a fan of some parts of the lcnc source code trough the complexity and or missing comments.
That's why i write custom code that will work with hal as rt engine.
Attached code works, but try to integrate such a code.. I think integrating the code is a more complex task then actual writing the
S-curve code.
@Rodw,
Yes almost, I think at that point, the motion controller already knows the start and end points of the acceleration segment and that procedure is calculating ( planning) where it needs to be right now (eg on that servo thread cycle). So if you thought of your S curve as being an offset either above or below the trapezoidal path at any given time (eg below on the concave start and above at the convex end), you could let that procedure calculate its normal trapezoidal settings, then apply the S curve offset for position and acceleration at that point.
I'am not sure if that will work, have to think about that.
If you can return the toolpoint, the acc and vel it will work.
I suspect you could calculate the maximum acceleration required to arrive the S curve
The max acceleration for the S-curve = 2* normal lineair acceleration. That max acceleration point is exact in the middle of the acc / dcc path, and is called "the inflection point".
@Arciera,
dm17ry
?? Hahaa.. What do you mean with this?
Next step will be to make the model of my RV-6SDL robot and try to get the joint positions from linuxcnc into the skynet app.
Making the stepmodel is oke. Try to get it working in the skynet app. Take a Skynet release that fit's your needs.
For actual controlling the robot, maybe wait a few day's until i have a release that is capable to control your robot.
Oke i have written the S-curve code in C comp style, and did a graph plot with Qt to show what is happening.
- Cyan is concave period
- Blue is convex period
- Gray is at speed period
- Green is acceleration
- startpoint x0 y0 z0, endpoint x600 y200 z0
- MaxVel = 20 mm/sec
- Acceleration = 1 mm/sec^2
- Output graph :
From S-curve document : doc
Ok. This looks oke to me.
Now i can try to expand the s-curve code to work with motioncommands.
QT source code with graph view : SourceCode
I Think what rodw is traying to indicate is that you can replace the void tpCalculateTrapezoidalAccel funtion with the one you write in order to test and implement.
Ahh. I didn't mention that.
If you ask me. I am not a fan of some parts of the lcnc source code trough the complexity and or missing comments.
That's why i write custom code that will work with hal as rt engine.
Attached code works, but try to integrate such a code.. I think integrating the code is a more complex task then actual writing the
S-curve code.
@Rodw,
Yes almost, I think at that point, the motion controller already knows the start and end points of the acceleration segment and that procedure is calculating ( planning) where it needs to be right now (eg on that servo thread cycle). So if you thought of your S curve as being an offset either above or below the trapezoidal path at any given time (eg below on the concave start and above at the convex end), you could let that procedure calculate its normal trapezoidal settings, then apply the S curve offset for position and acceleration at that point.
I'am not sure if that will work, have to think about that.
If you can return the toolpoint, the acc and vel it will work.
I suspect you could calculate the maximum acceleration required to arrive the S curve
The max acceleration for the S-curve = 2* normal lineair acceleration. That max acceleration point is exact in the middle of the acc / dcc path, and is called "the inflection point".
@Arciera,
dm17ry
?? Hahaa.. What do you mean with this?
Next step will be to make the model of my RV-6SDL robot and try to get the joint positions from linuxcnc into the skynet app.
Making the stepmodel is oke. Try to get it working in the skynet app. Take a Skynet release that fit's your needs.
For actual controlling the robot, maybe wait a few day's until i have a release that is capable to control your robot.
Oke i have written the S-curve code in C comp style, and did a graph plot with Qt to show what is happening.
- Cyan is concave period
- Blue is convex period
- Gray is at speed period
- Green is acceleration
- startpoint x0 y0 z0, endpoint x600 y200 z0
- MaxVel = 20 mm/sec
- Acceleration = 1 mm/sec^2
- Output graph :
From S-curve document : doc
Ok. This looks oke to me.
Now i can try to expand the s-curve code to work with motioncommands.
QT source code with graph view : SourceCode
Attachments:
Please Log in or Create an account to join the conversation.
Time to create page: 0.198 seconds