Ask Your Question

Revision history [back]

Extreme quantization of joint velocities on real PR2

Hi all,

I'm trying to run a self-written controller plugin for the PR2 that offers a velocity-resolved interface for the arms of the robot. In fact, it basically does what the joint_velocity_controller out of robot_mechanism_controllers does but for all 7 joints of one arm.

Anyway, my problem is that when I run the controller on a real PR2, send a step desired joint velocity, and plot the actual velocities with rxplot then the output graph shows a pretty extreme quantization of the output velocity. Interestingly, when I run the same controller on a simulated PR2 in gazebo the output graph looks smooth.

I attached links to several graphs to illustrate my problem. The plots show desired and actual velocities of the l_shoulder_pan_joint, x-axis unit is [s], and y-axis units is [rad/s]. As PID gains I used the ones provided in the pr2_controller_configuration package. Simulation and real robot both run fuerte.

Does anybody have an idea of why the actual velocity on the real robot looks so quantized?

I already played around with the gains to see if tuning helps -- but it doesn't. On a side-note, if I set the D-gain of the PID controller that I use to something (even pretty small) non-zero I get some nice-sounding vibrations because of these actual velocity readings. So, it would be desirable to fix this issue. ;)

I'm grateful for any hint/answer that points me the right way. Thanks already.

Best, Georg.

link to plots: toychest.ai.uni-bremen.de/wiki/projects:tuning_pr2_vel_resolved

click to hide/show revision 2
No.2 Revision

Extreme quantization of joint velocities on real PR2

Hi all,

I'm trying to run a self-written controller plugin for the PR2 that offers a velocity-resolved interface for the arms of the robot. In fact, it basically does what the joint_velocity_controller out of robot_mechanism_controllers does but for all 7 joints of one arm.

Anyway, my problem is that when I run the controller on a real PR2, send a step desired joint velocity, and plot the actual velocities with rxplot then the output graph shows a pretty extreme quantization of the output velocity. Interestingly, when I run the same controller on a simulated PR2 in gazebo the output graph looks smooth.

I attached links to several graphs to illustrate my problem. The plots show desired and actual velocities of the l_shoulder_pan_joint, x-axis unit is [s], and y-axis units is [rad/s]. As PID gains I used the ones provided in the pr2_controller_configuration package. Simulation and real robot both run fuerte.

Does anybody have an idea of why the actual velocity on the real robot looks so quantized?

I already played around with the gains to see if tuning helps -- but it doesn't. On a side-note, if I set the D-gain of the PID controller that I use to something (even pretty small) non-zero I get some nice-sounding vibrations because of these actual velocity readings. So, it would be desirable to fix this issue. ;)

I'm grateful for any hint/answer that points me the right way. Thanks already.

Best, Georg.

link to plots: toychest.ai.uni-bremen.de/wiki/projects:tuning_pr2_vel_resolved

Extreme quantization of joint velocities on real PR2

Hi all,

I'm trying to run a self-written controller plugin for the PR2 that offers a velocity-resolved interface for the arms of the robot. In fact, it basically does what the joint_velocity_controller out of robot_mechanism_controllers does but for all 7 joints of one arm.

Anyway, my problem is that when I run the controller on a real PR2, send a step desired joint velocity, and plot the actual velocities with rxplot then the output graph shows a pretty extreme quantization of the output velocity. Interestingly, when I run the same controller on a simulated PR2 in gazebo the output graph looks smooth.

I attached links to several graphs to illustrate my problem. The plots show desired and actual velocities of the l_shoulder_pan_joint, x-axis unit is [s], and y-axis units is [rad/s]. As PID gains I used the ones provided in the pr2_controller_configuration package. Simulation and real robot both run fuerte.

Does anybody have an idea of why the actual velocity on the real robot looks so quantized?

I already played around with the gains to see if tuning helps -- but it doesn't. On a side-note, if I set the D-gain of the PID controller that I use to something (even pretty small) non-zero I get some nice-sounding vibrations because of these actual velocity readings. So, it would be desirable to fix this issue. ;)

I'm grateful for any hint/answer that points me the right way. Thanks already.

Best, Georg.

link to plots: toychest.ai.uni-bremen.de/wiki/projects:tuning_pr2_vel_resolved

Extreme quantization of joint velocities on real PR2

Hi all,

I'm trying to run a self-written controller plugin for the PR2 that offers a velocity-resolved interface for the arms of the robot. In fact, it basically does what the joint_velocity_controller out of robot_mechanism_controllers does but for all 7 joints of one arm.

Anyway, my problem is that when I run the controller on a real PR2, send a step desired joint velocity, and plot the actual velocities with rxplot then the output graph shows a pretty extreme quantization of the output velocity. Interestingly, when I run the same controller on a simulated PR2 in gazebo the output graph looks smooth.

I attached links to several graphs to illustrate my problem. The plots show desired and actual velocities of the l_shoulder_pan_joint, x-axis unit is [s], and y-axis units is [rad/s]. As PID gains I used the ones provided in the pr2_controller_configuration package. Simulation and real robot both run fuerte.

Does anybody have an idea of why the actual velocity on the real robot looks so quantized?

I already played around with the gains to see if tuning helps -- but it doesn't. On a side-note, if I set the D-gain of the PID controller that I use to something (even pretty small) non-zero I get some nice-sounding vibrations because of these actual velocity readings. So, it would be desirable to fix this issue. ;)

I'm grateful for any hint/answer that points me the right way. Thanks already.

Best, Georg.

link to plots: toychest.ai.uni-bremen.de/wiki/projects:tuning_pr2_vel_resolved

Extreme quantization of joint velocities on real PR2

Hi all,

I'm trying to run a self-written controller plugin for the PR2 that offers a velocity-resolved interface for the arms of the robot. In fact, it basically does what the joint_velocity_controller out of robot_mechanism_controllers does but for all 7 joints of one arm.

Anyway, my problem is that when I run the controller on a real PR2, send a step desired joint velocity, and plot the actual velocities with rxplot then the output graph shows a pretty extreme quantization of the output velocity. Interestingly, when I run the same controller on a simulated PR2 in gazebo the output graph looks smooth.

I attached links to several graphs to illustrate my problem. The plots show desired and actual velocities of the l_shoulder_pan_joint, x-axis unit is [s], and y-axis units is [rad/s]. As PID gains I used the ones provided in the pr2_controller_configuration package. Simulation and real robot both run fuerte.

Does anybody have an idea of why the actual velocity on the real robot looks so quantized?

I already played around with the gains to see if tuning helps -- but it doesn't. On a side-note, if I set the D-gain of the PID controller that I use to something (even pretty small) non-zero I get some nice-sounding vibrations because of these actual velocity readings. So, it would be desirable to fix this issue. ;)

I'm grateful for any hint/answer that points me the right way. Thanks already.

Best, Georg.

link to plots: toychest.ai.uni-bremen.de/wiki/projects:tuning_pr2_vel_resolved

Extreme quantization of joint velocities on real PR2

Hi all,

I'm trying to run a self-written controller plugin for the PR2 that offers a velocity-resolved interface for the arms of the robot. In fact, it basically does what the joint_velocity_controller out of robot_mechanism_controllers does but for all 7 joints of one arm.

Anyway, my problem is that when I run the controller on a real PR2, send a step desired joint velocity, and plot the actual velocities with rxplot then the output graph shows a pretty extreme quantization of the output velocity. Interestingly, when I run the same controller on a simulated PR2 in gazebo the output graph looks smooth.

I attached links to several graphs to illustrate my problem. The plots show desired and actual velocities of the l_shoulder_pan_joint, x-axis unit is [s], and y-axis units is [rad/s]. As PID gains I used the ones provided in the pr2_controller_configuration package. Simulation and real robot both run fuerte.

Does anybody have an idea of why the actual velocity on the real robot looks so quantized?

I already played around with the gains to see if tuning helps -- but it doesn't. On a side-note, if I set the D-gain of the PID controller that I use to something (even pretty small) non-zero I get some nice-sounding vibrations because of these actual velocity readings. So, it would be desirable to fix this issue. ;)

I'm grateful for any hint/answer that points me the right way. Thanks already.

Best, Georg.

link to plots: toychest.ai.uni-bremen.de/wiki/projects:tuning_pr2_vel_resolved

Extreme quantization of joint velocities on real PR2

Hi all,

I'm trying to run a self-written controller plugin for the PR2 that offers a velocity-resolved interface for the arms of the robot. In fact, it basically does what the joint_velocity_controller out of robot_mechanism_controllers does but for all 7 joints of one arm.

Anyway, my problem is that when I run the controller on a real PR2, send a step desired joint velocity, and plot the actual velocities with rxplot then the output graph shows a pretty extreme quantization of the output velocity. Interestingly, when I run the same controller on a simulated PR2 in gazebo the output graph looks smooth.

I attached links to several graphs to illustrate my problem. The plots show desired and actual velocities of the l_shoulder_pan_joint, x-axis unit is [s], and y-axis units is [rad/s]. As PID gains I used the ones provided in the pr2_controller_configuration package. Simulation and real robot both run fuerte.

Does anybody have an idea of why the actual velocity on the real robot looks so quantized?

I already played around with the gains to see if tuning helps -- but it doesn't. On a side-note, if I set the D-gain of the PID controller that I use to something (even pretty small) non-zero I get some nice-sounding vibrations because of these actual velocity readings. So, it would be desirable to fix this issue. ;)

I'm grateful for any hint/answer that points me the right way. Thanks already.

Best, Georg.

link to plots: toychest.ai.uni-bremen.de/wiki/projects:tuning_pr2_vel_resolved

Extreme quantization of joint velocities on real PR2

Hi all,

I'm trying to run a self-written controller plugin for the PR2 that offers a velocity-resolved interface for the arms of the robot. In fact, it basically does what the joint_velocity_controller out of robot_mechanism_controllers does but for all 7 joints of one arm.

Anyway, my problem is that when I run the controller on a real PR2, send a step desired joint velocity, and plot the actual velocities with rxplot then the output graph shows a pretty extreme quantization of the output velocity. Interestingly, when I run the same controller on a simulated PR2 in gazebo the output graph looks smooth.

I attached links to several graphs to illustrate my problem. The plots show desired and actual velocities of the l_shoulder_pan_joint, x-axis unit is [s], and y-axis units is [rad/s]. As PID gains I used the ones provided in the pr2_controller_configuration package. Simulation and real robot both run fuerte.

Does anybody have an idea of why the actual velocity on the real robot looks so quantized?

I already played around with the gains to see if tuning helps -- but it doesn't. On a side-note, if I set the D-gain of the PID controller that I use to something (even pretty small) non-zero I get some nice-sounding vibrations because of these actual velocity readings. So, it would be desirable to fix this issue. ;)

I'm grateful for any hint/answer that points me the right way. Thanks already.

Best, Georg.

link to plots: toychest.ai.uni-bremen.de/wiki/projects:tuning_pr2_vel_resolved

Extreme quantization of joint velocities on real PR2

Hi all,

I'm trying to run a self-written controller plugin for the PR2 that offers a velocity-resolved interface for the arms of the robot. In fact, it basically does what the joint_velocity_controller out of robot_mechanism_controllers does but for all 7 joints of one arm.

Anyway, my problem is that when I run the controller on a real PR2, send a step desired joint velocity, and plot the actual velocities with rxplot then the output graph shows a pretty extreme quantization of the output velocity. Interestingly, when I run the same controller on a simulated PR2 in gazebo the output graph looks smooth.

I attached links to several graphs to illustrate my problem. The plots show desired and actual velocities of the l_shoulder_pan_joint, x-axis unit is [s], and y-axis units is [rad/s]. As PID gains I used the ones provided in the pr2_controller_configuration package. Simulation and real robot both run fuerte.

Does anybody have an idea of why the actual velocity on the real robot looks so quantized?

I already played around with the gains to see if tuning helps -- but it doesn't. On a side-note, if I set the D-gain of the PID controller that I use to something (even pretty small) non-zero I get some nice-sounding vibrations because of these actual velocity readings. So, it would be desirable to fix this issue. ;)

I'm grateful for any hint/answer that points me the right way. Thanks already.

Best, Georg.

link to plots: toychest.ai.uni-bremen.de/wiki/projects:tuning_pr2_vel_resolved

Extreme quantization of joint velocities on real PR2

Hi all,

I'm trying to run a self-written controller plugin for the PR2 that offers a velocity-resolved interface for the arms of the robot. In fact, it basically does what the joint_velocity_controller out of robot_mechanism_controllers does but for all 7 joints of one arm.

Anyway, my problem is that when I run the controller on a real PR2, send a step desired joint velocity, and plot the actual velocities with rxplot then the output graph shows a pretty extreme quantization of the output velocity. Interestingly, when I run the same controller on a simulated PR2 in gazebo the output graph looks smooth.

I attached links to several graphs to illustrate my problem. The plots show desired and actual velocities of the l_shoulder_pan_joint, x-axis unit is [s], and y-axis units is [rad/s]. As PID gains I used the ones provided in the pr2_controller_configuration package. Simulation and real robot both run fuerte.

Does anybody have an idea of why the actual velocity on the real robot looks so quantized?

I already played around with the gains to see if tuning helps -- but it doesn't. On a side-note, if I set the D-gain of the PID controller that I use to something (even pretty small) non-zero I get some nice-sounding vibrations because of these actual velocity readings. So, it would be desirable to fix this issue. ;)

I'm grateful for any hint/answer that points me the right way. Thanks already.

Best, Georg.

link to plots: toychest.ai.uni-bremen.de/wiki/projects:tuning_pr2_vel_resolved