search

osrf / drcsim

Repository for the DRC project.
16 stars 6 forks source link

Wobbling in BDI manipulation mode #476

Open osrf-migration opened 9 years ago

osrf-migration commented 9 years ago

Original report (archived issue) by GoRobotGo (Bitbucket: GoRobotGo).

The original report had attachments: test_manip.cpp, sway.mp4

The robot seems to wobble back and forth in BDI manipulation mode. To reproduce: roslaunch drcsim_gazebo atlas.launch model_args:="_v4" hand_suffix:="_robotiq_hands" Change to manipulation mode. A simple program to do so it attached.

After one minute the robot will start wobbling - side to side.

osrf-migration commented 9 years ago

Original comment by Nate Koenig (Bitbucket: Nathan Koenig).

I'm testing using:

  1. roslaunch drcsim_gazebo atlas.launch model_args:="_v4" hand_suffix:="_robotiq_hands"
  2. rostopic pub /atlas/control_mode std_msgs/String -- "Manipulate"

So far, it has been 8 minutes of real-time (2 minutes sim time) and Atlas is still stable. Could you try the above two commands to see if they work?

osrf-migration commented 9 years ago

Original comment by GoRobotGo (Bitbucket: GoRobotGo).

I wasn't able to change the mode using any of:

#!c++

rostopic pub /atlas/control_mode std_msgstring -- "Manipulate"
rostopic pub /atlas/control_mode std_msgs/String "Manipulate"
rostopic pub /atlas/control_mode std_msgs/String -- "Manipulate"

It remained in stand as verified by /atlas/atlas_sim_interface_state

osrf-migration commented 9 years ago

Original comment by GoRobotGo (Bitbucket: GoRobotGo).

This issue is still present in pre-release 4.2.5. Video attached

osrf-migration commented 9 years ago

Original comment by GoRobotGo (Bitbucket: GoRobotGo).

osrf-migration commented 9 years ago

Original comment by David Conner (Bitbucket: dcconner).

We see the swaying as well, but the controller keeps Atlas on its feet. I had some success modifying the control gains to improve, but not eliminate the behavior.

I haven't tried pinning the hip yet; it might help to determine if it is the torso driving leg motions, or leg motions from new library driving the torso sway.

osrf-migration commented 9 years ago

Original comment by Nate Koenig (Bitbucket: Nathan Koenig).

Pull request #519 will allow my above test commands to work.

I also see the sway. Looking into it.

osrf-migration commented 9 years ago

Original comment by John Hsu (Bitbucket: hsu, GitHub: hsu).

osrf-migration commented 9 years ago

Original comment by John Hsu (Bitbucket: hsu, GitHub: hsu).

Tried changing physics variable and tested with:

Same wobbling behavior exists. My conclusion is this is controller induced, not physics solver related. @BillBlank cc'd for comment.

I tried a few combinations of different joint damping settings, but no magic bullet yet (by tweaking model).

For reference, my test command is:

rostopic pub -1 /atlas/atlas_sim_interface_command atlas_msgs/AtlasSimInterfaceCommand '{behavior: 6, manipulate_params: { use_demo_mode: False, use_desired: True, desired: { pelvis_height: 0.7, pelvis_yaw: 0.0, pelvis_lat: 0.0  } }, k_effort: [ 255, 255, 255, 255, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255, 255] }'
osrf-migration commented 9 years ago

Original comment by Stefan Kohlbrecher (Bitbucket: Stefan_Kohlbrecher).

This bug makes drcsim unusable for manipulation tests while using the BDI interface. Are there any plans for a fix?

osrf-migration commented 9 years ago

Original comment by John Hsu (Bitbucket: hsu, GitHub: hsu).

I've looked at it some more, but it seems the problem is not easily patch-able from the gazebo model side.

osrf-migration commented 9 years ago

Original comment by Stefan Kohlbrecher (Bitbucket: Stefan_Kohlbrecher).

Agreed, the question is if we can expect an update from whoever else is reponsible ;)

osrf-migration commented 9 years ago

Original comment by John Hsu (Bitbucket: hsu, GitHub: hsu).

It helps if more people chime in for support on this issue. Squeakiest wheel gets the grease :)

osrf-migration commented 9 years ago

Original comment by David Conner (Bitbucket: dcconner).

From what I gather this requires a BDI fix on their simulation library. At present, the simulation is useless for manipulation unless we pin the hip. This makes testing multiple behaviors difficult. @BillBlank is this on BDI's radar for their simulation team to fix?

osrf-migration commented 9 years ago

Original comment by Bill Blank (Bitbucket: BillBlank).

David, I'm sorry, but due to resource limitations we unfortunately are not planning to address this issue. Our focus is 100% on the physical robots and keeping teams up and running up through the finals in June.