0
$\begingroup$

Gazebo Answers logo

Extracted from question 2665

When the simulator does actually launch, about 60% of the time, as soon the counter starts running and the atlas robot is initialized into the world, the atlas robot will immediately fall to the ground in a totally random direction without having received any input from the user, no input from any controller through rosrun, no input on keyboard or mouse, the atlas robot just falls in a random direction and crunches up.


Originally posted by gerkey on Gazebo Answers with karma: 1414 on 2013-05-07

Post score: 1


Original comments

Comment by gerkey on 2013-05-07:
What version of drcsim are you running? Starting with drcsim 2.5.0, the Atlas robot starts in BDI Stand mode, which should keep it standing.

Comment by aqiruse on 2013-05-10:
I am running drcsim 2.5. I have tried on both an Intel Card and an NVIDIA Card and it will sometimes fall over immediately after spawning.

Comment by aqiruse on 2013-05-10:
I am running drcsim 2.5. I have tried on both an Intel Card and an NVIDIA Card and it will sometimes fall over immediately after spawning.

$\endgroup$

2 Answers 2

0
$\begingroup$

Gazebo Answers logo

Update

should be fixed in this pull request.


Hi, If
roslaunch atlas_utils qual_task_1.launch

produces a fallen robot on startup, can you please send us the state log for debugging?

/tmp/qual_task_1/state.log

thanks.


Originally posted by hsu with karma: 1873 on 2013-05-11

This answer was ACCEPTED on the original site

Post score: 0


Original comments

Comment by cga on 2013-05-11:
logs for 3 falls and 3 stands are at http://www.cs.cmu.edu/~cga/q1fall/ If you want more or with another version of drcsim running (or from another type of machine), just let us know.

$\endgroup$
0
$\begingroup$

Gazebo Answers logo

This is not an answer but more data. We have been reluctant to report this data since it is basically intermittent (voodoo), and only occurs on our machine that is similar to the planned VRC machines (dual Intel(R) Xeon(R) CPU E5-2687W 0 @ 3.10GHz) and not on i7-3770 or i7-3770K-based machines. This machine has been running Gazebo well for months. We sacrificed many chickens to the gods before we decided this effect is real.

This is the drcsim-nightly distribution installed this morning (May 11) around 10 EST.

Test 1: Qualifier task 1: Just type:

roslaunch atlas_utils qual_task_1.launch

Of 10 launches, in 6 the robot had already fallen when first rendered.

Test 2: roslaunch atlas_utils atlas.launch: 10 trials, no falls.

Test 3: roslaunch atlas_utils atlas_sandia_hands.launch: of 10: 4 falls, 2 blank screens, 4 worked.

Test 4: roslaunch atlas_utils vrc_task_2.launch: All 5 launches worked. (we loaded the missing cinderblock model).

Test 5: roslaunch atlas_utils vrc_task_1.launch: All 5 launches worked. (we updated the polaris_ranger_ev model)

Test 6: roslaunch atlas_utils vrc_task_3.launch: All 5 launches worked.

Test 7: roslaunch atlas_utils atlas_sandia_hands.launch: 6 falls out of 10.

A student was so annoyed at being forced to use the "buggy" machine he made a video documentary, attempting to capture on film the evil spirits involved: http://www.youtube.com/watch?v=S7ONCWsY23U (he reverted to drcsim-2.5/gazebo-1.7 for the video).

Our hypotheses (for what it is worth):

  1. The BDI wiggle and foot slipping moves at startup are marginal (but the robot seems to fall before that?)

  2. Something about threading timing means the controller kicks in a little later and the robot is already doomed.

  3. A week ago we experimented with the position and velocity filtering ROS service on this machine. Maybe after several reboots, reinstalls of Gazebo (various versions), there is something from that that persists.

The good news is this effect is not showing up on the VRC tasks.

Aqiruse, you are not crazy. You will be vindicated.

Many chickens died to bring you this information.


Originally posted by cga with karma: 223 on 2013-05-11

This answer was NOT ACCEPTED on the original site

Post score: 1

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.