osrf / srcsim

Space Robotics Challenge
Other
9 stars 4 forks source link

Valkyrie can't walk after reharnessing following a fall #241

Open osrf-migration opened 7 years ago

osrf-migration commented 7 years ago

Original report (archived issue) by Steven Gray (Bitbucket: stgray).


After falling, when Valkyrie is reharnessed, the /world to /pelvis transform changes -- namely, upon release, her pelvis pose is between half a meter and a few meters lower. Attempting to command any footstep results in Valkyrie falling.

It looks like a whole body trajectory is given to the IHMC controller using Valkyrie's state prior to the fall (chest orientation, pelvis height, arm joint angles -- everything but the pelvis orientation and pelvis x,y position). Noticed this last part after Valkyrie fell and I skipped to a new checkpoint.

Here's a video trying to move the left foot forward only 5cm. I use the current foot in world position from TF and displace it 5cm. Valkyrie lifts her foot and essentially sits down.

The image shown is Valkyrie's position before falling in the task 1 start box (red) and her position after reharnessing at task 1 checkpoint 2. This is one of the more extreme vertical shifts I've seen. It also shows her torso tracking it's orientation from before the fall. 3266145333-val_before_after.png

(Splitting this out from #207 as it is separate from Valkyrie being in a weird configuration as it reharnesses and also happens after a successful reharness and detach)

osrf-migration commented 7 years ago

Original comment by Steven Gray (Bitbucket: stgray).


To quote @knitfoo from #207 -- this seems to be the same issue I'm seeing. Reharnessing succeeds and detaches, but Valkyrie cannot walk again.

Essentially, if you fall down, the robot is unusable from that point forward; 
you have to restart the sim to regain use of the legs. You can reharness 
two, three, a zillion times, and the first step you attempt to take results in 
disaster. It's not clear exactly why; I have some sense that the robot may 
have lost it's sense of foot position (i.e. we command a baby step, and 
Val tries to take a huge step).
osrf-migration commented 7 years ago

Original comment by Steven Gray (Bitbucket: stgray).


osrf-migration commented 7 years ago

Original comment by Rud Merriam (Bitbucket: rmerriam).


Not sure if this and #207 are same or different. Basically the re-harness is messy. I saw the pelvis issue also and mentioned it in $207. Got Val to work but took manually jiggling a bit. Even then she duck-walked to the finish box.

osrf-migration commented 7 years ago

Original comment by Steven Gray (Bitbucket: stgray).


@rmerriam The duck-walking you describe would be the same issue, where after reharnessing, Val's pelvis height in the world frame is different than it used to be. I made this issue for those times when reharnessing succeeds and Valkyrie starts in a configuration such that she should be able to walk. I've seen more extreme pelvis offsets, where Valkyrie is unable to take any steps at all.

osrf-migration commented 7 years ago

Original comment by Steve Peters (Bitbucket: Steven Peters, GitHub: scpeters).


I've created pull request #99, which would reset the arm and leg joints after harnessing. I think it might help with this issue.

osrf-migration commented 7 years ago

Original comment by Steven Gray (Bitbucket: stgray).


Thanks @scpeters. That'll help the tangling when lowering. Not sure about the torso moving to match its old orientation in the world prior to a fall even after it gets teleported to a new location. I'm also still seeing this issue even when Val is reharnessed successfully.

osrf-migration commented 7 years ago

Original comment by dan (Bitbucket: dan77062).


We're also seeing this behavior and noticed that we never see the harness detach. It goes to lowering... switching to high level control and then just stays there. So perhaps it constantly thinks it's in the harness and that's why it doesn't know its pelvis height? Or perhaps this is a different issue with similar symptoms.

osrf-migration commented 7 years ago

Original comment by Vinayak Jagtap (Bitbucket: vinayak_jagtap).


We had the same issue too. As per my observation, the z of world and pelvis is almost same when this occurs.

osrf-migration commented 7 years ago

Original comment by Rud Merriam (Bitbucket: rmerriam).


I'm also seeing were messages are being ignored after a harness release. I have a set pattern of actions to reestablish the robot's stature. For example, one is spreading the feet apart. If I put two 'spreading' calls in a row the second one takes affect. It's happening nearly all the time with that action. Others are adjusting the pelvis height and the chest position.

Also been doing a lot more running with all three tasks and it seems much worse with that setup.

osrf-migration commented 7 years ago

Original comment by Brooks Paige (Bitbucket: brx).


I'm also getting what seems like it might be the same issue even after being released from the initial harness, since updating srcsim to 0.7.0. The legs and feet appear in a crooked sort of position, and the transform to the world coordinate frame is unavailable.

osrf-migration commented 7 years ago

Original comment by Louise Poubel (Bitbucket: chapulina, GitHub: chapulina).


@brx, which version of SRCSim are you upgrading from? Did you run sudo apt-get upgrade?