Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

dartsim: support non-tree kinematics in AttachFixedJoint #352

Merged
merged 8 commits into from
Jun 29, 2022

Conversation

scpeters
Copy link
Member

@scpeters scpeters commented Jun 5, 2022

🦟 Bug fix

Alternative to #345 and #347

Summary

With inspiration from #345 and #347, this implements a variation on the approach from gazebo-classic for handling non-tree kinematics (including kinematic loops) with the reduced coordinate Featherstone formulation by splitting links and welding them back together with a constraint. This adds data to the LinkInfo struct and a SplitAndWeldLink helper function to Base.hh. The test made by @arjo129 is included as well.

This only implements support for kinematic loops in the AttachFixedJoint method, but it should be applicable in other Attach*Joint methods and the ConstructSdfJoint method to support non-tree kinematics in more cases.

This implementation differs slightly from the approach in gazebo-classic. In gazebo11, all joints that are created dynamically using gazebo::physics::Model::CreateJoint use the approach of splitting and welding a link, even if the joint could simply be added as part of the existing Skeleton (see the call to DARTModelPrivate::CreateLoopJointAndNodePair in DARTModel::CreateJointHelper in DARTModel.cc). Within DARTModlPrivate::CreateLoopJointAndNodePair, a new joint and node pair are created for the desired joint type by calling DARTModelPrivate::CreateJointAndNodePair and then welded to the desired child link with blended inertia by calling DARTLink::AddSlaveBodyNode.

In this implementation, the split-and-weld approach is only taken if the target child link already has a parent joint that is not FreeJoint. To minimize changes to the existing code, when an existing parent joint is detected, the split-and-weld is done first, by creating a new FreeJoint and node pair and passing it to the SplitAndWeldLink helper function, which blends the inertia between the nodes and creates a WeldJointConstraint. The newly welded body node is then moved to a WeldJoint using the moveTo API. This move requires the transform for the WeldJointConstraint to be updated to ensure consistency.

UPDATE: with thanks to @azeey for the changes in #367, the welded links are now properly removed when joints are detached.

Checklist

  • Signed all commits for DCO
  • Added tests
  • Updated documentation (as needed)
  • Updated migration guide (as needed)
  • Consider updating Python bindings (if the library has them)
  • codecheck passed (See contributing)
  • All tests passed (See test coverage)
  • While waiting for a review on your PR, please help review another open pull request to support the maintainers

Note to maintainers: Remember to use Squash-Merge and edit the commit message to match the pull request summary while retaining Signed-off-by messages.

In gazebo classic, kinematic loops are closed by
identifying a link with multiple parents and
creating a mirror body node that is welded to
the original. The inertia will be split evenly
between the body nodes. This change adds data
to store the welded mirror nodes, the weld
constraints, and the original inertial data.

Signed-off-by: Steve Peters <scpeters@openrobotics.org>
To support closing kinematic loops.

Signed-off-by: Steve Peters <scpeters@openrobotics.org>
If the child link already has a parent, split the link
and weld the new node to the original node.

Signed-off-by: Steve Peters <scpeters@openrobotics.org>
@codecov
Copy link

codecov bot commented Jun 5, 2022

Codecov Report

❗ No coverage uploaded for pull request base (ign-physics5@e91c58a). Click here to learn what that means.
The diff coverage is n/a.

@@               Coverage Diff               @@
##             ign-physics5     #352   +/-   ##
===============================================
  Coverage                ?   76.35%           
===============================================
  Files                   ?      129           
  Lines                   ?     5810           
  Branches                ?        0           
===============================================
  Hits                    ?     4436           
  Misses                  ?     1374           
  Partials                ?        0           

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update e91c58a...80bafb3. Read the comment docs.

Signed-off-by: Arjo Chakravarty <arjo@openrobotics.org>
arjo129
arjo129 previously requested changes Jun 6, 2022
Copy link
Contributor

@arjo129 arjo129 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I've added the unit tests in but they do not pass. Neither does the down stream pr in osrf/mbzirc#108 work. I'm not too sure whats wrong with it. Either ways I'm not on mbzirc right now. I believe @aaronchongth is taking over this aspect of it.

@aaronchongth
Copy link

I might be wrong, but could the unit test that was ported over not be related to kinetic loops since both the parents of the added fixed joints are free floating and not attached to each other?

Signed-off-by: Steve Peters <scpeters@openrobotics.org>
@scpeters
Copy link
Member Author

scpeters commented Jun 9, 2022

I might be wrong, but could the unit test that was ported over not be related to kinetic loops since both the parents of the added fixed joints are free floating and not attached to each other?

I think the test is ok. It may not precisely be a "loop", but it doesn't follow the tree structure because the middle box will have two parents after attaching the fixed joints.

I think I have a fix in 21a608d

@scpeters scpeters marked this pull request as ready for review June 9, 2022 22:46
@scpeters
Copy link
Member Author

scpeters commented Jun 9, 2022

I've updated the description and marked this ready for review

@scpeters scpeters changed the title dartsim: support kinematic loops in AttachFixedJoint dartsim: support non-tree kinematics in AttachFixedJoint Jun 9, 2022
@scpeters scpeters dismissed arjo129’s stale review June 9, 2022 23:52

The test has been fixed

@aaronchongth
Copy link

Thanks for explaining @scpeters !

Attaching works like a charm! I am having trouble detaching the models though, it looks like one of the models will refuse to detach.

detach_and_stuck-2022-06-10_11.55.40.mp4

It looks like the first parent that attaches, is unable to detach afterwards, but the second parent is free to attach and detach. After detaching, both the still-attached parent and the model will be stuck in space somehow, until the other previously detached parent bumps into it.

I tried dropping shapes onto the stuck platform, but it doesn't budge.

This exposes a current test failures

Signed-off-by: Steve Peters <scpeters@openrobotics.org>
Signed-off-by: Steve Peters <scpeters@openrobotics.org>
@aaronchongth
Copy link

After testing it more downstream, I faced some interesting behaviors when detaching the models one after another. Consistently, detaching one of the joints does not seem to do anything, while detaching the other one causes both to be detached.

detach_1.mp4
detach_2.mp4

I duplicated the test in this PR to detach one at a time, but could not seem to replicate this behvaior. I was wondering if you knew what might be the cause?

@iche033
Copy link
Contributor

iche033 commented Jun 13, 2022

After testing it more downstream, I faced some interesting behaviors when detaching the models one after another. Consistently, detaching one of the joints does not seem to do anything, while detaching the other one causes both to be detached.

I looked into this a bit and turns out that it's a downstream issue. We are generating the same detach topic name for both UAVs (one overriding another). So something to be fixed on our end.

@scpeters
Copy link
Member Author

scpeters commented Jun 13, 2022

After testing it more downstream, I faced some interesting behaviors when detaching the models one after another. Consistently, detaching one of the joints does not seem to do anything, while detaching the other one causes both to be detached.

I looked into this a bit and turns out that it's a downstream issue. We are generating the same detach topic name for both UAVs (one overriding another). So something to be fixed on our end.

ok, I've also done some testing and if I only detach one joint, I would expect both boxes to fall, but only one is doing so. There may still be issues in the Detaching functionality here

@iche033
Copy link
Contributor

iche033 commented Jun 16, 2022

ok, I've also done some testing and if I only detach one joint, I would expect both boxes to fall, but only one is doing so. There may still be issues in the Detaching functionality here

@aaronchongth was also able to reproduce the issue. Once the joint between a drone and an object is detached, we noticed that the drone is able to continue flying but the object remains static. Let us know if there's anything we can help here.

@scpeters
Copy link
Member Author

ok, I've also done some testing and if I only detach one joint, I would expect both boxes to fall, but only one is doing so. There may still be issues in the Detaching functionality here

@aaronchongth was also able to reproduce the issue. Once the joint between a drone and an object is detached, we noticed that the drone is able to continue flying but the object remains static. Let us know if there's anything we can help here.

I've tried debugging it, but I may need some advice from @mxgrey and @azeey about how to figure out what is going on

@arjo129
Copy link
Contributor

arjo129 commented Jun 17, 2022

Just a random thought: I don't see us removing the WeldConstraint in the detach function. Does this have something to do with it.

@mxgrey
Copy link
Contributor

mxgrey commented Jun 17, 2022

I haven't looked at the code yet, but based on the conversation I might suspect that if the WeldConstraint is being kept alive inside the World where one BodyNode was removed from the constraint while the other was kept inside the constraint, then DART will believe that the BodyNode that remains inside the constraint is meant to be rigidly attached to "the world" because a whenever a BodyNode* has a nullptr value it's treated as "the world".

I'll take a look to check my hypothesis later today.

@mxgrey
Copy link
Contributor

mxgrey commented Jun 17, 2022

Like @arjo129 suggested, I don't see ConstraintSolver::removeConstraint ever get called to remove the weld constraint that gets added here. I believe this is almost certainly the cause of the problem.

What I would suggest is to maintain a hash map of BodyNode* -> ConstraintBasePtr where each BodyNode* in the map is a shard that was created for the sake of this WeldConstraint technique. When DetachJoint is called, we should check that hash map to see if the child BodyNode of the joint is one of the keys. If it is, then we should delete the shard, redistribute the inertia across all the remaining shards for that link, and then finally remove the ConstraintBasePtr weld constraint from the world.

All that being said, it's worth noting that this link splitting technique is not actually needed for this PR. The ordinary way of using the weld constraint would completely suffice for AttachFixedJoint. The motive for splitting the link back in Gazebo classic was to allow other joint types (e.g. Revolute, Prismatic), to form non-tree structures. If a link ever gets assigned more than one parent where one of the parents is a non-fixed joint, then we split the link into BodyNode shards, glue the shards together with weld constraints, and give each shard a different parent joint. Doing that allows each parent joint to be any joint type. But if we are only implementing this for fixed joint then there's no benefit to the splitting.

I might suggest reworking this PR so that the link splitting logic gets applied to all the joint types instead of just the fixed joint. Otherwise it would be much cleaner to just remove the link splitting entirely.

* Remove welded dummy bodies when detaching a joint.

Also fixes a bug with integer divisions.

Signed-off-by: Addisu Z. Taddese <addisu@openrobotics.org>
@scpeters
Copy link
Member Author

I've merged #367 into this branch, so I think this is ready for re-review

@iche033
Copy link
Contributor

iche033 commented Jun 29, 2022

tested and confirmed that this PR works for us. I'll leave it to @azeey to give the final blessing.

const auto poseParent2 = dartBody3->getTransform();
const auto poseChild2 = dartBody2->getTransform();
auto poseParentChild2 = poseParent2.inverse() * poseChild2;
auto fixedJoint2 = model2Body->AttachFixedJoint(model3Body);
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It would be great to add test expectations on the resulting masses after a WeldJointConstraint is created and destroyed, but feel free to punt for now.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'll do this in a follow-up

@scpeters scpeters merged commit 8aee408 into ign-physics5 Jun 29, 2022
@scpeters scpeters deleted the scpeters/dart_kinematic_loops branch June 29, 2022 18:34
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
🏯 fortress Ignition Fortress
Projects
Archived in project
Development

Successfully merging this pull request may close these issues.

6 participants