Note to Participants: Gazebo bug fix & changes to friction properties of the cable

Competitions AI for Industry Challenge
, ,
1

Hi teams!

We hope you’re enjoying working on the challenge so far.

I’m Kaushik :waving_hand: - I’m a Robotics Software Engineer supporting the AI for Industry Challenge hosted by Intrinsic. I’ve enjoyed interacting with a few of you during the course of the challenge, and I’m really excited to see the solutions everyone is building.

Recently, we received a few reports [#444, #450] noting unexpected phantom collisions during insertion after a few trials when invoked via our aic_engine. However, participants did not see this issue when running individual trials alone.

Our team investigated this and initially found that the cable’s friction and inertia, as previously defined, resulted in unexpectedly high forces when it makes contact with the table and task board in certain configurations. We have since fixed this by tuning the friction and inertia values.

More importantly, we found and fixed another bug where collision meshes from nested models do not get cleared when these models are removed. This fix would not have been possible without the collaboration of the participants in this challenge! We’re also glad to share that this patch will be included upstream (see this PR) in the next release of Gazebo.

:hammer_and_wrench: Required Actions

Please note these recommended actions as a result of changes relating to this issue:

  1. Pull the latest Docker image: Our eval image has switched to using these changes. Please pull the latest aic_eval image which uses the updated gz-physics package which contains the fix.

  2. Update local builds: If you are building locally, please run the vcs import step and rebuild your workspace.

  3. Re-evaluate your model: Due to the changes in the friction and inertia properties of the cable, we strongly recommend re-evaluating your model’s performance with these changes in effect. Please consider augmenting your dataset as necessary if you see a divergence in performance. Once you are ready, we encourage all teams to make a new submission to update the leaderboard with your latest scores based on these changes!

Feel free to let us know if this patch works and if you face any further issues.

Happy learning!

2 Likes