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

The team behind Intrinsic's AI for Industry Challenge has resolved a significant physics simulation bug in the Gazebo robotics simulator, thanks to collaborative efforts with competition participants. Robotics Software Engineer Kaushik announced that reports from teams (issues #444 and #450) identified unexpected phantom collisions during cable insertion tasks when run through the challenge's `aic_engine`. The root cause was traced to incorrectly tuned friction and inertia properties for the cable model, which generated abnormally high forces upon contact with the table and task board in specific configurations.

More importantly, the investigation uncovered a deeper bug where collision meshes from nested models were not being cleared when those models were removed from the simulation, leading to persistent ghost collisions. The fix involved tuning the cable's physical parameters and patching the mesh-clearing logic. This correction is now live in the challenge's evaluation Docker image (`aic_eval`), which uses an updated `gz-physics` package. Participants are required to pull the latest image, rebuild their local workspaces, and crucially, re-evaluate their AI models' performance, as the changed physics may affect task success rates.

The collaboration proved so valuable that the patch has been submitted upstream and is slated for inclusion in the next official release of Gazebo. This demonstrates how open challenges can drive improvements in foundational open-source tools used by the broader robotics and AI community. Teams are now encouraged to update their submissions to the leaderboard, reflecting their solutions' performance under the new, more accurate simulation environment.