Error after submission and the container isn't getting cleanly terminated locally
A critical bug in the AI for Industry Challenge platform prevents Docker containers from exiting cleanly, halting submissions.
A significant technical bug is disrupting the "AI for Industry Challenge," a competition likely focused on deploying AI in industrial or robotic settings. A participant, Sreekar, detailed on April 20, 2026, that their model submission fails because the evaluation environment's Docker containers do not exit cleanly. The core of the problem is an assertion failure (`Assertion 'numThreadHandles < 128' failed`) in the Ogre3D graphics engine's threading code, which is being used within a ROS 2 (Robot Operating System 2) component container. This crash causes the container process to die with exit code -6, leaving the Docker Compose stack hanging and requiring manual termination with Ctrl+C.
Despite the evaluation logic completing successfully, this termination failure prevents the submission system from finalizing the run and providing a score. The participant confirmed that the proper shutdown parameter (`shutdown_on_aic_engine_exit:=true`) was set, indicating the bug is external to their code. The issue is compounded by a lack of accessible error logs from the submission system, despite recent platform changes intended to increase log visibility. This has created a blocker for multiple participants, as evidenced by related forum topics discussing similar submission and container errors, effectively stalling progress in the competition.
- A fatal assertion failure in the Ogre3D threading library (`numThreadHandles < 128`) crashes the ROS 2 component container.
- The bug prevents Docker containers from terminating cleanly post-evaluation, forcing manual intervention and causing submissions to fail silently.
- The issue blocks scoring and is widespread, with multiple related support threads from other participants experiencing similar container errors.
Why It Matters
This bug halts progress in a major AI competition, preventing the evaluation of industrial AI models and wasting developer time.