Hi everyone,
I’m building an open-source tool called Metriplane.
Robot incidents often end up as loose logs, screenshots and bug reports that are difficult to reproduce. Metriplane turns recorded workcell state into a clear incident report, a verified evidence bundle and a regression test that can be run again after something changes.
You can try the latest release with:
python -m pip install "metriplane==0.3.0"
metriplane demo --open
The demo recreates an assembly delay caused by a missing torque driver. It needs no robot, camera, GPU, Docker or ROS installation.
Recent work has expanded Metriplane beyond its original demo. The repository now includes reproducible paths for ManiSkill and robomimic recordings. Both use the same incident and regression engine. After conversion, they can be evaluated without either simulator being installed.
I have also added a small Source Adapter SDK and an initial ROS 2, MCAP and TF2 format profile. The current ROS 2 recording example is synthetic. The next step is testing the approach with suitable public recordings from real robots or simulators.
The ROS 2 bridge is also indexed for Jazzy and can publish Metriplane frames, alerts and incidents into ROS 2.
I would love feedback from people working with ROS 2, simulation, robot testing or manufacturing workcells.
If you have two minutes, run the demo and reply with your operating system, Python version and whether it passed. Bug reports and honest criticism are welcome.
Reproduction report: Sign in to GitHub · GitHub