Metriplane v0.3.0 turns robot incidents into repeatable regression tests
Open-source tool replays robot failures with no GPU, Docker, or ROS required.
Metriplane v0.3.0, developed by Miko997, is an open-source physical observability tool for robotics workcells. It transforms messy recordings—logs, screenshots, bug reports—into structured incident reports and regression tests that can be re-run after changes. The latest release includes a self-contained demo: `pip install metriplane==0.3.0` and `metriplane demo --open` recreates an assembly delay caused by a missing torque driver without any robot, camera, GPU, Docker, or ROS installation. This lowers the barrier for evaluating deterministic replay and evidence generation, making the concept easy to try in minutes.
Beyond the demo, the repository now provides reproducible paths for ManiSkill and robomimic recordings, reusing the same incident and regression engine, so users can evaluate incidents without installing those simulators. A new Source Adapter SDK and initial ROS 2, MCAP, and TF2 format profile have been added, with the ROS 2 bridge indexed for Jazzy and capable of publishing frames, alerts, and incidents into ROS 2. The current ROS 2 example is synthetic, and the author is seeking feedback from ROS 2, simulation, and manufacturing workcell communities—requesting users to run the demo and report OS, Python version, and success, while open to bug reports and criticism.
- Install and demo: `pip install metriplane==0.3.0` then `metriplane demo --open` reproduces a missing-torque-driver assembly delay with no robot or simulator.
- Expanded support: reproducible paths for ManiSkill and robomimic recordings, evaluated without installing those simulators.
- ROS 2 integration: Source Adapter SDK plus initial ROS 2, MCAP, and TF2 profile; bridge indexed for Jazzy publishes frames and incidents.
Why It Matters
For robotics teams, incident replay with regression tests can cut debugging time and prevent recurring failures in production workcells.