Understand what went wrong in a recorded workcell run—and turn it into a repeatable test.
Give Metriplane timestamped object positions and process rules. It creates an incident timeline, a human-readable report, a verified evidence bundle, and a regression check you can run again.
In the bundled example, a required torque driver is missing and an assembly step is delayed by 35.0 seconds. The generated check lets an engineer rerun that same case after the software or process rules change. Metriplane analyzes recordings; it does not control machinery or make safety or quality decisions.
In this bounded, observe-only demo, a finished pallet is ready, but the recorded AGV state does not enter the configured pickup zone within 60 seconds. See how Metriplane turns the missed handoff into an inspectable incident, an integrity-verified evidence bundle, and a repeatable regression check.
Watch the full 52-second 4K film on YouTube →
python -m pip install "metriplane==0.4.0.post2"
metriplane demo --openThe example is camera-free, GPU-free, Docker-free, ROS-free, and offline after
installation. It writes an HTML report and requests that the browser open it.
Headless users can omit --open.
This is a real replay of the package's inspectable recorded JSONL state and
process rules: Metriplane runs the normal incident engine, writes a fresh
report and evidence bundle, verifies that bundle, and reruns the generated
regression check. Copy the exact starter inputs with
metriplane demo --export-inputs example-inputs.
Metriplane bundled demo
Scenario:
A required torque driver is missing during an assembly step.
The fastening step is delayed by 35.0 seconds.
Result:
PASS Incident timeline: 6 events
PASS Incident report: 1 incident
PASS Evidence bundle: verified
PASS Repeatable regression check: passed
Browser: open request sent
If no browser opens, use the Report path above.
Demo complete.
Input
Timestamped object positions + process rules
Metriplane
Replays the recorded run and checks what happened
Output
Incident timeline
Incident report
Verified evidence bundle
Repeatable regression check
Use Metriplane to:
- explain a recorded workcell delay, such as a required tool going missing;
- preserve an incident as a checksummed bundle that another engineer can verify;
- rerun that incident as a regression check after software or process-rule changes.
- Recorded run: timestamped observations saved from a workcell session.
- Event: one detected change or process condition in that run.
- Incident: related events grouped into one problem worth reviewing.
- Evidence bundle: the incident files plus checks that reveal missing or changed contents.
- Regression check: a repeatable test generated from an incident and rerun after a change.
- Process rules: the expected tools, locations, steps, and timing for the work.
- Deterministic replay: replay that gives the same software result from the same validated input; it does not prove that the original physical measurements were accurate.
- Current installable software release:
v0.4.0.post2— replacement publication identity for the reduced Truth Recovery core release; no DOI and no new research measurement boundary - Retired unpublished production candidate:
v0.4.0.post1— locked qualification and TestPyPI staging passed, but production stopped before lease creation or upload and no post1 bytes reached PyPI - Preserved failed publication tag:
v0.4.0— immutable history of the attempt that stopped before any registry package or GitHub Release was published - Prior usability and adoption software release:
v0.3.0(no DOI) - Frozen DOI-archived research artifact:
v0.2.0 - TIM evaluated software boundary:
v0.1.3
Metriplane source code is preserved in the Software Heritage Archive.
- Software Heritage:
swh:1:rev:dfe9401891596215081a4eff0b828504217a3590 - Archived Git revision:
dfe9401891596215081a4eff0b828504217a3590 - Archived research-artifact DOI (
v0.2.0):10.5281/zenodo.20736619 - Source repository: https://github.com/Miko997/metriplane
- Official site: https://www.metriplane.com/
- Research Software Directory: https://research-software-directory.org/software/metriplane
- SciCrunch registry: Metriplane (RRID:SCR_028813)
- Python package: https://pypi.org/project/metriplane/
- ROS 2 Jazzy package index: https://index.ros.org/p/metriplane_ros/#jazzy
- v0.4.0.post2 reduced Truth Recovery core release: https://github.com/Miko997/metriplane/releases/tag/v0.4.0.post2
- v0.3.0 usability and adoption release: https://github.com/Miko997/metriplane/releases/tag/v0.3.0
- Product roadmap: ROADMAP.md
- 3-minute v0.2.0 demo: https://www.youtube.com/watch?v=7U5nbBbGGbw
- v0.2.0 release: https://github.com/Miko997/metriplane/releases/tag/v0.2.0
- Zenodo DOI: https://doi.org/10.5281/zenodo.20736619
- SSRN manuscript preprint: https://doi.org/10.2139/ssrn.7166858
- External reproduction issue: #6
- Short feedback form: https://docs.google.com/forms/d/e/1FAIpQLSfnMZ4b3fSVVtwA89hZt3A09gf85eLfhW00FDD76TGRLNpirQ/viewform
Robotics teams often need more than raw logs after an incident. Metriplane explores a structured evidence layer where replayed workcell state becomes an inspectable incident bundle, a verification target, and a generated regression check.
replayed workcell state
→ physical event
→ incident
→ Cell Truth Report
→ evidence bundle
→ bundle verification
→ generated regression check
The author-run evidence package included in the archived release records:
- 580 tests passed
- deterministic replay pass=true
- 6 physical events
- 1 incident
- 35.0 second missing-tool delay
- bundle verify: pass=true
- generated regression check: PASS
The core SoftwareX reproduction is camera-free, uses the exact v0.2.0 tag,
and writes rerun outputs to temporary directories rather than the archived
evidence package.
git clone --branch v0.2.0 --depth 1 https://github.com/Miko997/metriplane.git
cd metriplane
python3 -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install -e .
python -m metriplane.cli doctor
RUNS=/tmp/metriplane-softwarex-runs \
./tools/mp.sh deterministic-replay datasets/demo/session_001.jsonl
metriplane atlas validate-pack configs/domain_packs/assembly_cell
metriplane atlas run \
--session-jsonl datasets/demo/atlas/assembly_cell_missing_tool.jsonl \
--pack configs/domain_packs/assembly_cell \
--out /tmp/metriplane-softwarex-atlas \
--overwrite
metriplane atlas bundle verify \
/tmp/metriplane-softwarex-atlas/evidence_bundles/INC-0001.zip
metriplane atlas test \
/tmp/metriplane-softwarex-atlas/regression_tests/INC-0001.yaml \
--jsonThe full maintainer gate is separate from the core artifact path and adds test and browser dependencies:
python -m pip install -e .
python -m pip install pytest playwright
python -m playwright install chromium --with-deps
PYTEST_DISABLE_PLUGIN_AUTOLOAD=1 python -m pytest -q- observe-only
- local-first
- replay-first, camera-compatible
- bounded workcell scoped
- research-software oriented
- focused on evidence, not robot control
- no robot or machine control
- no safety certification
- no quality-release approval
- no people recognition
- no marker-free tracking claim
- no full 3D reconstruction claim
- no production-factory deployment validation
- no factory-wide deployment readiness
The metriplane_ros bridge is listed in the official ROS Index for Jazzy:
metriplane_ros. The bridge
connects to a running Metriplane WebSocket stream and republishes Metriplane
frames, alerts, and incidents as std_msgs/String JSON topics.
This source-level rosdistro indexing does not by itself imply that Metriplane is
available as a ROS binary package through apt.
Robotics teams often debug incidents using recorded sensor data, ROS bags, logs, traces, and simulation replays. Those artifacts are valuable raw evidence. Metriplane is not a replacement for ROS bags or logs. It explores the next structured layer: incident context, evidence bundle, verification result, and generated regression check.
External technical feedback and public discussion links are collected on the official site:
https://www.metriplane.com/feedback/
If you use or evaluate Metriplane v0.2.0, cite the archived release:
https://doi.org/10.5281/zenodo.20736619
For research-resource identification, use Metriplane (RRID:SCR_028813).
The SoftwareX manuscript evaluates the archived Metriplane v0.2.0 release. Reproduction commands, expected outputs, and evidence provenance are documented in docs/softwarex_reproducibility.md.
The public Python package and command-line entry points remain named metriplane.
| Area | Path | Purpose |
|---|---|---|
| Package | metriplane/ |
Python package and CLI implementation |
| Domain packs | configs/domain_packs/ |
Workcell-specific Atlas configuration |
| Demo datasets | datasets/demo/ |
Checked-in replay inputs for reproduction |
| Evidence | evidence/ |
Release evidence, manifests, and experiment artifacts |
| External-source proof | proofs/maniskill-pickcube-v1/ |
Bounded ManiSkill PickCube publication candidate and evaluator packet |
| Synthetic format proof | proofs/massrobotics-amr-offline-replay-v1/ |
Bounded MassRobotics-format offline replay with reference-only upstream identity |
| Tools | tools/ |
Supported local helper scripts |
| Docs | docs/ |
Technical documentation and runbooks |
| Web UI | web/ |
Local operator and review interfaces |
Use the Quick Reproduction Path above for the camera-free core artifact. The same commands, expected outputs, evidence provenance, and the separate maintainer-gate sequence are maintained in docs/softwarex_reproducibility.md and the review kit.
- First-time-user front door: docs/README.md
- Supported environments: docs/SUPPORTED_ENVIRONMENTS.md
- Development and contribution: docs/development.md and CONTRIBUTING.md
- Technical integration reference: docs/INTEGRATIONS.md
- Exact-version research reproduction: docs/softwarex_reproducibility.md
- ManiSkill PickCube external-fixture proof candidate: proof landing page
- MassRobotics AMR synthetic offline-replay profile: mapping note
MIT License. See LICENSE.