Evidence Infrastructure for Embodied AI

Find the events.
Engineers find the cause.

VeluneOS reconstructs evidence timelines from robotics logs and shows where a Target execution differs from a user-selected private Reference set.

Stop searching the entire log.

Large robotics logs force engineers to manually inspect topics, timestamps, gaps, jitter, commands, and motion. VeluneOS narrows the investigation surface and produces reproducible evidence reports for human review.

Local-first

Process MCAP and ROS2-derived evidence without requiring centralized raw-log upload.

Reference comparison

Compare one explicit Target against every user-selected private Reference.

Reproducible handoff

Preserve JSON as the source of truth with a bounded Markdown summary for engineering review.

Validated end to end.

The current MVP has been tested using real-derived robotics Core Report Bundles and a 10GB Target validation corpus.

10GB+

Large-corpus validation

2

Private Reference comparisons

316

Automated tests passing

0

Compatibility warnings in the real Service E2E

Evidence, not judgment.

VeluneOS reports observed differences and evidence windows. It does not automatically determine root cause, fault, liability, safety, severity, normality, or regression.

Validate it on your own logs.

Select one Target execution and a small private Reference set. VeluneOS will run a local compatibility preflight, generate pairwise evidence comparisons, and provide reproducible JSON and Markdown outputs.