Observed arrival · 2026-10-15
Nemesis finds the point of no return in autonomy failures
A failure-investigation tool for autonomy simulations that rewinds failed runs, tests alternative decisions, and checks whether the simulator contributed to the failure.
- For
- Autonomy-simulation teams investigating failed planner runs
- Worth noticing
- Its checkpoint search matched exhaustive search in 467 of 468 cases, using about three replays of compute.
Field notes
The published benchmark uses 468 planner failures across 20 scenes from the Waymo Open Motion Dataset, simulated in Waymax with a reactive test planner and injected faults. The site reports that checkpoint search matched exhaustive search in 467 of 468 cases at roughly three replays of compute, and that 183 of 722 failures disappeared when traffic was replayed from a recorded path. Exporting failures as regression tests and running against a user's own simulator are marked in development.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue