Jeff LiuExperiment Reports
Back to all reports
Reinforcement & robot learning

Failure cases: training metrics versus displacement

Original report

Open full-page reader

Historical report preserved from the July 2026 Robot Learning Dashboard archive. Text, diagrams and experimental data remain in their original language.

An analysis of G1 policies with favorable training metrics but inadequate commanded motion, using measured displacement to examine capability claims.

Experiment date
Original authors
The original archive credits Jeff Liu; the individual report has no separate byline.
Report language
Chinese

Main observations

The report contrasts sprint, stairs and rough policies with a working locomotion policy, explaining why still frames and tracking cameras can obscure displacement.

Scope and limitations

Conditions, data and conclusions are preserved from the July 2026 archive. These simulation experiments do not establish hardware validation.

Source

View the original source and its recorded version

Preserved as a historical record. The report describes the software, conditions and observations of its original experiment.