AgenticJevNav · Simulation Experiments

Manuscript Tables I-II and Figure 3 · Supplementary archives · HSSD simulation

Loading results

Scoring and comparability

SR: for present targets, the final position is within 1.6 m horizontally of an acceptable target AABB, with heading error at most 60 degrees. A2 T6 is scored using ground-truth absence and a not-found report within budget. NE covers T1–T5; SPL covers static T1–T4; Time covers successful episodes only: the first qualifying pose (after revision for T5) or report timestamp for T6. Path and request costs cover complete rollouts. Wrong goal is an independent diagnostic, not the complement of SR. Waiting and requests follow the recorded execution.

Historical budgets, seeds, adapters, and execution dates differ across the six-task comparison, so it is descriptive only. The current manuscript uses Table I for the main comparison, Table II for ablations, and Figure 3 for the 16 current A2 failures. Matched Sol controls and native benchmark summaries remain supplementary. Archived paired statistics use the October 27 historical reference and must not be mixed with the current A2 cohort. Navmesh blocking is a simulation proxy, not a physical collision measurement. This page contains recorded simulation experiments; physical robot results are available on the Physical Robot Experiments page.

Episode replays

Open 3D replay

✓ Success× Failure

Cohort and excluded attempts

3D scene · Recorded trajectory and timeline
Robot RGB · Captured frames · 6×
Recorded robot RGB GIF