dronova.

Planar survey simulation · USGS-derived scalar fields

Adaptive survey

Development selection and held-out terrain

Baseline retained. The selected policy failed all six lawnmower RMSE promotion bounds.

Download full report Study methods Result data
Simulation executions
516
36 development + 480 held-out
RMSE promotion gates passed
0 / 6
Selected policy versus lawnmower
Unsafe executions
33
Study-wide, across all evaluated policies
Terrain split
3 / 3 / 3
Train / validation / test locations

Recorded evidence

Results

24 cases

36 development executions and 480 held-out executions on nine terrain fields.

Case 4 of 24utm13_e350000_n3800000 · nominal_full · lawnmower

16 episodes per arm. Safety counts are totals; RMSE and coverage are means.
MetricSelected guarded_earlylawnmower
Normalized map RMSE0.1135330.116816
Synthetic DEM-sensing RMSE m0.4646080.478044
Observed area fraction0.9915770.989685
Squared-speed energy proxy20.464718.8493
Uncertified steps (total)00
Unsafe episodes00
Paired RMSE statistics at this location and horizon.
StatisticValue
Selected minus comparator RMSE-0.00328332
95% interval lower-0.0220418
95% interval upper0.0154751
Paired seeds16
Missing pairs0
Promotion one-sided upper bound0.0204237
Promotion one-sided alpha0.00833333
Safety gateTrue
RMSE gateFalse
Case notes & acceptance

Promotion comparison.

null = missing or not applicable, never zero. Intervals and sample sizes are reported in each table. Scroll tables horizontally to see every column.

USGS 3DEP / Field navigator

Terrain atlas

9 fields · 32 × 32 samples

Actual elevation rasters used by the study. North-up, with colour scaled within each patch.

Field 01 / 9UTM zone 13N · EPSG:26913

Elevation field utm13_e350000_n3800000North-up USGS raster. 32 by 32 samples across 100 metres between edge sample centers. Relative elevation spans 4.09 metres above a patch minimum of 1,675.37 metres. Contours linearly interpolate the supplied samples.005050100100East from southwest sample (m)North (m)N
0 mRelative elevation4.09 m
Patch minimum
1,675.37 m
Relief
4.09 m
Mean above minimum
2.04 m
Metres from USGS source values; vertical datum unresolved.Download field ↓
Raster processing & units

Elevation = offset + normalized value × scale. Colours show the actual raster; contours use linear interpolation on a fixed triangulation at 12.5% intervals of the normalized range. Sample centers span 100 m, with spacing 100/31 m. This is output sampling, not native resolution or accuracy. Per-patch colour scaling makes shapes comparable, not physical relief.

The navigator shows elevation, not rendered flight terrain or obstacles. Frozen train / validation / test assignments are in the original terrain figure below and the acquisition manifest.

Dataset methods →

Report figures

Original published assets
Nine USGS terrain fields and their frozen data splits.
01Nine USGS terrain fields and their frozen data splits.
Paired held-out map RMSE differences.
02Paired held-out map RMSE differences.
Methods & interpretation

The development-only choice guarded_early fails all six lawnmower RMSE promotion bounds; limited-time mean RMSE is worse at all three test locations. Nine fresh USGS patches are split 3 train / 3 validation / 3 test, not a nationwide geography sample. Infrastructure hardening did not rerun episodes. Paired Student-t intervals are descriptive within each location/horizon; only lawnmower uses the six-cell Bonferroni promotion rule. Adaptive Lloyd bandit uses oracle evaluation reward. Per-patch normalization is not terrain sampling accuracy; DEMs are not obstacle or ecological maps.

Methods overviewFull figure archive →Download report ↓