suave_requirements

Runtime monitoring of the SUAVE RELAX requirements (requirement_monitor) and offline plotting of their fulfillment from recorded campaign bags.

Plotting requirement fulfillment

scripts/plot_requirement_fulfillment.py writes one comparison figure per requirement topic (thruster availability, search footprint, and the combined, thruster, water-visibility and battery reaction times) plus a coverage.json provenance report. Run it with the workspace sourced:

B=~/suave/results/batches/<batch>
python3 src/suave/suave_requirements/scripts/plot_requirement_fulfillment.py $B
python3 src/suave/suave_requirements/scripts/plot_requirement_fulfillment.py \
  $B/campaigns/<campaign>

The path may be a batch, its campaigns/ folder, its state.json, or one campaign folder (with runs.json). Each campaign gets its own folder, by default <batch>/campaign_results/<campaign>/requirement_fulfillment/ (a campaign outside a batch uses its parent folder instead of <batch>); --output-dir DIR writes to DIR/<campaign>/ instead.

Options: --managers (default: all six exp3 managers), --samples (grid points per run, default 1001, minimum 2), --formats (default png pdf). The rebetmc manager is labelled TacticalRetreat in figures.

Which bags are used

Bags are selected from each campaign folder with suave_tools.runs.accepted_runs, never by scanning directory names:

• A trial counts only if its entry in the campaign’s runs.json (schema_version: 2) has "status": "completed".
• A completed trial must have "bag": "accepted"; its bag is rosbags/<manager>_<mission>_<run_id>_<attempt_id>/. A completed trial without an accepted bag, or an unsupported runs.json, is an error, not a silent omission.
• Trials run with recording disabled ("bag": null) are reported as recording_disabled.
• Bags under rosbags/pending/ or rosbags/failed/ are never loaded.

An unreadable selected bag is an error. A bag that is readable but lacks a topic is allowed and reported in coverage.json.

Timing and statistics

• Time is normalized recording time: 0–100 % of each bag’s metadata start/duration, using the same origin for every topic. This is not mission time; mission boundaries are not inferred.
• Each run is resampled with previous-value hold: no interpolation and no lookahead. Before a topic’s first message the value is unknown (gap), never zero. At duplicate timestamps the last message wins.
• Non-finite values, and values outside [0, 1] by more than 1e-6, are counted as invalid samples and excluded.
• Curves are the mean over runs with a value at that grid point. Bands are approximate 95 % confidence intervals (1.96 * SD / sqrt(n), ddof=1) for n >= 2; with n = 1 only the mean is drawn. Legends show the number of contributing runs, and coverage.json holds per-grid-point counts.
• Managers with no observations for a topic are named in the figure caption and listed under no_observations in coverage.json; managers with no accepted runs at all are listed separately under no_accepted_runs. A topic with no data at all gets an explicit no-data figure.