LV SWARM Benchmarking Coordination Policies in a Lake Victoria Inspired Simulation

Coordination policies can respond differently to depth structure, drifting targets and heterogeneous item mass in aquatic collection tasks. LV-SWARM represents these processes in a configured Lake Victoria-inspired simulation. This study analyses 700 archived trials covering twelve coordination mechanisms and two reference policies, with fifty seed labels per policy, twenty robots, six hundred initial items and a 2,000-second horizon. Collection fraction and simulated collected mass are compared using within-seed contrasts, bootstrap intervals and multiplicity-adjusted paired tests. Across the six environment-derived mechanisms, mean collection fraction was 4.77 percentage points lower than across the six general-purpose mechanisms (95% pointwise interval −5.40 to −4.11). The exploratory TSS–PDC–HPCA subset instead exceeded the general-purpose set by 7.50 percentage points (6.88 to 8.11). TSS had the highest median collection fraction, 0.6317, whereas HPCA had the highest median collected mass, 266.65 kg. Nine mechanisms exceeded the RANDOM reference in adjusted paired comparisons. Several comparisons with the myopic full-field SMOB reference did not detect a difference, but no equivalence margin was specified. A separate 100-trial MOUNS input ablation estimated a collection change of +1.14 percentage points (−1.48 to +3.70), leaving effects in either direction compatible with the results. These findings compare fixed implementations with unequal information access within one scenario. The environmental quantities are model assumptions, and the results establish neither field remediation performance nor ecological benefit. The archived records, analysis, figures and identified source snapshot are provided for computational reuse.