ResearchBenchmarksRNN-based cooperative multi-agent verification on BoxPushing (BP) 10x10 environmentFollow1.15Average Violation RateRNN-ProVe1.14881.15691.1651.1731May 14, 2026Evaluation ResultsMethodMethodLinksAverage Violation RateSuccess Rate (1 - δ)Epsilon Error Rate (ϵ)Execution Time (s)RNN-ProVeEnv size=10×10, GRU si...Env size=10×10, GRU size=16, Feas. Oracle=Approx., 1 - ê=99.98 | 99.95, # samples=100k, |h*|=203202026.051.15992.450.0072RNN-ProVeEnv size=10×10, GRU si...Env size=10×10, GRU size=16, Feas. Oracle=Approx., 1 - ê=99.98 | 99.95, # samples=1M, |h*|=410612026.051.18991.760.0179