ResearchBenchmarksRNN-based cooperative multi-agent verification on BoxPushing (BP) 20x20 environmentFollow1.51Avg Violation RateRNN-ProVe1.50121.56061.621.6794May 14, 2026Evaluation ResultsMethodMethodLinksAvg Violation RateVerification Success Rate (1 - δ)Error Tolerance (ϵ)Time (s)RNN-ProVeEnv size=20×20, GRU si...Env size=20×20, GRU size=32, Feas. Oracle=Approx., 1 - ê=99.23 | 99.99, # samples=100k, |h*|=324132026.051.51992.690.014RNN-ProVeEnv size=20×20, GRU si...Env size=20×20, GRU size=32, Feas. Oracle=Approx., 1 - ê=99.23 | 99.99, # samples=1M, |h*|=650272026.051.73992.120.0476