ResearchBenchmarksSafe Reinforcement Learning on SMARTS Intersection 512K steps (test)Follow12Collision CountCOX-Q5.1651.3397.5143.67Mar 25, 2026Evaluation ResultsMethodMethodLinksCollision CountOff-road CountRule Violation CountTimeout CountCOX-Qsteps=512K, stochastic...steps=512K, stochastic runs=20002026.0312200TQC-ORACsteps=512K, stochastic...steps=512K, stochastic runs=20002026.03181012CALsteps=512K, stochastic...steps=512K, stochastic runs=20002026.0323101SACLagsteps=512K, stochastic...steps=512K, stochastic runs=20002026.03332180CPPOPIDsteps=512K, stochastic...steps=512K, stochastic runs=20002026.031832290