ResearchBenchmarksG3 efficient on walker2d medium-replay v2Follow581.5Mean Raw MuJoCo ReturnFrozen148.964261.257373.55485.843Apr 23, 2026Evaluation ResultsMethodMethodLinksMean Raw MuJoCo ReturnFrozen2026.04581.5Prior2026.04501.3PoEalpha=0.1alpha=0.12026.04466.5Additivealpha=0.5alpha=0.52026.04290.4AWR2026.04287.5PoEalpha=0.5alpha=0.52026.04280KL-Regalpha=0.5alpha=0.52026.04280CQL2026.04258.4IQL2026.04165.6