List your company
ResearchFormerly WizWand

Multi-task Reinforcement Learning on Meta-World MT10 V2

99.5Success RateARS-LoRA
68.19676.32384.4592.577May 12, 2026
Updated 5mo ago

Evaluation results

MethodLinks
ARS-LoRAParams=3878K2026.0599.5
ARS-LN (400)Params=2530K2026.0598.6
ARSParams=2517K2026.0597.3
TOPPOParams=685K2026.0597.2
+LN-c +PopArtParams=685K2026.0596.8
+LN-cParams=685K2026.0596.2
+LN-c +PopArt +FG-cParams=685K2026.0594.8
SMTParams=–2026.0586.8
MOOREParams=6890K2026.0586
Vanilla MT-PPOParams=684K2026.0585.2
PaCoParams=–2026.0582.4
MT-MH-SACParams=2546K2026.0577.2
PCGrad-SACParams=2517K2026.0574.8
Soft ModularizationParams=3454K2026.0574.4
SAC-MTParams=2517K2026.0569.4