Meta-Reinforcement Learning research benchmarks

Benchmarks

Dataset NameSOTA methodMetricTrendResultsLast Updated

2-D point robot navigation (meta-test)

DR-MAML+

-19.2Average Return

Feb 26, 2026

Cheetah vel-ood

CSP

0.53FLOPs (k)

Feb 26, 2026

Lunarlander g

CSP

0.015FLOPs (k)

Feb 26, 2026

Cartpole fl-ood

CSP

0.004FLOPs (k)

Feb 26, 2026

InvDoublePend params

CSP

39FLOPs (k)

Feb 26, 2026

Hopper params

CSP

0.51FLOPs (k)

Feb 26, 2026

Walker2d params

CSP

3.11FLOPs (k)

Feb 26, 2026

MuJoCo Ant Body variation (test)

RoML

1,368CVaR 0.05 Return

Feb 26, 2026

MuJoCo Ant Mass variation (test)

RoML

1,385CVaR 0.05 Return

Feb 26, 2026

MuJoCo Ant Goal variation (test)

RoML

-454CVaR 0.05 Return

Feb 26, 2026

MuJoCo Humanoid Body variation (test)

RoML

1,365CVaR 0.05 Return

Feb 26, 2026

MuJoCo Humanoid Mass variation (test)

RoML

1,378CVaR 0.05 Return

Feb 26, 2026

MuJoCo Humanoid Velocity variation (test)

RoML

833CVaR 0.05 Return

Feb 26, 2026

LunarLander

—

—Primary metric

Feb 26, 2026

Highway

—

—Primary metric

Feb 26, 2026

CartPole

—

—Primary metric

Feb 26, 2026