AI Paper Atlas

Research map / Reinforcement learning / Policy gradient and actor-critic

User-Oriented Robust Reinforcement Learning

Haoyi You, Beichen Yu, Haiming Jin, Zhaoxing Yang, Jiahui Sun

AAAI 2023 · Reinforcement learning › Policy gradient and actor-critic · approach: Model-based RL with guarantees · 1 citations (OpenAlex)

Read the paper

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Optimize the user-weighted UOR objective with the known environment distribution (DB-UOR-RL)UOR-RL · MuJoCo tasksUOR E₁ (k=1): DB-UOR-RL is best in each environmentImproved
Optimize the user-weighted UOR objective without a known environment distribution (DF-UOR-RL)UOR-RL · MuJoCo tasksUOR E₁ (k=1): 13.63 ± 0.59 vs 12.57± 0.59 (Reacher); 3213 ± 399.13 vs 2307 ± 192.94 (Ant)Improved
Use UOR-RL with preference parameter k=0, equivalent to average-return optimizationUOR-RL · MuJoCo tasksAverage return: 19.97 ± 1.11 vs 19.28 ± 1.21 (Reacher); 2910 ± 135.06 vs 2652 ± 125.13 (Half Cheetah)Mixed
Set robustness degree k=21, weighting poor-performing trajectories more stronglyUOR-RL · MuJoCo tasksWorst 10% trajectory average return: 3.82 ± 0.85 vs -0.14± 1.79 (Reacher); 2497 ± 404.51 vs 2523 ± 195.72 (Half Cheetah)Mixed
Increase preference parameter k, shifting emphasis toward lower-performing trajectories (ablation)DB-UOR-RL and DF-UOR-RL · Half CheetahAverage return of trajectories by sorted groupMixed
Use the known environment-parameter distribution to evaluate UOR rather than estimating it distribution-freeUOR-RL · MuJoCo tasksUOR E₁ (k=1): 14.38 ± 0.86 vs 13.63 ± 0.59 (Reacher); 3212 ± 187.34 vs 3213 ± 399.13 (Ant)Mixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy