AI Paper Atlas

Research map / Reinforcement learning / Visual and representation RL

Learning Generalizable Skill Policy with Data-Efficient Unsupervised RL

Jongchan Park, Seungjun Oh, Seungho Baek, Yusung Kim

ICML 2026 · Reinforcement learning › Visual and representation RL · approach: Unsupervised skill discovery · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

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
add Complementary Information Bottleneck (CIB) (ablation)GenDa · Quadruped-Pixels unseen Unicolor backgroundperformance: 137 -> 268Improved
add CIB (ablation)GenDa · Quadruped-Pixels unseen Gradation backgroundperformance: 78 -> 206Improved
add CIB (ablation)GenDa · Quadruped-Pixels original backgroundperformance: 335 -> 331No clear effect
add skill relabeling (ablation)GenDa · Quadruped-Numeric pre-trainingstate coverageImproved
add CIB (ablation)GenDa · Quadruped-Numeric downstream (RS, RG)success rateImproved
increase pre-training from 2M to 10M steps (ablation)GenDa · Humanoid-Numeric MazeHsuccess rate: 0.14 -> 0.41Improved
increase pre-training from 2M to 10M steps (ablation)GenDa · Humanoid-Numeric (RS, RG)success rate: 0.77 -> 0.68Hurt
vary beta from 1.0 to 4.0 (ablation)GenDa · Humanoid-Numericstate coverage: 2192 -> 2595No clear effect
vary beta from 1.0 to 0.1 (ablation)GenDa · Quadruped-Pixelstate coverage: 331 -> 286No clear effect
change c-tau mixture ratio from 0.5:0.5 to 0.25:0.75 (ablation)GenDa relabeling · Quadruped-Numericstate coverage: 1851 -> 1820No clear effect
raise z_roll threshold from 0.4 to 0.95 (ablation)GenDa relabeling · Quadruped-Numericstate coverage: 1851 -> 1678Hurt
use GenDa instead of METRA in random-start downstreamGenDa · Dog-Numeric (RS, RG)success rate: 0.00 -> 0.53Improved

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