AI Paper Atlas

Research map / Reinforcement learning / Offline RL

Offline Reinforcement Learning with Universal Horizon Models

Hojun Chung, Junseo Lee, Songhwai Oh

ICML 2026 · Reinforcement learning › Offline RL · approach: Simple practical offline RL · 0 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
add lambda scheduling for UHM training (ablation)UHM · OGBench (antmaze-giant-navigate-1, puzzle-3x3-play-2, cube-single-play-1)success rateImproved
add terminal state handling (absorbing terminal states, no bootstrapping) (ablation)UHM · OGBench (Hmaze-medium-navigate-1, puzzle-3x3-play-4, cube-single-play-2)success rateImproved
winsorize horizon at quantile q=0.1 or 0.2 instead of q=1e-8 (ablation)UHM · Long-horizon OGBench taskssuccess rateImproved
behavior mixing coefficient beta 0.3 versus 0.0 and 1.0 (ablation)UHM · Representative standard OGBench tasksaverage success rate: 0.63 / 0.66 / 0.59 for beta 0.0 / 0.3 / 1.0Mixed
use UHM n-step future prediction instead of GHM geometric horizon model (ablation)Model-based value expansion · OGBench 25 long-horizon tasksaverage success rate (%): 16 -> 22Improved
use UHM instead of GHM on standard tasks (ablation)Model-based value expansion · OGBench 50 standard tasksaverage success rate (%): 48 -> 55Improved
use UHM on-policy imagined rollouts instead of single-step dynamics model rollouts MBTD(λ) (ablation)TD(λ) value learning · OGBench 50 standard tasksaverage success rate (%): 45 -> 55Improved
use dataset trajectories (DTD(λ)) instead of imagined on-policy trajectories (ablation)TD(λ) value learning · OGBench 25 noisy tasksaverage success rate (%): 39 -> 23Hurt
use UHM instead of dataset trajectories (DTD(λ)) on long-horizon tasks (ablation)TD(λ) value learning · OGBench 25 long-horizon tasksaverage success rate (%): 13 -> 22Improved
use n-step return TD instead of one-step TD (ablation)Offline RL critic · OGBench 50 standard tasksaverage success rate (%): ReBRAC† 35 -> DTD(λ) 52Improved
use single-step dynamics model rollouts for long-horizon imaginationMBTD(λ) · OGBench 25 long-horizon tasksaverage success rate (%): 5 vs ReBRAC† 5No clear effect
model-based methods with uncertainty-based reward penaltiesMOPO and MOBILE · OGBench sparse-reward standard taskssuccess rate: MOPO 4, MOBILE 10 averageHurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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