AI Paper Atlas

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

Langevin Rollout Optimization for Modelic Reinforcement Learning

Tianyi Zhang, Likun Wang, Guojian Zhan, Feihong Zhang, Yang Guan, Yao Lyu, Shengbo Eben Li

ICML 2026 · Reinforcement learning › Policy gradient and actor-critic · approach: Policy gradients for special settings

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 LaRO (Langevin gradient refinement) in parallel with MPPI via MLAPBOOM · H-Bench Hand (7 tasks avg)Total Average Return: 634.2 -> 773.3Improved
add LaRO via MLAPBOOM · DMC Suite Hard (7 tasks avg)Total Average Return: 802.7 -> 847.2Improved
add LaRO via MLAPBOOM · H1hand-runTotal Average Return: 329.3 -> 823.0Improved
use learned score network instead of exact BPTT gradients (ablation)LaRO · H1hand-runTotal Average ReturnNo clear effect
use learned score network instead of BPTT gradients (ablation)LaRO · H1hand-runlatency (ms): 12.72 -> 1.79Improved
lower Langevin temperature alpha (towards deterministic gradient ascent) (ablation)LaRO · H1hand-runTotal Average ReturnHurt
use 3 Langevin steps instead of 5 (ablation)LaRO · H1hand-runTotal Average ReturnHurt
use 8 Langevin steps instead of 5 (ablation)LaRO · H1hand-runTotal Average ReturnHurt
LaRO-only planner versus MPPI-only planner (ablation)BOOM-L · H1hand-runTotal Average ReturnNo clear effect
combine MPPI and LaRO with max look-ahead selection (MLAP) vs either alone (ablation)BOOM-L · H1hand-runTotal Average ReturnImproved
add LaRO and score network on top of BOOMBOOM-L · Average of 4 tasksinference latency (ms): 28.95 -> 34.76Hurt
add LaRO and score network on top of BOOMBOOM-L · Average of 4 taskstraining time (h): 20.08 -> 25.51Hurt

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