AI Paper Atlas

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

SPAR: Support-Preserving Action Rectification

Jiaxin Zhao, Weihang Pan, Xun Liang, Binbin Lin

ICML 2026 · Reinforcement learning › Policy gradient and actor-critic · approach: Trajectory reuse in policy gradients · 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 latent self-imitation to residual policy (ablation)SPAR residual · Pen-Clonednormalized score: 67.8 -> 76.2Improved
add latent self-imitation to residual policy (ablation)SPAR residual · Hopper medium-replaynormalized score: 72.7 -> 101.9Improved
add latent self-imitation to residual policy (ablation)SPAR residual · AntMaze Large-diversenormalized score: 40.0 -> 71.0Improved
restrict self-imitation to the anchored residual space rather than global action space (ablation)SPAR · Hopper medium-replaynormalized score: 12.1 -> 101.9Improved
restrict self-imitation to the anchored residual space rather than global action space (ablation)SPAR · AntMaze Large-diversenormalized score: 0.0 -> 71.0Improved
add Stage III value-gated rectification (ablation)SPAR · Hopper medium-replaynormalized score: 49.5 -> 101.9Improved
add Stage III value-gated rectification (ablation)SPAR · AntMaze Large-diversenormalized score: 60.0 -> 71.0Improved
add Stage III value-gated rectification (ablation)SPAR · HalfCheetah medium-expertnormalized score: 96.1 -> 97.0Improved
add uncertainty penalty (lambda_u=0 to 0.5) (ablation)SPAR-PROJ · Pen-Clonednormalized score: 39.2 -> 76.2Improved
add uncertainty penalty (lambda_u=0 to 0.5) (ablation)SPAR-MLP · Hopper medium-replaynormalized score: 55.2 -> 101.9Improved
increase uncertainty penalty from 0.5 to 2 (ablation)SPAR-PROJ · AntMaze Large-diversenormalized score: 71.0 -> 60.0Hurt
use soft rather than hard advantage filtering (ablation)SPAR-PROJ · AntMaze Large-diversenormalized score: 30.0 -> 71.0Improved

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