AI Paper Atlas

Research map / Reinforcement learning / Visual and representation RL

ReLAM: Learning Anticipation Model for Rewarding Visual Robotic Manipulation

Nan Tang, Jing-Cheng Pang, Guanlin Li, Chao Qian, Yang Yu

ICML 2026 · Reinforcement learning › Visual and representation RL · approach: Goal-conditioned policy learning

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
use 4 keypoints instead of 8 or 12 (ablation)ReLAM · Meta-World Drawer Opensuccess rateImproved
use piecewise linear reward instead of exponential or logistic (ablation)ReLAM · Meta-World Drawer Opensuccess rateImproved
use piecewise linear reward instead of pure linear (ablation)ReLAM · Meta-World Drawer Opensuccess rateImproved
generate subgoals with the anticipation model instead of ground-truth keypoint subgoals (ablation)ReLAM · Meta-World / ManiSkill taskssuccess rateNo clear effect
represent subgoals as keypoints instead of images (ablation)ReLAM · Meta-World Drawer Open (online RL)mean success rate (%): 4.0 -> 100.0Improved
represent subgoals as keypoints instead of images (ablation)ReLAM · ManiSkill Push Cube (offline RL)mean success rate (%): 46.7 -> 89.3Improved
use ReLAM keypoint distance reward instead of Diffusion RewardReLAM · Meta-World Drawer Open (online RL)mean success rate (%): 80.0 -> 100.0Improved
use ReLAM keypoint distance reward instead of DACfOReLAM · ManiSkill Pick Cube (offline RL)mean success rate (%): 78.7 -> 88.0Improved
use ReLAM instead of Diffusion Reward on Door OpenReLAM · Meta-World Door Open (online RL)mean success rate (%): 100.0 -> 100.0No clear effect
use ReLAM instead of ground-truth image subgoal OracleReLAM · Meta-World Button Press Wall (online RL)mean success rate (%): 76.7 -> 75.8Hurt

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