AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

General Process Reward Modeling for Robotic Reinforcement Learning

Huajie Tan, Sixiang Chen, Yijie Xu, Zixiao Wang, Cheng Chi, Yuheng Ji, Yaoxu Lyu, Zhongxia Zhao, Xiansheng Chen, Peterson Co, Shaoxuan Xie, Guocai Yao and 3 more

CVPR 2026 · Reinforcement learning › RL from feedback · approach: RL for LLM reasoning

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 policy-invariant reward shaping instead of raw progress increment reward (ablation)Dopamine-RL · real-world 3 tasksaverage success rate (%): 41.3 -> 85.0Improved
add one-shot GRM adaptation from a single demonstration (ablation)Dopamine-RL · real-world 3 tasksaverage success rate (%): 63.2 -> 85.0Improved
add multi-perspective progress fusion over incremental-only (ablation)Dopamine-RL · real-world 3 tasksaverage success rate (%): 70.0 -> 85.0Improved
add fusion over forward-anchored-only (ablation)Dopamine-RL · real-world 3 tasksaverage success rate (%): 65.7 -> 85.0Improved
add fusion over backward-anchored-only (ablation)Dopamine-RL · real-world 3 tasksaverage success rate (%): 62.5 -> 85.0Improved
use multi-view instead of single-view inputs for GRM-8B (ablation)GRM-8B · task completion classificationaccuracy (%): 83.9% -> 92.8%Improved
use multi-view instead of single-view for GRM-3B (ablation)GRM-3B · VOC average, dense samplingValue-Order Correlation: 0.87 -> 0.93Improved
scale GRM from 3B to 8B (single-view) (ablation)GRM · VOC average, sparse samplingValue-Order Correlation: 0.91 -> 0.92No clear effect
add dense rewards from GRM via Dopamine-RL vs sparse-reward RLRL policy · Simulation 10 taskssuccess rate (%): 79.9 -> 81.0Improved
add dense rewards from GRM vs sparse-reward RLreal-world RL · Real-world 8 taskssuccess rate (%): 68.0 -> 95.2Improved
add Dopamine-RL to improve OOD generalization over BCpolicy · Insert Square, Circuit, Cap Pen OODavg relative drop (%): 57.1 -> 19.3Improved
denser temporal sampling for baseline reward modelsVLAC-2B and GVL · VOCValue-Order CorrelationNo clear effect

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