Feat2Go: Visual Feature-Grounded Value Estimation for Embodied Reinforcement Learning

arXiv 2026 45.1 method

TLDR

Feat2Go uses a pretrained visual world model to derive progress targets and train a value model for embodied RL, improving VLA policy performance on manipulation tasks.

Reasoning

The paper presents a novel framework that leverages a pretrained visual world model for fine-grained value estimation, showing strong empirical gains on simulated benchmarks. However, it lacks real-world validation and the reliance on a pretrained world model may limit generalizability.

Read-first score

Read-first score 45.1, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 22.

Recency 6%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Citation impact 18%

90.1

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.90130989

Methodology quality 18%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=experiment

Topical relevance 29%

31.4

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Citation velocity 12%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FoundationFrontierBridge

Rank sensitivity

Stability: volatile; rank range: 362.

Keyword Scores

world model

model-based reinforcement learning world model

world dynamics prediction

world simulator

generative world model

interactive world model

video world model

Deep Analysis

Innovations

Fine-grained value estimation framework using patch-level similarity to subgoal states from a pretrained visual world model
Trend-based clustering to partition episodes into semantic stages for continuous progress targets
Embodied value model trained to predict structural progress from current observation and task instruction, reshaping terminal rewards
Compatibility with existing VLA policy RL pipelines (PPO, GRPO) without manual reward engineering

Methodology

Feat2Go first derives a continuous progress target from a pretrained visual world model by measuring patch-level similarity to subgoal states and partitioning episodes into semantic stages using trend-based clustering. It then trains an embodied value model to predict this structural progress from the current observation and task instruction, and uses the predicted value to reshape terminal rewards during policy optimization, integrating with existing VLA policy RL pipelines like PPO and GRPO.

Key Results

On ManiSkill3, Feat2Go improves OpenVLAOFT from 17.5% to 82.9% average out-of-distribution success while retaining 96.9% in-distribution performance. On RoboTwin 2.0, it achieves an average success rate of 88.8% in domain-randomized task settings, outperforming prior reinforcement learning methods.

Resources

Paper