GigaBrain-0.5M: a VLA That Learns From World Model-Based Reinforcement Learning

arXiv 26.2 2026 70 method, system, application

TLDR

GigaBrain-0.5M integrates world model-based RL into VLA, achieving ~30% improvement on real-world robotic tasks.

Reasoning

The paper presents a novel integration of world model-based reinforcement learning with vision-language-action models, showing substantial empirical gains on challenging real-world tasks. However, the abstract lacks detailed methodology for RAMP and does not compare against other world model approaches.

Read-first score

Read-first score 70, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 54.

Recency 8%

100

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

Methodology quality 25%

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

Topical relevance 42%

77.1

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 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=github

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 91.

Keyword Scores

world model

model-based reinforcement learning world model

video world model

world dynamics prediction

generative world model

world simulator

interactive world model

Deep Analysis

Innovations

Integration of world model-based reinforcement learning (RAMP) into VLA models for cross-task adaptation
GigaBrain-0.5M* model that combines large-scale robotic pre-training (10,000 hours) with world model-based RL
Demonstration of substantial performance gains (~30%) over RECAP baseline on challenging manipulation tasks

Methodology

GigaBrain-0.5M* is built upon GigaBrain-0.5, which is pre-trained on over 10,000 hours of robotic manipulation data. It further integrates world model-based reinforcement learning via the RAMP (Reinforcement leArning via world Model-conditioned Policy) method to enable robust cross-task adaptation. The model is evaluated against the RECAP baseline on tasks including Laundry Folding, Box Packing, and Espresso Preparation, with real-world deployment videos for validation.

Key Results

RAMP achieves approximately 30% performance improvements over the RECAP baseline on challenging tasks such as Laundry Folding, Box Packing, and Espresso Preparation. GigaBrain-0.5M* exhibits reliable long-horizon execution, consistently accomplishing complex manipulation tasks without failure as validated by real-world deployment videos.

Resources

Paper Project