DreamX-Phi 1.0: Action-Conditioned Video World Model for Robotic Manipulation

arXiv 2026 47.6 method, application

TLDR

DreamX-Phi 1.0 is an action-conditioned video world model for robotic manipulation, predicting future observations with geometric consistency and efficient distillation, ranking first in WorldArena 2.0.

Reasoning

The paper introduces a strong video world model with novel geometric encoding and object-consistency techniques, validated by challenge rankings. Weaknesses include lack of detailed experimental methodology and limited evidence beyond benchmark results.

Read-first score

Read-first score 47.6, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 58.

Recency 6%

100

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

Topical relevance 29%

82.9

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

Methodology quality 18%

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

Reproducibility 18%

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

Citation impact 18%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. cited_by_count=0

Citation velocity 12%

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

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 557.

Keyword Scores

world model

video world model

generative world model

world dynamics prediction

world simulator

interactive world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Action-conditioned video world model that predicts future observations from an observed frame, a language instruction, and a prescribed action sequence of end-effector poses and gripper states.
Injection of per-arm SE(3) transformations into attention via PRoPE-style geometric encoding to preserve arm identity and rigid-motion structure during prediction.
Addition of a lightweight depth branch to provide scene-level geometry constraints.
Use of SAM3 masks with a frozen V-JEPA teacher to maintain object consistency throughout grasping.
Distribution-matching distillation of the multi-step generator into a few-step student for efficient deployment.

Methodology

DreamX-Phi is an action-conditioned video world model that takes an observed frame, a language instruction, and an action sequence of end-effector poses and gripper states, and predicts future frames. It incorporates per-arm SE(3) transformations into attention via PRoPE-style geometric encoding, a lightweight depth branch for scene-level geometry, and SAM3 masks with a frozen V-JEPA teacher for object consistency. The multi-step generator is distilled into a few-step student using distribution-matching distillation for efficient deployment.

Key Results

At the time of writing, DreamX-Phi achieves first place on Track 1 and second place on Track 2 of the WorldArena 2.0 Challenge.

Resources

Paper