⌁ Awesome World Model Hub Papers · Datasets · Projects
中文 GitHub
← Back to papers

EnerVerse: Envisioning Embodied Future Space for Robotics Manipulation

AgiBot 2025 59.6 method, system, application

TLDR

EnerVerse is a generative robotics foundation model using video diffusion to predict future embodied spaces for manipulation, achieving SOTA in sim and real tasks.

Reasoning

Strengths include a novel chunk-wise autoregressive video diffusion framework, multi-view representation for 3D grounding, and a data engine reducing sim-to-real gap. Weaknesses are limited detail on the policy head and scope of real-world evaluation.

Read-first score

Read-first score 59.6, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 51.

Recency 8%
86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Topical relevance 42%
72.9

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 25%
50

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

Reproducibility 25%
38

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

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 375.

Keyword Scores

generative world model
9
video world model
9
world model
8
world dynamics prediction
8
world simulator
7
interactive world model
6
model-based reinforcement learning world model
4

Deep Analysis

Innovations

  • Chunk-wise autoregressive video diffusion framework for predicting future embodied spaces
  • Sparse context memory for long-term reasoning
  • Multi-view video representation to address motion ambiguity and 3D grounding
  • Data engine pipeline combining generative modeling with 4D Gaussian Splatting for a self-reinforcing data loop
  • Policy head EnerVerse-A that reuses features from the first denoising step and predicts action chunks

Methodology

EnerVerse employs a chunk-wise autoregressive video diffusion framework to predict future embodied spaces from instructions, enhanced by a sparse context memory for long-term reasoning. It adopts a multi-view video representation to model the 3D robotics world, and uses a data engine (EnerVerse-D) that combines generative modeling with 4D Gaussian Splatting to reduce the sim-to-real gap. The policy head (EnerVerse-A) translates 4D world representations into physical actions by reusing features from the first denoising step and predicting action chunks.

Key Results

EnerVerse achieves state-of-the-art performance in both simulation and real-world tasks, with an efficiency of about 280 ms per 8-step action chunk on a single RTX 4090.

Resources

PaperWebsite

Tags