OrbiSim: World Models as Differentiable Physics Engines for Embodied Intelligence

arXiv 2026 53.6 method, system

TLDR

OrbiSim redefines world models as a differentiable physics engine for embodied intelligence, enabling end-to-end differentiability for simulation and policy optimization.

Reasoning

The paper introduces a novel paradigm that bridges structured scene assets, neural dynamics, and RL through full differentiability, with empirical results showing improved predictive fidelity and control. However, the abstract lacks explicit details on real-world experiments or specific benchmarks, and some claims about potential are not fully supported by visible evidence.

Read-first score

Read-first score 53.6, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 46.

Recency 6%

100

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

Citation impact 18%

80.7

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.80689593

Topical relevance 29%

65.7

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=result

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: 358.

Keyword Scores

world model

world dynamics prediction

model-based reinforcement learning world model

world simulator

interactive world model

video world model

generative world model

Deep Analysis

Innovations

Redefines world models as a fully differentiable physics engine for embodied intelligence
Establishes a unified, physically-grounded pathway bridging structured scene assets, neural dynamics, and downstream reinforcement learning
Enables end-to-end differentiability throughout the entire simulation loop, from explicit state transitions to visual observation generation
Supports tasks traditionally intractable for classical simulators, such as differentiable contact modeling, gradient-based policy optimization under sparse rewards, and intuitive physical inference

Methodology

OrbiSim is a robotic simulation paradigm that integrates structured scene assets, neural dynamics, and reinforcement learning into a fully differentiable physics engine. It achieves end-to-end differentiability across the entire simulation loop, covering state transitions and visual observation generation, enabling gradient-based optimization for policy learning and physical inference.

Key Results

OrbiSim significantly outperforms state-of-the-art world models in both predictive fidelity and control performance, and demonstrates consistent responsiveness to asset configurations and physical parameters.

Resources

Paper