World-Ego Modeling for Long-Horizon Evolution in Hybrid Embodied Tasks

2026-05-19

AI Summary

Existing world models for embodied intelligence suffer performance limitations in long-horizon hybrid tasks—such as those intertwining navigation and manipulation—due to the tight coupling between world and ego dynamics. This work proposes a world-ego modeling paradigm that decouples these dynamics through three complementary perspectives: motion, semantics, and intent. We introduce a unified World-Ego Model (WEM), integrating an implicitly disentangled planner with a cascaded parallel mixture-of-experts diffusion generator. For the first time, we formalize this decoupling paradigm, devise three distinct decoupling strategies, and establish HTEWorld, the first benchmark specifically designed for long-horizon hybrid world modeling. Evaluated on HTEWorld—which comprises 125K video clips and 300 multi-episode trajectories—WEM achieves substantial performance gains while maintaining competitive results on pure manipulation benchmarks.

Abstract

World models are widely explored in embodied intelligence, yet they typically predict distinct evolutions of the world and the ego within a single stream, where the world captures persistent instruction-agnostic scene regularities and the ego captures robot-centric instruction-conditioned dynamics. This world-ego entanglement leads to a degradation in long-horizon embodied scenarios, particularly in hybrid tasks with interleaved navigation and manipulation behaviors. In this paper, we introduce \emph{World-Ego Modeling}, a new conceptual paradigm that decomposes future evolution into world and ego components. We define the world-ego boundary from three perspectives, i.e., motion-, semantic-, and intention-based views, and analyze three disentanglement strategies with post-, pre-, and full disentanglement. Further, we instantiate this paradigm as the World-Ego Model (WEM), a unified embodied world model that couples an implicit separate world-ego planner with a cascade-parallel mixture-of-experts (CP-MoE) diffusion generator. To enable rigorous evaluation, we further construct HTEWorld, the first benchmark for long-horizon world modeling with hybrid navigation-manipulation tasks, providing 125K video clips (over 4.5M frames) with fine-grained action annotations and 300 multi-turn evaluation trajectories (over 2K instructions). Extensive experiments show that WEM achieves state-of-the-art performance on HTEWorld while remaining competitive on existing manipulation-only benchmarks.

Problem

• world models
• embodied intelligence
• long-horizon tasks
• hybrid tasks
• world-ego entanglement

Innovation

• World-Ego Modeling
• disentanglement
• long-horizon embodied tasks
• hybrid navigation-manipulation
• diffusion-based world model