2 Related Work
2.1 Interactive Video World Models Large-scale video generators [ Brooks et al., 2024 ; Polyak et al., 2024 ; Google DeepMind, 2025 ; Kong et al., 2024 ; Wan Team et al., 2025 ] establish visual and temporal priors for generative simulation by synthesizing scene appearance and motion from text or visual inputs. Interactive video world models build on these capabilities by conditioning future observations on user-controlled inputs. Action-conditioned systems [ Bruce et al., 2024 ; Valevski et al., 2025 ; Li et al., 2025b ; He et al., 2025b ; Jiang et al., 2026 ] support movement and interaction, while language-directed methods [ Mao et al., 2026 ; Tang et al., 2025 ] provide semantic control over unfolding events. Explicit camera-control approaches [ Wang et al., 2024 ; He et al., 2025a ] further expose viewpoint trajectories as a generation interface. Together, these interfaces support exploring a generated environment through successive choices, introducing dependencies between current outputs, earlier observations, and the user’s interaction history. Streaming and open-domain systems [ Robbyant Team et al., 2026 ; Gao et al., 2026 ; DreamX Team et al., 2026 ; Zhu et al., 2026b ; Zhu et al., 2026a ; Duan et al., 2026 ; Huang et al., 2026 ] pursue sustained, responsive exploration across diverse environments. Beyond generating locally coherent transitions, maintaining a world through exploration requires relating new observations to previously visited regions. History-conditioned generation and context memory [ Li et al., 2025b ; Sun et al., 2025 ] seek to preserve information from past observations when extending a rollout. These developments also connect to the view of world models as interactive proxies providing feedback for agents [ Yang et al., 2026a ] . They motivate evaluating the stability of a generated environment alongside visual quality and control adherence, particularly when users leave a region and later return. Agentic world modeling connects predictive simulation to decision-making across environments [ Chu et al., 2026 ] . Visual navigation approaches couple language-conditioned future observation prediction with policy learning [ Dong et al., 2026b ] , or integrate visual foresight, planning, and memory within a unified model [ Dong et al., 2026a ] . For autonomous driving, vision-language-action systems bring together perception, linguistic reasoning, and actionable outputs [ Hu et al., 2025a ] ; OneVL uses future visual prediction and language explanations to supervise compact latent reasoning for planning [ Lu et al., 2026b ] . These applications make the persistence of spatial evidence relevant to both generation and downstream decisions. A complementary line examines how generative capabilities are learned and aligned. DanceOPD composes generative capability fields on student-induced states [ Zhou et al., 2026b ] , while on-policy diffusion self-distillation converts reward guidance into intermediate prediction targets [ Zhou et al., 2026a ] . These image-generation studies provide a training perspective on capability alignment that complements evaluation of consistency during interactive rollouts.
2.2 3D Interactive World Models Explicit scene representations enable navigable world generation and editing [ Yu et al., 2025a ; Liang et al., 2025 ; Szymanowicz et al., 2025 ; World Labs, 2025 ; World Labs, 2026 ] . These systems associate generated content with a scene that can be queried from different viewpoints. Complementary methods couple generative priors with reconstructed geometry or spatial conditioning [ Yu et al., 2025b ; Ren et al., 2025 ; Huang et al., 2025b ; Lu et al., 2026a ; Yang et al., 2026b ; InSpatio Team et al., 2026 ; Chen et al., 2026a ] . Point clouds, Gaussian representations, and warped observations provide camera-aligned evidence for synthesizing unseen regions, while allowing learned priors to complete missing appearance. Geometry may directly determine a rendered output or instead guide a generator that retains freedom to revise appearance and complete occluded regions. This combination connects the flexibility of image and video generation with structured constraints on viewpoint changes. Geometry-grounded weather video synthesis offers a related editing perspective, combining semantic appearance conditioning, simulated particle dynamics, and scene geometry to control weather effects [ Qian et al., 2026a ] . The representation taxonomy of Kong et al. [2025b] places video, occupancy, and LiDAR generation within a common 3D/4D world-modeling landscape. Sensor-based world modeling provides complementary representations of dynamic scenes. DynamicCity models semantic 4D occupancy [ Bian et al., 2025 ] , while LiDARCrafter generates and edits temporally coherent LiDAR sequences [ Liang et al., 2026a ] and U4D introduces uncertainty-conditioned generation [ Xu et al., 2026b ] . Cross-sensor pretraining [ Kong et al., 2026 ] and multimodal scene-consistency learning [ Kong et al., 2025a ] address representation learning in these sensor-rich settings. Together, these works motivate assessing scene structure and dynamics through evidence appropriate to each representation. Spatial memory extends these mechanisms beyond short temporal contexts. Surfel-indexed view retrieval [ Li et al., 2025c ] , geometry-guided correspondences [ Shen et al., 2026 ] , and geometry-aligned or external memory [ AlayaWorld Team et al., 2026 ; Yin et al., 2026 ; Qian et al., 2026b ] reconnect current viewpoints with previously observed content. These advances address spatial forgetting and drift, and existing studies include reconstruction, revisit, or geometric-consistency diagnostics. Nevertheless, using 3D information does not by itself establish whether original input content is faithfully retained, newly generated regions persist after leaving view, and repeated observations agree as one 3D world. \texorpdfstring \pdfxform AOrbia evaluates these complementary requirements jointly under controlled exploration, enabling comparisons across explicit 3D, geometry-guided, and memory-augmented approaches.
2.3 World Model Evaluation
Video benchmarks characterize visual quality, temporal consistency, and prompt alignment [ Huang et al., 2024 ; Liu et al., 2024 ] , while faithfulness-oriented evaluations probe physical plausibility and commonsense [ Zheng et al., 2025 ; Bansal et al., 2025 ; Meng et al., 2025 ; Li et al., 2025a ] . Embodied benchmarks connect generated dynamics to action-level behavior and functional utility [ Qin et al., 2024 ; Hu et al., 2025b ; Shang et al., 2026 ] . WorldLens [ Liang et al., 2026b ] evaluates driving world models across generation, reconstruction, action following, downstream tasks, and human preference. Complementing this benchmark, 3D and 4D World Modeling: A Survey [ Kong et al., 2025b ] organizes evaluation perspectives spanning generation, forecasting, planning, reconstruction, and downstream utility. Together, perceptual and functional criteria answer related but distinct questions about whether generated videos are visually convincing and useful for interaction.
Evaluation reliability also depends on evidence grounding. DriveBench shows that plausible VLM answers can reflect linguistic priors rather than visual input [ Xie et al., 2025 ] , while Open-o3-Video makes video reasoning traceable through timestamps, objects, and bounding boxes [ Meng et al., 2026 ] . These findings support pairing language-based assessments with explicit visual and geometric evidence.
Interactive suites examine control responsiveness, trajectory adherence, and geometric consistency [ Duan et al., 2025 ; Xu et al., 2026c ; Fang et al., 2026 ; Zhao et al., 2026 ] , while multi-turn protocols assess interaction adherence and resulting state changes [ Ying et al., 2026 ; Tang et al., 2025 ; Wu et al., 2026a ] . Complementary diagnostics isolate deformation and occlusion inconsistency [ Gu et al., 2025 ] , static-scene reconstructability [ Fan and Favaro, 2026 ] , and reconstruction-grounded manipulation [ Wang et al., 2026b ] . Geometric evaluation therefore already addresses several failure modes, with different protocols emphasizing different aspects of spatial coherence.
Memory-oriented tests examine returns to previously observed scenes and recovery of persistent or dynamically updated states [ Ye et al., 2026 ; Lu et al., 2026c ; Chen et al., 2026b ; Chen et al., 2026d ] , revisit geometry [ Xu et al., 2026a ] , and calibrated revisit similarity [ Gu et al., 2026b ] . Long-horizon tests track drift and state-returning behavior [ Wu et al., 2026b ; Gu et al., 2026a ] , while agent-based evaluators provide transition and persistence judgments [ Chen et al., 2026c ] . Building on these complementary perspectives, \texorpdfstring \pdfxform AOrbia jointly measures preservation of the input scene, persistence of newly generated regions, and 3D self-consistency across repeated exploration. This organization separates fidelity to observed evidence from retention of generated content and compatibility of generated views. Table 1 summarizes the coverage of representative benchmarks.
3 The \texorpdfstring \pdfxform AOrbia Benchmark
\texorpdfstring \pdfxform AOrbia systematically evaluates the 3D consistency of interactive video world models. In this section, we introduce the evaluation principles and protocols (Section 3.1 ), the data construction and annotation pipeline (Section 3.2 ), and the dataset scale and coverage across scene sources and exploration horizons (Section 3.3 ).