WorldReward: Reward Modeling for Camera-Conditioned World Models

• URL: http://arxiv.org/abs/2609.03952v1
• Date: Thu, 03 Sep 2026 14:53:53 GMT
• Title: WorldReward: Reward Modeling for Camera-Conditioned World Models
• Abstract summary: We present WorldReward, a VLM-based pairwise preference reward model for camera-conditioned world models.<n>WorldReward decomposes paired videos into action-aligned chunks, organizes each chunk into structured visual evidence, and aggregates chunk-level decisions by voting into separate video-level action and visual-quality preferences.<n>WorldReward achieves the highest agreement on all three dimensions, exceeding GPT-5.5 by 3.42, 1.45, and 3.56 percentage points, respectively.
• Score: 56.14617966299295
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Camera-conditioned world models generate interactive videos in which commanded actions should induce the expected scene changes while appearance, geometry, and temporal dynamics remain coherent. Existing rewards assess these requirements separately: geometry-based rewards estimate trajectory execution but cannot judge the visual quality of the executed motion, whereas image-based rewards measure frame quality without capturing action execution or temporal dynamics. We posit that a vision-language model (VLM) offers a shared reasoning space for relating actions to their visual outcomes. However, judging a complete long video against its full action sequence creates a lengthy, noisy context in which short-lived local action evidence can be missed or diluted. We present WorldReward, a VLM-based pairwise preference reward model that unifies action-consistency and visual-quality evaluation for camera-conditioned world models. WorldReward decomposes paired videos into action-aligned chunks, organizes each chunk into structured visual evidence, and aggregates chunk-level decisions by voting into separate video-level action and visual-quality preferences. To train it, we construct a large-scale reasoning-augmented preference dataset using structured judgments generated by a frontier VLM and refined through tool-based agent auditing and targeted human review. We further introduce WorldReward-Bench, a human-annotated benchmark measuring reward-model agreement with human preferences across action consistency, appearance quality, and motion quality. WorldReward achieves the highest agreement on all three dimensions, exceeding GPT-5.5 by 3.42, 1.45, and 3.56 percentage points, respectively. When used for RL post-training of HY-WorldPlay 1.5, it consistently improves both action execution and visual quality across short- to long-term horizons.

Related papers

• VOR-Bench: A Human Perception-Driven Benchmark for Video Object Removal [41.67024349583551]
  We introduce VOR Bench, which advances VOR evaluation through three integrated components.<n>First, we present the VOR dataset (VORD), the first benchmark dataset providing both paired edited videos and graffiti masks.<n>Second, we develop rMPAF, a realistic Motion- capable Paired-video Acquisition Framework.<n>Third, we introduce VOR-MDSM, the first perception-driven VLM-based scoring model specifically designed for mask-guided VOR.
  arXiv  Detail & Related papers  (2026-09-15T09:08:00Z)
• PlayWorld: Benchmarking World Models with Agent Players over Long-Horizon Objectives [86.3073067035531]
  Video world models simulate future states conditioned on current observations and user actions.<n>Recent systems have demonstrated impressive video consistency and action controllability over long sequences.<n>But fairly comparing these interactive models remains challenging.<n>We employ multi-modal Agent Players to interact with world models toward specified long-horizon objectives.
  arXiv  Detail & Related papers  (2026-08-13T17:59:30Z)
• WorldExam: Benchmarking World Models from Apparent Appearance to Inherent Reactivity [75.96851880065331]
  We introduce WorldExam, a diagnostic benchmark spanning four levels: Visual Quality, Control Adherence, Spatial Consistency, and World Reactivity.<n>It comprises 1,474 cases across eight dedicated tasks and supports unified evaluation of camera-, action-, and language-driven model paradigms.<n> Evaluation of 20 representative models reveals a clear capability split. Camera-driven models excel at camera control, but their interfaces do not support dynamic interaction.
  arXiv  Detail & Related papers  (2026-08-03T17:59:54Z)
• SC3-Eval: Evaluating Robot Foundation Models via Self-Consistent Video Generation [73.2142090645987]
  SC3-Eval is a self-consistent video generation recipe that adapts a pre-trained video foundation model into an accurate policy evaluator.<n>We demonstrate SC3-Eval rollouts reproduce the failure modes that policies exhibit in real-world rollouts, supporting fine-grained diagnostic comparison.
  arXiv  Detail & Related papers  (2026-06-17T02:15:46Z)
• WorldOlympiad: Can Your World Model Survive a Triathlon? [39.69359039523777]
  We introduce WorldOlympiad, a benchmark for diagnosing video-based world models across physical faithfulness, geometric consistency, and interaction fidelity.<n>WorldOlympiad decomposes world-model evaluation into three complementary dimensions.
  arXiv  Detail & Related papers  (2026-06-09T17:24:36Z)
• DisCo: World Models with Discrete Camera Motion Control [79.86256515640231]
  We propose DisCo, a controllable video world model that conditions generation on a compact set of discrete action primitives to improve action separability.<n>DisCo achieves significantly more reliable action following while preserving visual quality.
  arXiv  Detail & Related papers  (2026-06-06T03:50:45Z)
• Towards High-Consistency Embodied World Model with Multi-View Trajectory Videos [24.111891848073288]
  Embodied world models aim to predict and interact with the physical world through visual observations and actions.<n>MTV-World introduces Multi-view Trajectory-Video control for precise visuomotor prediction.<n>MTV-World achieves precise control execution and accurate physical interaction modeling in complex dual-arm scenarios.
  arXiv  Detail & Related papers  (2025-11-17T02:17:04Z)
• SVAG-Bench: A Large-Scale Benchmark for Multi-Instance Spatio-temporal Video Action Grounding [48.64661382961745]
  We introduce Spatio-temporal Video Action Grounding (SVAG), a novel task that requires models to simultaneously detect, track, and temporally localize all referent objects in videos.<n>To support this task, we construct SVAG-Bench, a large-scale benchmark comprising 688 videos, 19,590 annotated records, and 903 unique verbs.<n> Empirical results show that existing models perform poorly on SVAG, particularly in dense or complex scenes.
  arXiv  Detail & Related papers  (2025-10-14T22:10:49Z)
• DynamicEval: Rethinking Evaluation for Dynamic Text-to-Video Synthesis [17.750053029702222]
  Existing text-to-video (T2V) evaluation benchmarks, such as VBench and EvalCrafter, suffer from two limitations.<n>We introduce DynamicEval, a benchmark consisting of systematically curated prompts emphasizing dynamic camera motion.<n>For background scene consistency, we obtain the interpretable error maps based on the Vbench motion smoothness metric.<n>Our proposed metrics achieve stronger correlations with human preferences at both the video level and the model level.
  arXiv  Detail & Related papers  (2025-10-08T18:41:04Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.