{
  "id": 148,
  "url": "https://arxiv.org/abs/2607.06522v1",
  "title": "Bridging Physical Reasoning and Task Generalization via Visual Action Outcome Reasoning Alignment",
  "summary": "Vision-language models (VLMs) struggle to generalize in interactive physical reasoning, particularly under unseen tasks and environments. Two key failure modes are prominent: hallucinated chain-of-thought (CoT) reasoning that contradicts physical reality, and misalignment between the model's reasoning and actions. We present VAORA (Visual Action Outcome Reasoning Alignment), a novel reward design that directly addresses both issues. VAORA introduces two complementary rewards: Visual Alignment Re",
  "authors": "Han-Jun Ko, Jr-Jen Chen, Haobo Yuan, Hsin-Ying Lee, Tiancheng Shen, Ming-Hsuan Yang et al.",
  "category": "research",
  "topics": "safety-alignment,environment",
  "orgs": null,
  "regions": null,
  "published_at": "2026-07-07T17:27:59.000Z",
  "fetched_at": "2026-07-14T14:14:19.969Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/148",
  "original_url": "https://arxiv.org/abs/2607.06522v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}