The paper introduces a new benchmark for vision‑language models that tests their ability to decide whether to answer a physics question immediately or to request additional experimental evidence. Each problem presents one measurement image and four possible physical worlds defined by two masses and two values of another property; the model must either stop and answer or choose the cheapest experiment that resolves the question. Across six open models and 144 parameter sets, the models almost always repeat the same action even when the optimal choice changes, and only a single model gets both decisions correct on 5.9% of cases.
arXiv:2609.21942v1 Announce Type: cross Abstract: A robot that fails at a task faces the first decision in corrective dialogue: act on its own diagnosis, consult another onboard sensor, or interrupt...
arXiv:2610.08791v1 Announce Type: new Abstract: Video world models can produce visually convincing yet physically inconsistent sequences, raising concerns about their reliability for prediction and p...
InfiniteWorld is a new benchmark that pairs physical‑reasoning questions with simulator videos whose physical parameters are systematically altered, allowing researchers to test whether models rely on visual evidence or on familiar physics. The benchmark’s generation pipeline connects simulation, scene screening, paired video translation, question construction, and input checks, and it includes five tasks—law identification, parameter estimation, temporal reasoning, analytic counterfactual comparison, and law induction. Evaluation across five matched input settings (text‑only, ordered, shuffled observations) reveals that video‑language models often default to Earth‑normal physics, showing strong accuracy on normal questions but poor recognition of altered dynamics. "whyItMatters":"The benchmark provides a measurable way to separate visual evidence from familiar physics, exposing persistent biases in current video‑language models toward Earth‑normal rules."
arXiv:2509. 12263v3 Announce Type: replace Abstract: Large multimodal models (LMMs) encode physical laws observed during training, such as momentum conservation, as parametric knowledge.
arXiv:2609.13225v1 Announce Type: cross Abstract: Benchmarks agree that vision-language models reason poorly about low-level manipulation, but an aggregate accuracy score does not say which step fail...
arXiv:2609.13152v1 Announce Type: new Abstract: Large language models (LLMs) perform strongly on static science benchmarks, yet their ability to reason about the physical world through active experim...
arXiv:2609.13308v1 Announce Type: cross Abstract: A companion evaluation found that naming the target part in a manipulation prompt increased action accuracy by 0.32-0.63 across eight vision-language...
arXiv:2603. 06828v2 Announce Type: replace-cross Abstract: We uncover a behavioral law of long-horizon vision-language models: models that maintain temporally grounded beliefs generalize better.
The paper introduces BedBench, a benchmark of thirteen procedurally generated tasks designed to test vision‑language models’ ability to recover and consistently use exact 3D object properties. Across twelve open‑weight and four frontier models, the study finds a stark gap: humans achieve 93.3 % accuracy on per‑color counting, while the best model scores only 7.2 %. Even targeted fine‑tuning can raise a model’s accuracy to 68.6 % but leaves its answers about the most common colors inconsistent with its own per‑color counts, highlighting that recovering a property does not guarantee its consistent use.
arXiv:2609.22588v1 Announce Type: new Abstract: Vision-language models (VLMs) perform strongly on visual question answering benchmarks, yet often make decisions that contradict visual evidence they h...
arXiv:2606. 30686v1 Announce Type: cross Abstract: Vision-Language-Action (VLA) systems, built on pretrained vision-language models (VLMs), have shown rapidly improving performance on robot manipulation benchmarks.
The paper introduces a five-task diagnostic experiment that separates perceptual and reasoning failures in multimodal large language models on physics and geometry benchmarks. It finds that misinterpreting diagrams hurts performance even on text-only solvable problems, and that accuracy improves when models receive human-authored captions. The study shows that correcting captions can recover many errors, revealing distinct reasoning bottlenecks that differ by domain, while a heavily pretrained model still underperforms and often truncates reasoning traces.