Principia is a new benchmark that tests video models on Newtonian physics by evaluating relational consistency between paired objects, independent of calibration. It covers eight phenomena—gravity, restitution, friction, rotational inertia, projectile motion, momentum, pendulum, and mass‑spring oscillation—across translational, rotational, collisional, and oscillatory dynamics using real‑world scenes. The benchmark introduces a calibration‑independent consistency score and shows that current state‑of‑the‑art video generators perform poorly on it, with the best vision‑language model achieving only 67% accuracy on detecting physics violations.
By Varun Varma Thozhiyoor, Shivam Tripathi, Venkatesh Babu Radhakrishnan, Anand Bhattad
CALIPER is a new benchmark that tests whether pretrained visual encoders can infer physical properties such as mass and friction from images. The test involves striking an object twice at known speeds, showing a third strike only up to contact, and asking a linear readout on frozen features to predict how far the object slides. Results show that in clean, fixed‑camera scenes all representations perform similarly, but when camera, lighting, and clutter are varied, only encoders that truly infer physics—like V‑JEPA 2—maintain performance, while random or raw pixel representations fail.
By Aman Mehta, Riya Baviskar
CompAdapt is a physics-consistent text-to-video generation framework that extends diffusion-based models to handle composite physical behaviors such as coupled motions, multi-stage transitions, and multi-object collisions. It translates natural language prompts into structured physical semantics, enabling end-to-end specification of motion types, temporal relations, and initial parameters. The system introduces dynamics-aware prior matching for one-shot adaptation to new physical environments and a physics-aware latent feature fusion module to enhance visual fidelity during fast, complex motion, outperforming existing physics-constrained baselines on physics-focused T2V benchmarks.
By Haoran Qin (Harbin Institute of Technology, China), Renlong Wu (Harbin Institute of Technology, China), Tianyu Huang (Harbin Institute of Technology, China), Yukang Ding (Taobao, Alibaba Group, China), Hui Li (Harbin Institute of Technology, China), Wangmeng Zuo (Harbin Institute of Technology, China)
arXiv:2606. 28128v1 Announce Type: cross Abstract: Video generation models have emerged as a promising paradigm for embodied world simulation.
By Peiwen Zhang, Yufan Deng, Shangkun Sun, Juncheng Ma, Duomin Wang, Jonas Du, Zilin Pan, Ye Huang, Hao Liang, Songyan Huang, Ruihua Zhang, Enze Xie, Ming-Yu Liu, Daquan Zhou
arXiv:2606.09646v2 Announce Type: replace-cross
Abstract: We study whether pretrained video foundation models encode intuitive-physics information in their frozen representations, and how this inform...
By Samuele Punzo, Niccol\`o Caselli, Ippokratis Pantelidis, Francesco Massafra, Salvatore Lo Sardo, Mohammadreza Salehi
arXiv:2608. 05948v1 Announce Type: new Abstract: Physics engines facilitate large-scale training and evaluation for embodied intelligence, while generative video world models are emerging as implicit simulators of future states and interactions.
By Shuai Wang, Yaxin Feng, Xuekun Jiang, Shihan Tian, Ningyu Yan, Xing Shen, Chaoyang Lyu, Hui Wang, Yunsong Zhou, Hanqing Wang, Jiangmiao Pang, Yang Xiang, Xing Gao, Chunhua Shen, Weinan Zhang
PhysicsLENS is a new dataset and benchmark designed to evaluate how well video generation models capture physical properties relevant to robotics. It consists of matched scenario pairs that share the same conditioning frame and task but differ in underlying physics, and it covers seven physical domains such as collision, gravity, and friction. The benchmark includes over 400 human-annotated labels from four video generation models, revealing that many plausible-looking videos still ignore the specified physical property.
By Isaiah Milkey, Som Sagar, Aditya Taparia, Xinyuan Liu, Jiqing Wen, Ransalu Senanayake
arXiv:2610.03154v1 Announce Type: cross
Abstract: Video generation models produce strikingly realistic sequences and are increasingly proposed as world models, yet recent benchmarks reveal pronounced...
By Jonas Kneifl, Jakub Skalski, Bart{\l}omiej Twardowski, Kamil Deja
arXiv:2606. 09646v1 Announce Type: cross Abstract: We study whether pretrained video foundation models encode intuitive-physics information in their frozen representations, and how this information varies across model families, layers, and probe types.
By Samuele Punzo, Niccol\`o Caselli, Ippokratis Pantelidis, Francesco Massafra, Salvatore Lo Sardo, Mohammadreza Salehi
Physics engines facilitate large-scale training and evaluation for embodied intelligence, while generative video world models are emerging as implicit simulators of future states and interactions. However, existing evaluations of physical fidelity are often conducted in isolation and rely heavily on perceptual similarity or human judgments, providing limited insight into which physical principles or parameters are violated.
PhysiCalWorld introduces a method for calibrating action‑conditioned world models to instance‑specific physics by encoding an exploratory interaction clip into a compact physics descriptor token. This token grounds future predictions in the actual physical behavior of a scene without requiring fine‑tuning or explicit physics labels, and can be reused across rollouts with minimal compute overhead. Experiments across four diverse environments demonstrate significant improvements in rollout prediction accuracy and visual planning success, while the learned descriptors form a physically structured latent space that generalizes beyond the training physics ranges.
By Yunhao Luo, Wentao He, Yinpei Dai, Hyelin Nam, Hongze Fu, Joyce Chai, Nima Fazeli, Jeong Joon Park
JEPA-x is a cross‑predictive physics grounding method that aligns visual latent dynamics with privileged physical trajectories. By treating visual observations and physical states as two views of the same action‑conditioned trajectory and sharing a predictor, it forces the model to learn a common transition rule for both modalities. The physical branch is only used during training, so deployment incurs no extra cost, and the approach significantly reduces rollout drift and boosts control success across a multi‑task suite.
By Kehan Wen, Ziming Li, Siyuan Luo, Fan Shi