Evaluating Newtonian Mechanics in Video Generative Models with Real Physical Systems

Antonios Tragoudaras, Chenyu Zhang, Daniil Cherniavskii, Antonios Vozikis, Thijmen Nijdam, Derck W. E. Prinzhorn, Mark Bodracska, Nicu Sebe, Andrii Zadaianchuk, Stratis Gavves
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:121912-121951, 2026.

Abstract

Recent advances in image and video generation raise hopes that these models possess world modeling capabilities—the ability to generate realistic, physically plausible videos. This could revolutionize applications in robotics, autonomous driving, and scientific simulation. However, before treating these models as world models, we must ask: Do they adhere to physical laws? Current evaluation methods rely on subjective judgments or trajectory matching, limiting their usage for physical reasoning estimation, where many generations could be physically plausible. Thus, we introduce Morpheus, one of the first physics-informed evaluation frameworks for measuring the ability of video generation models to comprehend Newtonian dynamics. Morpheus features 130 real-world videos capturing physical phenomena, guided by conservation laws. Using those as conditioning for video generation, we assess physical plausibility leveraging interpretable metrics evaluated with respect to infallible conservation laws known per physical setting, leveraging advances in physics-informed neural networks and vision-language foundation models. Importantly, Morpheus targets controlled Newtonian rigid-body settings to enable quantitative checks. Our findings reveal that even with advanced prompting and video conditioning, contemporary models struggle to encode physical principles despite generating aesthetically pleasing videos. Code and data available here.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-tragoudaras26a, title = {Evaluating Newtonian Mechanics in Video Generative Models with Real Physical Systems}, author = {Tragoudaras, Antonios and Zhang, Chenyu and Cherniavskii, Daniil and Vozikis, Antonios and Nijdam, Thijmen and Prinzhorn, Derck W. E. and Bodracska, Mark and Sebe, Nicu and Zadaianchuk, Andrii and Gavves, Stratis}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {121912--121951}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/tragoudaras26a/tragoudaras26a.pdf}, url = {https://proceedings.mlr.press/v306/tragoudaras26a.html}, abstract = {Recent advances in image and video generation raise hopes that these models possess world modeling capabilities—the ability to generate realistic, physically plausible videos. This could revolutionize applications in robotics, autonomous driving, and scientific simulation. However, before treating these models as world models, we must ask: Do they adhere to physical laws? Current evaluation methods rely on subjective judgments or trajectory matching, limiting their usage for physical reasoning estimation, where many generations could be physically plausible. Thus, we introduce Morpheus, one of the first physics-informed evaluation frameworks for measuring the ability of video generation models to comprehend Newtonian dynamics. Morpheus features 130 real-world videos capturing physical phenomena, guided by conservation laws. Using those as conditioning for video generation, we assess physical plausibility leveraging interpretable metrics evaluated with respect to infallible conservation laws known per physical setting, leveraging advances in physics-informed neural networks and vision-language foundation models. Importantly, Morpheus targets controlled Newtonian rigid-body settings to enable quantitative checks. Our findings reveal that even with advanced prompting and video conditioning, contemporary models struggle to encode physical principles despite generating aesthetically pleasing videos. Code and data available here.} }
Endnote
%0 Conference Paper %T Evaluating Newtonian Mechanics in Video Generative Models with Real Physical Systems %A Antonios Tragoudaras %A Chenyu Zhang %A Daniil Cherniavskii %A Antonios Vozikis %A Thijmen Nijdam %A Derck W. E. Prinzhorn %A Mark Bodracska %A Nicu Sebe %A Andrii Zadaianchuk %A Stratis Gavves %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-tragoudaras26a %I PMLR %P 121912--121951 %U https://proceedings.mlr.press/v306/tragoudaras26a.html %V 306 %X Recent advances in image and video generation raise hopes that these models possess world modeling capabilities—the ability to generate realistic, physically plausible videos. This could revolutionize applications in robotics, autonomous driving, and scientific simulation. However, before treating these models as world models, we must ask: Do they adhere to physical laws? Current evaluation methods rely on subjective judgments or trajectory matching, limiting their usage for physical reasoning estimation, where many generations could be physically plausible. Thus, we introduce Morpheus, one of the first physics-informed evaluation frameworks for measuring the ability of video generation models to comprehend Newtonian dynamics. Morpheus features 130 real-world videos capturing physical phenomena, guided by conservation laws. Using those as conditioning for video generation, we assess physical plausibility leveraging interpretable metrics evaluated with respect to infallible conservation laws known per physical setting, leveraging advances in physics-informed neural networks and vision-language foundation models. Importantly, Morpheus targets controlled Newtonian rigid-body settings to enable quantitative checks. Our findings reveal that even with advanced prompting and video conditioning, contemporary models struggle to encode physical principles despite generating aesthetically pleasing videos. Code and data available here.
APA
Tragoudaras, A., Zhang, C., Cherniavskii, D., Vozikis, A., Nijdam, T., Prinzhorn, D.W.E., Bodracska, M., Sebe, N., Zadaianchuk, A. & Gavves, S.. (2026). Evaluating Newtonian Mechanics in Video Generative Models with Real Physical Systems. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:121912-121951 Available from https://proceedings.mlr.press/v306/tragoudaras26a.html.

Related Material