SoftPhy: Soft-Body Physical Concept Learning and Reasoning from Videos
Primary Area: neurosymbolic & hybrid AI systems (physics-informed, logic & formal reasoning, etc.)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: Neuro-symbolic Visual Reasoning, Physical Reasoning
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Abstract: We introduce the Soft-Body Physical Dataset (SOPHY), a novel benchmark for evaluating machine models in physical reasoning across diverse scenarios for soft bodies. The SOPHY is specifically designed to be complementary with existing physical reasoning benchmarks by encompassing diverse physical property inferences for soft bodies like physical parameters such as mass and density across dynamic situations and predicting corresponding dynamics. This comprehensive dataset enables the development and assessment of AI models with human-like visual reasoning abilities in understanding both rigid objects and soft objects’ visual attributes, physical properties, and dynamics while devising goal-oriented solutions. We evaluated a range of AI models and found that they still struggle to achieve satisfactory performance, which shows that current AI models still lack physical commonsense for soft objects and illustrates the value of the proposed dataset. We hope the SOPHY fosters advancements in AI perception and reasoning in diverse physical environments, bridging the gap between human and machine intelligence in the physical world.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 6857
Loading