Bridging the reality gap: A benchmark for physical reasoning in general world models with various physical phenomena beyond mechanics

Pengyu Zhao, Jinan Xu, Ning Cheng, Huiqi Hu, Xue Zhang, Xiuwen Xu, Zijian Jin, Fandong Meng, Jie Zhou, Wenjuan Han

Abstract: Highlights•A physical reasoning benchmark covering four physical phenomena.•The first physical reasoning benchmark constructed from real-world videos.•Experiments show general world models have limited physical reasoning capabilities.•Analyze error cases and propose improvement strategies.