Horizon Robotics (China)

2 related articles

Subscribe to updates

arXiv Oct 12, 2026
Same object, new material prompt: PhysForge fits physics on contact and deformation windows
PhysForge treats material-conditioned dynamics from a video diffusion model as references for fitting the physical parameters of 3D assets, so a user-specified material actually changes how an object contacts and deforms. Previously such parameters were largely inferred from appearance, making it hard to assign a material to the same object or to estimate properties from complex interactions between objects and semantic parts. On PhysAssets and Google Scanned Objects it reaches the best CLIP and physical controllability scores for both common and uncommon materials, with average physical controllability 4.3 versus 3.8 for the strongest baseline.
PhysForge treats material-conditioned dynamics from a video diffusion model as references for fitting the physical parameters of 3D assets, so a user-specified material actually changes how an object contacts and deforms. Previously such parameters were largely inferred from appearance, making it hard to assign a material to the same object or to estimate properties from complex interactions between objects and semantic parts. On PhysAssets and Google Scanned Objects it reaches the best CLIP and physical controllability scores for both common and uncommon materials, with average physical controllability 4.3 versus 3.8 for the strongest baseline.
Nanjing University Shenzhen University Horizon Robotics (China)
Yuchen Wang（First author）
PhysForge PhysFit PhysComp Physical-Property Estimation Material Point Method
arXiv Oct 2, 2026
Multimodal Flow unifies language and vision as continuous hyperchunks under one flow-matching backbone, with MF-1 scoring 0.821 on GenEval and 83.44 on DPG-Bench at 150B tokens
The work introduces Multimodal Flow, which encodes text blocks and images as ordered continuous hyperchunks and models both modalities in their own embedding spaces with a shared chunk-causal Flow Matching backbone, instantiated as MF-1: continued pretraining improves multimodal modeling across 0.6B, 1.2B, and 1.6B scales, and the 1.6B model reaches 0.821 on GenEval and 83.44 on DPG-Bench with only 150B pretraining tokens and an average of 75.3 across VQAv2, MMBench, and POPE, outperforming representative hybrid and discrete models under matched data, optimization, and parameter budgets.
The work introduces Multimodal Flow, which encodes text blocks and images as ordered continuous hyperchunks and models both modalities in their own embedding spaces with a shared chunk-causal Flow Matching backbone, instantiated as MF-1: continued pretraining improves multimodal modeling across 0.6B, 1.2B, and 1.6B scales, and the 1.6B model reaches 0.821 on GenEval and 83.44 on DPG-Bench with only 150B pretraining tokens and an average of 75.3 across VQAv2, MMBench, and POPE, outperforming representative hybrid and discrete models under matched data, optimization, and parameter budgets.
Huazhong University of Science and Technology Beijing Jiaotong University Horizon Robotics (China)
Hongyuan Tao（First author）
Multimodal Flow MF-1 Flow Matching Text-to-Image Generation Visual Question Answering

You’re all caught up

Same object, new material prompt: PhysForge fits physics on contact and deformation windows

Multimodal Flow unifies language and vision as continuous hyperchunks under one flow-matching backbone, with MF-1 scoring 0.821 on GenEval and 83.44 on DPG-Bench at 150B tokens