Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
URDF-Anything: Constructing Articulated Objects with 3D Multimodal Language Model
Authors: Zhe Li, Xiang Bai, Jieyu Zhang, Zhuangzhe Wu, Che Xu, Ying Li, Chengkai Hou, Shanghang Zhang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on both simulated and real-world datasets demonstrate that our method significantly outperforms existing approaches regarding geometric segmentation (m Io U 17% improvement), kinematic parameter prediction (average error reduction of 29%), and physical executability (surpassing baselines by 50%). |
| Researcher Affiliation | Academia | Zhe Li1, , Xiang Bai1, , Jieyu Zhang2, Zhuangzhe Wu1, Che Xu1, Ying Li1, Chengkai Hou1, Shanghang Zhang1, 1State Key Laboratory of Multimedia Information Processing, School of Computer Science, Peking University, 2University of Washington |
| Pseudocode | No | The paper only describes the methodology using prose and diagrams (e.g., Figure 2) without explicit pseudocode or algorithm blocks. |
| Open Source Code | Yes | 1Project page: https://lzvsdy.github.io/URDF-Anything/ and We provide supplemental material, including instruction to download and prepare data and how to use our code and derive the result. We also include our code. |
| Open Datasets | Yes | Experimental evaluation on the Part Net-Mobility dataset [12], including both in-distribution and challenging out-of-distribution objects, demonstrates the superior performance of URDF-Anything compared to existing methods. We train and evaluate our framework on the Part Net-Mobility dataset [12], a large collection of 3D articulated objects with URDF annotations. |
| Dataset Splits | Yes | The dataset is partitioned into In-Distribution (ID) and Out-of-Distribution (OOD) subsets based solely on object categories. Since our method utilizes image input, we generated visual data by rendering multi-view and single-view RGB images from the dataset s 3D models within a simulation environment. Ground truth kinematic and geometric information from the original URDF files was processed and reorganized into a compact JSON format matching our model s output structure. The dataset is divided into standard training and testing sets. |
| Hardware Specification | Yes | We adopt one NVIDIA 80G A800 GPU for training. Our model was fine-tuned in 2.5 hours on a single NVIDIA A800 (80GB) GPU. |
| Software Dependencies | No | We employ Shape LLM [22] as our 3D MLLM backbone, with Shape LLM7B-general-v1.0 checkpoint as the default settings. For the 3D backbone, We use Uni3D [45] to extract dense geometric features. We employ Lo RA [46] for efficient fine-tuning... We use Adam W optimizer [47]... (No specific version numbers for software like PyTorch, Python, CUDA, etc. are provided.) |
| Experiment Setup | Yes | We use Adam W optimizer [47] with the learning rate and weight decay set to 0.0003 and 0, respectively. We use the cosine learning rate scheduler, with the warm-up iteration ratio set to 0.03. The batch size per device is set to 2, and the gradient accumulation step is set to 10. |