AssemblyBench Dataset and AssemblyDyno Model for Industrial Assembly
AssemblyBench comprises a synthetic collection of 2,789 industrial items, featuring multimodal instruction guides, 3D component models, and assembly paths. The related AssemblyDyno model leverages these instructional materials and 3D representations to forecast the sequence of assembly and the corresponding trajectories. In terms of pose estimation and trajectory viability, AssemblyDyno surpasses previous research, as confirmed through physics-based simulations.
Key facts
- AssemblyBench contains 2,789 industrial objects.
- The dataset includes multimodal instruction manuals, 3D part models, and assembly trajectories.
- AssemblyDyno is a transformer-based model.
- AssemblyDyno predicts assembly order and part assembly trajectories.
- AssemblyDyno outperforms prior works in assembly pose estimation.
- Trajectory feasibility is evaluated by physics-based simulations.
- Existing datasets overlook shape complexities and assembly trajectories in industrial assemblies.
- The work is from the field of Computer Vision and Pattern Recognition.
Entities
Institutions
- arXiv