Embodied AI Glossary中文

PartNet-Mobility

PartNet-Mobility 数据集Advanced

A library of articulated 3D object models with joint-motion annotations, ready to load directly into a simulator.

PartNet-Mobility is an articulated-object dataset released by Hao Su's group at UC San Diego alongside the SAPIEN simulation environment (CVPR 2020). Articulated objects are things like cabinet doors, drawers, and laptops, made of parts connected by joints that can move relative to each other. The dataset contains 46 categories, 2,346 objects, and 14,068 moveable parts; each joint is annotated with its type (revolute, prismatic, or screw), its range of motion, and its parent-child relationship, and comes with textures and a URDF file that loads directly into a physics simulator. Earlier part datasets were mostly static geometry with no way to actually “pull open a drawer.” It's the most commonly used asset source for research on articulated-object manipulation, and benchmarks like ManiSkill build their door-opening and drawer-opening tasks on top of it.

ExampleLoading a PartNet-Mobility cabinet's URDF file into SAPIEN or ManiSkill, a robot arm is trained to grasp the handle and pull the drawer open to a target distance.

Also called
PartNet-Mobility Dataset
Related
SAPIEN (SimulAted Part-based Interactive ENvironment) · Articulated Object · Articulated Object Manipulation · Unified Robot Description Format · ManiSkill · Articulation Estimation
Sources
SAPIEN: A SimulAted Part-based Interactive ENvironment (arXiv 2003.08515)
SAPIEN 论文 HTML 全文(PartNet-Mobility 统计) (Chinese)
As of
2020-03

See it in the full glossary →