Only released in EOL distros:  


This is the documentation for the code and data from the paper An Object-Based Semantic World Model for Long-Term Change Detection and Semantic Querying. Reading the paper will help make the rest of this make sense.

The worldmodel stack implements a set of tools for object-based semantic world modeling, including a perceptual pipeline and database system to support semantic queries. The resulting object database is made available using a web interface; see below. The system is specifically tailored to long-term operation with minimal perceptual assumptions; it requires only a Microsoft Kinect (mounted some distance from the floor) and a localized mobile base. Generating and querying your own semantic maps is straightforward!

Web Interface

The system has been run on a large dataset of the interior of Willow Garage; the web interface to the resulting database (and the semantic querying applications built atop it) is available at at


With the demise of Willow Garage, there is no easy host for the dataset (which is roughly 330GB). Please contact Mac Mason (, and I'll figure out the best way to send you the files.

To save space, the (Kinect) data are stored as pairs of images; you'll need the openni_record_player node is the worldmodel package to turn them back into point clouds. The other included data (robot pose, 2d and 3d laser, etc.) are all in their original form, and require no extra processing.


Download the source code from KForge (linked above), or install the ros-electric-worldmodel package. This stack is fairly dependency-heavy, as it includes code for actively pointing a PR2 head for improved data collection. Note that a PR2 is not necessary to use this stack! (You will need to install the dependencies nonetheless.)

Getting Started

For data collection, see semanticmodel. Once you've collected some data, see semantic_model_web_interface. If you have a continuous operation system running, see worldmodel_ops.

Report a Bug


Wiki: worldmodel (last edited 2015-06-12 01:04:44 by MacMason)