A Map for Robots
Using information gathered from drivers across the United States, we are creating a map that goes beyond even physical data.
Using machine learning, our researchers are crearting an encyclopedia of driving behavior in various settings and environments, all annotated alongside point cloud information.
Papers: Drone Flying Area Estimation Method based on Deep Learning
A Platform for HD Maps
and automatic generation of HD maps from video
- Point cloud data is extractied from video via deep learning the resulting information forming the first layer of our maps.
- Annotations such as lanes, signs, and signals are extracted, forming the second layer.
- Batcheds of point cloud and annotation data are fused together into a single cohesive output.
- The now fused map information is georeferenced, aligning point cloud and annotation information to their respective coordinates.
- Vector maps are produced in the OpenDrive format, allowing for autonomous-driving simulations and research.
- Human behavioral data while driving is extracted and annotated using machine learning, producing the third layer of our maps.