SURFACE OBJECT DETECTION RADAR FOR AUTONOMOUS AND SEMI-AUTOMATIC SYSTEMS AND APPLICATIONS

Abstract

In various examples, objects in surface areas can be classified based on their radar signatures. Ground truth data can be generated using cross-modal sensor fusion to identify objects detected by radar data but not by corresponding LiDAR or imagery data. One or more classes can be used to represent these objects on the surface, and corresponding ground truth annotations can be generated and used to train a machine learning model to explicitly recognize objects on the surface based on their radar signatures, thus safely ignoring detected objects on the surface (e.g., manhole covers, railroad tracks, drain grates, etc.).Accordingly, the machine learning model can be deployed and used to detect and ignore objects in the surface, to mark them as a traversable space, and / or to transmit a representation of these detections to one or more downstream components, such as the control stack of an ego machine, in order to enable safe and convenient planning and control of the ego machine.

Images