Training data generation for dynamic objects using high definition map data

A processor comprising:

• processing circuitry to cause performance of operations, the operations comprising:
• identifying one or more instances of point clusters comprising a spatial cluster of one or more points of a frame set generated during navigation of an environment, wherein at least one individual frame set of the one or more frame sets comprises a point cloud representation of one or more points in three-dimensional (3D) space;
• classifying an individual instance of the one or more instances as dynamic or static based at least on applying a deep learning algorithm to the individual instance;
• auditing the classifying of the individual instance of point clusters based at least on comparing the individual instance to the one or more points of an occupancy map (OMap) representation of the environment, the OMap representation comprising one or more points corresponding to static objects in the environment;
• and
• updating one or more parameters of the deep learning algorithm based at least on using, as ground truth data, the individual instance and the audited classifying associated therewith.

A system comprising:

• one or more processors comprising processing circuitry configured to cause performance of operations, the operations comprising:
• identifying one or more instances of point clusters comprising a spatial cluster of one or more points of a frame set generated during navigation of an environment, wherein at least one individual frame set of the one or more frame sets comprises a point cloud representation of one or more points in three-dimensional (3D) space;
• classifying an individual instance of the one or more instances as dynamic or static based at least on applying a deep learning algorithm to the individual instance;
• auditing the classifying of the individual instance of point clusters based at least on comparing the individual instance to the one or more points of an occupancy map (OMap) representation of the environment, the OMap representation comprising one or more points corresponding to static objects in the environment;
• generating a bounding shape corresponding to the at least instance based at least on the auditing;
• and
• updating one or more parameters of the deep learning algorithm based at least on using the bounding shape as ground truth data.