Scene embedding for visual navigation

A computer-implemented method, comprising:

• obtaining first image data representative of a current location of an object in a physical environment;
• obtaining second image data representative of a destination location for the object in the physical environment;
• processing the first image data and the second image data using a trained embedding network, the trained embedding network trained using a set of image triplets, the image triplets each including image data for a representative image, a similar image, and a disparate image representative of respective portions of the physical environment;
• and
• receiving, from the trained embedding network, a navigation vector indicating a relative direction and distance to the destination location from the current location of the object.

A system, comprising:

• at least one processor;
• and
• memory including instructions that, when executed by the at least one processor, cause the system to:
• obtain first image data representative of a current location of an object in a physical environment;
• obtain second image data representative of a destination location for the object in the physical environment;
• process the first image data and the second image data using a trained embedding network, the trained embedding network trained using a set of image triplets, the image triplets each including image data for a representative image, a similar image, and a disparate image representative of respective portions of the physical environment;
• and
• receive, from the trained embedding network, a navigation vector indicating a relative direction and distance to the destination location from the current location of the object.