Training a machine-learning model to predict location using wheel motion data

Abstract

A shopping cart's tracking system receives wheel motion data from a plurality of wheel sensors coupled to a plurality of wheels of the shopping cart, wherein the wheel motion data describes rotation of the plurality of wheels and orientation of the plurality of wheels. The tracking system predicts an estimated location of the shopping cart by applying a machine-learning location model to the wheel motion data. The machine-learning location model is trained with training examples that are generated by: receiving prior wheel motion data from the plurality of wheel sensors, partitioning the prior wheel motion data into a plurality of segments using a time window, receiving one or more baseline locations at one or more prior timestamps, and generating one or more training examples, each training example comprising a segment of prior wheel motion data and a baseline location with a timestamp overlapping the segment.