Back to results
US 12,586,026 B1Granted
Delivery arrival time prediction system
AssigneeAmazon Technologies, Inc.
InventorsRui Yan, Hansu Gu, Anupama Umesh Kochrekar, Sridevi Venkatramanan, Feng Zhu, Tian Chen, Tao Hu
Life of this patent
Estimating the term…
Abstract
A system for estimating a delivery is disclosed. The system generates delivery time predictions, and generates a confidence score for each time prediction. The system also generates a publication determination for each particular time prediction based on the confidence score associated with each particular time prediction. The publication determination indicates whether to publish the particular time prediction or to not publish the particular time prediction.
Claims19
A computer-implemented method, comprising:
- training, by a computing system, one or more knowledge artificial intelligence (AI) engines of a knowledge embedding layer of a predictor of an object delivery time estimation system, the predictor being configured to generate object delivery time predictions;
- training, by the computing system, one or more route artificial intelligence (AI) engines of a route encoder of the predictor of the object delivery time estimation system;
- training, by the computing system, one or more evaluator artificial intelligence (AI) engines of an evaluator of the object delivery time estimation system, the evaluator being configured to generate a confidence score for each object delivery time prediction;
- generating a route encoding by:
- providing an input vector to a linear transform matrix system, wherein the input vector includes an address of a finished stop and addresses for each uncompleted stop on a delivery route, wherein the linear transform matrix system linearly transforms the input vector;
- receiving a key, a query, and a value based at least on an output of the linear transform matrix system;
- providing the key, the query, and the value to a multi-headed attention algorithm;
- receiving, from the multi-headed attention algorithm, vectors of self-attention;
- and
- generating, based at least in part on the vectors of self-attention, a route encoding;
- receiving, by the computing system, information that identifies a finished stop on a delivery route;
- generating, by the predictor, a plurality of object delivery time predictions for each uncompleted stop on the delivery route, wherein the plurality of object delivery time predictions are generated based at least in part on the route encoding generated using the multi-headed attention algorithm;
- generating, by the evaluator, a confidence score for each object delivery time prediction of the plurality of object delivery time predictions for each uncompleted stop on the delivery route;
- detecting, by the computing system, a position of a delivery vehicle based at least in part on the information;
- adjusting, by the computing system, the delivery route based at least in part on the position of the delivery vehicle and the route encoding;
- causing, by the computing system, execution of a delivery to one or more uncompleted stops on the delivery route based at least in part on the adjusted delivery route;
- generating, by a confidence determination system, a publication determination for each particular object delivery time prediction based on the confidence score associated with each particular object delivery time prediction, the publication determination indicating whether to publish the particular object delivery time prediction or to not publish the particular object delivery time prediction;
- publishing, by the computing system, each particular object delivery time predictions on the delivery route having confidence scores indicating confidence greater than a confidence threshold to corresponding devices;
- and
- not publishing each particular object delivery time predictions on the delivery route having confidence scores indicating confidence less than the confidence threshold to corresponding devices.
17 more elements
A computer system, comprising:
- one or more processors;
- and
- a memory storing executable instructions that, upon execution by the one or more processors, cause the computer system to execute operations to:
- generate a route encoding by:
- providing an input vector to a linear transform matrix system,
- wherein the input vector includes an address of a finished stop and addresses for each uncompleted stop on a delivery route, wherein the linear transform matrix system linearly transforms the input vector;
- receiving a key, a query, and a value based at least on an output of the linear transform matrix system;
- providing the key, the query, and the value to a multi-headed attention algorithm;
- receiving, from the multi-headed attention algorithm, vectors of self-attention;
- and
- generating, based at least in part on the vectors of self-attention, a route encoding;
- receive information that identifies the finished stop on the delivery route;
- generate, by a predictor of an object delivery time estimation system, a plurality of object delivery time predictions for each uncompleted stop on the delivery route, the predictor comprising one or more trained predictor artificial intelligence (AI) systems collectively configured to generate object delivery time predictions, wherein the plurality of object delivery time predictions are generated based at least in part on the route encoding generated using the multi-headed attention algorithm;
- generate, by an evaluator of the object delivery time estimation system, a confidence score for each object delivery time prediction of the plurality of object delivery time predictions for each uncompleted stop on the delivery route, the evaluator comprising one or more trained evaluator artificial intelligence (AI) systems collectively configured to generate the confidence scores;
- detect, by the computer system, a position of a delivery vehicle based at least in part on the information;
- adjust, by the computer system, the delivery route based at least in part on the position of the delivery vehicle and the route encoding;
- cause, by the computer system, execution of a delivery to one or more uncompleted stops on the delivery route based at least in part on the adjusted delivery route;
- generate, by a confidence determination system of the object delivery time estimation system, a publication determination for each particular object delivery time prediction based on the confidence score associated with each particular object delivery time prediction, the publication determination indicating whether to publish the particular object delivery time prediction or to not publish the particular object delivery time prediction;
- publish, by a publication system, each particular object delivery time predictions on the delivery route having confidence scores indicating confidence greater than a confidence threshold to corresponding devices;
- and
- not publish each particular object delivery time predictions on the delivery route having confidence scores indicating confidence less than the confidence threshold to corresponding devices.
18 more elements
A computer-readable storage medium storing computer-executable instructions that, when executed by one or more processors of a computing device, cause the computing device to:
- generate a route encoding by:
- providing an input vector to a linear transform matrix system, wherein the input vector includes an address of a finished stop and addresses for each uncompleted stop on a delivery route, wherein the linear transform matrix system linearly transforms the input vector;
- receiving a key, a query, and a value based at least on an output of the linear transform matrix system;
- providing the key, the query, and the value to a multi-headed attention algorithm;
- receiving, from the multi-headed attention algorithm, vectors of self-attention;
- and
- generating, based at least in part on the vectors of self-attention, a route encoding;
- receive information that identifies the finished stop on the delivery route;
- generate, by a predictor of an object delivery time estimation system, a plurality of object delivery time predictions for each uncompleted stop on the delivery route, wherein the plurality of object delivery time predictions are generated based at least in part on the route encoding generated using the multi-headed attention algorithm;
- generate, by an evaluator of the object delivery time estimation system, a confidence score for each object delivery time prediction of the plurality of object delivery time predictions for each uncompleted stop on the delivery route;
- detect, by the object delivery time estimation system, a position of a delivery vehicle based at least in part on the information;
- adjust, by the object delivery time estimation system, the delivery route based at least in part on the position of the delivery vehicle and the route encoding;
- cause, by the object delivery time estimation system, execution of a delivery to one or more uncompleted stops on the delivery route based at least in part on the adjusted delivery route;
- generate, by a confidence determination system of the object delivery time estimation system, a publication determination for each particular object delivery time prediction based on the confidence score associated with each particular object delivery time prediction, the publication determination indicating whether to publish the particular object delivery time prediction or to not publish the particular object delivery time prediction;
- publish, by a publication system, each particular object delivery time predictions on the delivery route having confidence scores indicating confidence greater than a confidence threshold to corresponding devices;
- and
- not publish each particular object delivery time predictions on the delivery route having confidence scores indicating confidence less than the confidence threshold to corresponding devices.
14 more elements
Dependencies are inferred from the claim text.