Abstract: A method includes providing, by server computer to a service provider, an onboarding interface for onboarding a service provider to accept returns initiated by end users and from a plurality of available transporters, and then receiving, by the server computer, selections from the service provider regarding different options for processing returns via the onboarding interface. The server computer can receive a plurality of return orders for a plurality of items from a plurality of end users. and process the return orders according to the stored and selected options.

Abstract: The present disclosure generally pertains to systems and methods of order fulfillment using a fulfillment engine, which optimizes the allocation of client or user orders to third party merchants for fulfillment through the use of linear programming or machine learning. This optimization occurs by loading a series of relevant merchant attributes and client attributes into an optimizer with the order and analyzing the attributes to determine an optimal allocation of client orders to third party merchants. These orders are issued to the merchants and where necessary the system arranges delivery of the orders to the client. The system may also automatically update future orders in response to client feedback.

Abstract: Provided are various systems and processes for improving last-mile delivery of real-time, on-demand orders for perishable goods. In one aspect, a method for operating an autonomous vehicle comprises receiving real-time perception data at a planner module located on the autonomous vehicle, and generating an initial motion plan based on the real-time perception data. The autonomous vehicle is maneuvered along a constrained route based on the initial motion plan without user input. An alert condition is detected from the real-time perception data, and a notification is displayed at an operator device. The notification includes a request for user input to adjust the initial motion plan. User input is received at the planner module and a modified motion plan is generated by adjusting the initial motion plan based on the user input. A virtual representation of objects identified by the AV is locally rendered for display at the operator device.

Abstract: A method includes a computer generating a multimodal embedding using a first machine learning model. The computer obtains a labeled training query. The computer trains a second machine learning model using the labeled training query and the multimodal embedding to predict items related to multimodal embeddings based on queries.

Abstract: A computer receives a query from an end user device. The computer determines a segmentation result based on the query using a first language model. The computer determines a query embedding based on the query. The computer determines a plurality of candidate labels based on the query embedding. The computer determines a linking result based on the query and the plurality of candidate labels using a second language model. The computer retrieves one or more documents from a documents database using the linking result and the segmentation result. The computer provides the one or more documents to the end user device.

Abstract: Method includes inputting log data to perception, including sensor data of objects around AV and information about events occurring with respect to AV and obstacle, information including obstacle parameters, each having actual value detected by sensors; detecting, by perception from sensor data, coordinate region within which obstacle appears; outputting, by perception, data including values for obstacle parameters, each associated with actual or synthetic value; inputting values into planner, to generate values for parameters of AV; during re-run of perception, detecting, by perception, coordinate region in log data and detecting obstacle, as object overlapping coordinate region; outputting, by perception, updated data including updated values for obstacle parameters, each associated with actual value or updated synthetic value; inputting updated values into planner, to generate updated values for parameters of AV; and generating validation result based on values for parameters for AV and updated values for

Abstract: Various embodiments can reduce variance in a target metric (e.g., experiment outcome). Embodiments can use historical pre-experiment outcomes to predict a metric (forecasts) that is expected for a future measurement time. The forecasts can then be used as a covariate to reduce the variance of the target metric. When predicting forecasts, various embodiments can use an automation pipeline that can generate better and quicker forecasts. When there are multiple covariates that may be considered to reduce variance in the target metric, various embodiments can use a closed from solution for determining optimal coefficient of each covariate.

Abstract: A method is disclosed. The method includes receiving, by a server computer, a plurality of images associated with one or more service providers. The server computer then receives an inquiry request, and determines an image of the plurality of images. The image is selected in response to a composite score based on a scoring algorithm scoring each image in the plurality of images. The scoring algorithm comprises a conversion component and an uncertainty component. The server computer provides an inquiry response comprising the image to the end user device operated by the end user.

Abstract: A method includes a server computer receiving data relating to a plurality of delivery orders from service providers to end users, and data relating to a plurality of delivery orders from end users to service providers. The server computer determines a plurality of routes corresponding to the plurality of delivery orders. The server computer can then determine a set of optimal route plans by combining delivery order routes. The server computer can then receive acceptances from a plurality of transporters that will execute the set of optimal route plans. The server computer can then facilitate execution of the optimal route plans.

Abstract: A method is disclosed. The method includes receiving, by a server computer from a client computer, a query relating to stored records stored in a database, the query being from a user. The method also includes determining, by the server computer and using a large language model, a query structure from the query, and determining, by the server computer, features useful to answering the query. The method also includes generating, by the server computer, a database query using at least the query structure and the features, retrieving, by the server computer, data from the database using the database query, and outputting, the server computer, an answer to the query to the client computer.

Abstract: Method includes inputting log data to perception, including sensor data of objects around AV and information about events occurring with respect to AV and obstacle, information including obstacle parameters, each having actual value detected by sensors; detecting, by perception from sensor data, coordinate region within which obstacle appears; outputting, by perception, data including values for obstacle parameters, each associated with actual or synthetic value; inputting values into planner, to generate values for parameters of AV; during re-run of perception, detecting, by perception, coordinate region in log data and detecting obstacle, as object overlapping coordinate region; outputting, by perception, updated data including updated values for obstacle parameters, each associated with actual value or updated synthetic value; inputting updated values into planner, to generate updated values for parameters of AV; and generating validation result based on values for parameters for AV and updated values for

Abstract: A method includes providing, by server computer to a service provider, an onboarding interface for onboarding a service provider to accept returns initiated by end users and from a plurality of available transporters, and then receiving, by the server computer, selections from the service provider regarding different options for processing returns via the onboarding interface. The server computer can receive a plurality of return orders for a plurality of items from a plurality of end users. and process the return orders according to the stored and selected options.

Abstract: A method includes retrieving, by the user device from a database on the user device, historical data, and real-time client device data. The method includes determining an agitation score based on the real-time client device data and the historical data, and then performing an action based on the agitation score.

Abstract: A method includes a server computer receiving delivery data associated with a plurality of deliveries in a geographical area. The server computer can determine a plurality of adverse delivery zones in the geographical area. The server computer can receive a fulfillment request to deliver an item to an end user at a specific location and determine if the specific location is in one of the adverse delivery zones. If the specific location is in one of the adverse delivery zones, the server computer can take one or more additional actions to ensure that the item is delivered to the end user at the specific location.

Abstract: Methods, systems, and machine learning models for providing accurate estimate time of arrival (ETA) predictions are disclosed, particularly in the context of item fulfillment services. Input features including continuous, numerical, categorical, and time series features can be processed using an initial set of encoders. The resulting embeddings (and other applicable data) can be applied to a set of “expert” encoders. The embeddings produced by the expert encoders can be combined and processed using a multilayer perceptron, which can return one or more estimated arrival time predictions. Such predictions can correspond to multiple tasks and can include both point estimate predictions and distribution estimate predictions, e.g., predictions describing a probability density function of estimated arrival times.

Abstract: The method includes determining, by a central server computer using a machine learning model, aggregates of items that are likely to be ordered by end users. The method also includes initiating placement of the items in the aggregates of items in one or more locations based on locations of the end users, receiving a request for an aggregate of items, and initiate the fulfillment of the request for the aggregate of items, where the items in the aggregates of items are at the one or more locations.

Abstract: A method includes a computer receiving a user query. The computer generates a summary of the user query using a first large language model. The computer determines a user issue from a first database based on the summary. The computer determines digital document from a second database based on the user issue. The computer generates a prompt based on the digital document and a prompt template. The computer generates a response based on the prompt using a second large language model.

Abstract: A method is described. The method includes determining, by a central server computer using a machine learning model, aggregates of items that are likely to be ordered by end users. The method also includes initiating placement of the items in the aggregates of items in one or more locations based on locations of the end users, receiving a request for an aggregate of items, and initiate the fulfillment of the request for the aggregate of items, where the items in the aggregates of items are at the one or more locations.

Abstract: A system includes a carousel apparatus, a support structure configured to hold a receptacle, and a linear actuator apparatus configured to move the support structure. The carousel apparatus includes: a holder in which a first plurality of openings and a second plurality of openings are formed, and a first motor coupled to the holder and adapted to rotate the holder around an axis. A linear actuator apparatus includes: a linear guide which extends underneath the holder and is coupled to the support structure, and a second motor coupled to the linear guide and adapted to move the support structure along the linear guide underneath the holder.

Abstract: A method is disclosed. The method includes receiving, by a server computer, a plurality of images associated with one or more service providers. The server computer then receives an inquiry request, and determines an image of the plurality of images. The image is selected in response to a composite score based on a scoring algorithm scoring each image in the plurality of images. The scoring algorithm comprises a conversion component and an uncertainty component. The server computer provides an inquiry response comprising the image to the end user device operated by the end user.