Route selection using machine-learned safety model

Abstract

Systems and methods of configuring and using a machine-learned safety risk model to predict a corresponding risk of vehicular collision for different candidate routes are disclosed herein. In some example embodiments, a computer system obtains accident data and feature data for historical routes that have been communicated electronically as navigation guidance, trains a safety risk model using the accident data and the feature data of the historical routes as training data in a machine learning process, and then evaluates one or more routing algorithms by generating a corresponding set of routes for each routing algorithm and generating a corresponding performance measurement for each set of routes using the trained safety risk model.

What is claimed is: 
     
         1 . A computer-implemented method performed by a computer system having at least one hardware processor, the computer-implemented method comprising:
 for each one of a plurality of historical routes that have been communicated electronically as navigation guidance, obtaining corresponding accident data and corresponding feature data, the corresponding accident data indicating whether a vehicular accident occurred in association with the historical route, the feature data being based at least in part on edges and maneuvers that form the historical route;   training a safety risk model using the accident data and the feature data of the plurality of historical routes as training data in a machine learning process;   generating a first set of routes using a first routing algorithm;   generating a second set of routes using a second routing algorithm;   generating a first performance measurement for the first routing algorithm based on the first set of routes using the trained safety risk model;   generating a second performance measurement for the second routing algorithm based on the second set of routes using the trained safety risk model; and   causing an evaluation of the first routing algorithm or the second routing algorithm to be displayed within a user interface on a computing device of a user, the evaluation being based on the first performance measurement and the second performance measurement.   
     
     
         2 . The computer-implemented method of  claim 1 , further comprising:
 receiving a request for a route from a starting geographic location to a destination geographic location;   identifying a route from the starting geographic location to the destination geographic location using the trained safety risk model; and   transmitting the identified route to another computing device.   
     
     
         3 . The computer-implemented method of  claim 2 , wherein the transmitting of the identified route to the computing device comprises causing the identified route to be displayed within a user interface on the computing device. 
     
     
         4 . The computer-implemented method of  claim 2 , wherein the transmitting of the identified route to the computing device comprises transmitting the identified route to an autonomous vehicle for use by the autonomous vehicle in navigating from the starting geographic location to the destination geographic location. 
     
     
         5 . The computer-implemented method of  claim 2 , wherein the identifying of the route comprises:
 generating a plurality of candidate routes from the starting geographic location to the destination geographic location; and   selecting one of the plurality of candidate routes using the trained safety risk model, the selected one of the plurality of candidate routes being the identified route.   
     
     
         6 . The computer-implemented method of  claim 5 , wherein the selecting the one of the plurality of candidate routes comprises:
 generating a corresponding safety risk score for each one of the plurality of candidate routes using the trained safety risk model; and   selecting the one of the plurality of candidate routes based at least in part on the corresponding safety risk score for the selected one of the plurality of candidate routes.   
     
     
         7 . The computer-implemented method of  claim 1 , wherein each one of the plurality of historical routes was communicated electronically to a corresponding client device of a corresponding provider of a transportation service in association with a corresponding request for the transportation service, the corresponding accident data and the corresponding feature data for each one of the plurality of historical routes being stored in a database in association with the corresponding request for the transportation service. 
     
     
         8 . The computer-implemented method of  claim 1 , wherein the feature data comprises one or more edge statistics calculated based on the edges of the historical route. 
     
     
         9 . The computer-implemented method of  claim 8 , wherein the one or more edge statistics comprises one or more of a total number of the edges of the historical route, an edge distance statistic based on distances of the edges of the historical route, an edge duration statistic based on durations associated with the edges of the historical route, an edge speed statistic based on travelling speeds associated with the edges of the historical route, and a road class statistic based on one or more classes of roads associated with the edges of the historical route. 
     
     
         10 . The computer-implemented method of  claim 1 , wherein the feature data comprises one or more maneuver statistics calculated based on the maneuvers of the historical route. 
     
     
         11 . The computer-implemented method of  claim 10 , wherein the one or more maneuver statistics comprises one or more of a total number of the maneuvers of the historical route, a maneuver distance statistic based on distances of the maneuvers of the historical route, a maneuver duration statistic based on durations associated with the maneuver of the historical route, a maneuver speed statistic based on speeds associated with the maneuvers of the historical route, a heading change statistic based on degrees of heading changes associated with the maneuvers of the historical route, and a compound maneuver statistic based on a total number of compound maneuvers of the historical route. 
     
     
         12 . The computer-implemented method of  claim 1 , wherein the safety risk model comprises one or more models selected from a group of models, the group of models consisting of a gradient boosting decision tree model, a deep learning model, and a generalized linear model. 
     
     
         13 . The computer-implemented method of  claim 1 , wherein each one of the plurality of historical routes comprises a corresponding plurality of road segments, the corresponding accident data for each one of the plurality of historical routes comprising corresponding segment accident data for each one of the corresponding plurality of road segments of the historical route, the segment accident data indicating whether a vehicular accident occurred. on the corresponding road segment, the corresponding feature data for each one of the plurality of historical routes comprising segment feature data for each one of the corresponding plurality of road segments of the historical route, the segment feature data being based at least in part on edges or maneuvers corresponding to the road segment. 
     
     
         14 . The computer-implemented method of  claim 13 , further comprising:
 receiving a request for a route from a starting geographic location to a destination geographic location;   identifying a route from the starting geographic location to the destination geographic location using the trained safety risk model; and   transmitting the identified route to another computing device.   
     
     
         15 . The computer-implemented method of  claim 14 , wherein the transmitting the identified route to the computing device comprises causing an indication of an elevated risk associated with at least one road segment of the identified route to be displayed in association with the identified route using the trained safety risk model based on corresponding segment feature data of the identified route, the corresponding segment feature data of the identified route being based at least in part on edges or maneuvers corresponding to the at least one road segment of the identified route. 
     
     
         16 . The computer-implemented method of  claim 14 , wherein the identifying of the route comprises:
 generating a corresponding safety risk score for each one of a plurality of candidate route segments using the trained safety risk model; and   generating the identified route using the corresponding safety risk scores for the plurality of candidate route segments within a routing algorithm to form the identified route, the identified route being formed from at least a portion of the plurality of candidate route segments.   
     
     
         17 . A system comprising:
 at least one hardware processor; and   a machine storage medium embodying a set of instructions that, when executed by the at least one hardware processor, cause the at least one processor to perform operations comprising:
 for each one of a plurality of historical routes that have been communicated electronically as navigation guidance, obtaining corresponding accident data and corresponding feature data, the corresponding accident data indicating whether a vehicular accident occurred in association with the historical route, the feature data being based at least in part on edges and maneuvers that form the historical route; 
 training a safety risk model using the accident data and the feature data of the plurality of historical routes as training data in a machine learning process; 
 generating a first set of routes using a first routing algorithm; 
 generating a second set of routes using a second routing algorithm; 
 generating a first performance measurement for the first routing algorithm based on the first set of routes using the trained safety risk model; 
 generating a second performance measurement for the second routing algorithm based on the second set of routes using the trained safety risk model; and 
 causing an evaluation of the first routing algorithm or the second routing algorithm to be displayed within a user interface on a computing device of a user, the evaluation being based on the first performance measurement and the second performance measurement. 
   
     
     
         18 . The system of  claim 17 , wherein the operations further comprise:
 receiving a request for a route from a starting geographic location to a destination geographic location;   identifying a route from the starting geographic location to the destination geographic location using the trained safety risk model; and   transmitting the identified route to another computing device.   
     
     
         19 . The system of  claim 18 , wherein the transmitting of the identified route to the computing device comprises causing the identified route to be displayed within a user interface on the computing device. 
     
     
         20 . A machine storage medium embodying a set of instructions that, when executed by at least one hardware processor, cause the processor to perform operations comprising:
 for each one of a plurality of historical routes that have been communicated electronically as navigation guidance, obtaining corresponding accident data and corresponding feature data, the corresponding accident data indicating whether a vehicular accident occurred in association with the historical route, the feature data being based at least in part on edges and maneuvers that form the historical route;   training a safety risk model using the accident data and the feature data of the plurality of historical routes as training data in a machine learning process;   generating a first set of routes using a first routing algorithm;   generating a second set of routes using a second routing algorithm;   generating a first performance measurement for the first routing algorithm based on the first set of routes using the trained safety risk model;   generating a second performance measurement for the second routing algorithm based on the second set of routes using the trained safety risk model; and   causing an evaluation of the first routing algorithm or the second routing algorithm to be displayed within a user interface on a computing device of a user, the evaluation being based on the first performance measurement and the second performance measurement.