Generalized risk routing for human drivers

1. A transportation management system comprising:

one or more processors; and

one or more memory resources storing:

a machine-learned risk regressor, wherein the machine-learned risk regressor is trained using log data from vehicles operated throughout a given region to determine fractional risk values for path segments of the given region, the fractional risk values being specific to at least one of an autonomous vehicle (AV) type or AV software version and indicative of a probability that a harmful event will occur for traversal of the path segments; and

instructions that, when executed by the one or more processors, cause the transportation management system to:

receive live data from a plurality of vehicles in the given region;

determine a fractional risk value for traversing each path segment of the given region, wherein the fractional risk value is determined using the machine-learned risk regressor using the live data as input; and

route vehicles operating within the given region along lowest risk route options based on the fractional risk value for each path segment.

2. The transportation management system of claim 1 , wherein the executed instructions cause the transportation management system to:

for each determination of a fractional risk value for a particular path segment, temporally correlate the fractional risk value to a current set of conditions based on sensor data from the vehicles operating within the given region, wherein the fractional risk value is temporally correlated using the machine-learned risk regressor.

3. The transportation management system of claim 2 , wherein the current set of conditions comprises at least one of weather conditions or road conditions.

4. The transportation management system of claim 2 , wherein the executed instructions cause the transportation management system to:

receive a plurality of transport requests, wherein each of the plurality of transport requests indicates a pick-up location and a drop-off location; and, for each of the plurality of transport requests:

determine a plurality of route options between the pick-up location and the drop-off location associated with the respective transport request; and

determine an aggregate risk value for each route option of the plurality of route options based at least in part on the current set of conditions and the correlated fractional risk value of each path segment along the respective route option.

5. The transportation management system of claim 4 , wherein the executed instructions cause the transportation management system to further determine the aggregate risk value for each route option of the plurality of route options based on predicted conditions at a time a vehicle selected to traverse the route option will arrive at a particular path segment of the respective route option.

6. The transportation management system of claim 4 , wherein the executed instructions cause the transportation management system to determine the lowest risk route option from the plurality of route options based on the aggregate risk value for each route option of the plurality of route options.

7. The transportation management system of claim 2 , wherein the log data comprises sensor data received from driver devices, the sensor data comprising at least one of inertial measurement data and image data.

8. The transportation management system of claim 2 , wherein the log data is autonomous vehicle log data.

9. The transportation management system of claim 8 , wherein the log data comprises at least one of LIDAR, RADAR, SONAR, or image data.

10. The transportation management system of claim 8 , wherein the log data comprises input data corresponding to one or more autonomous vehicle control inputs for at least one control mechanism of an autonomous vehicle.

11. The transportation management system of claim 2 , wherein the sensor data is sensor data received from at least one driver device.

12. The transportation management system of claim 1 , wherein the executed instructions cause the transportation management system to:

manage an on-demand transportation service by receiving transport requests from requesting users and matching the requesting users with servicing drivers.

13. The transportation management system of claim 12 , wherein the executed instructions cause the transportation management system to manage the on-demand transportation service by matching requesting users with safety-driven autonomous vehicles and fully autonomous vehicles.

14. The transportation management system of claim 13 , wherein each transport request indicates a pick-up location and a drop-off location, and wherein the executed instructions further cause the transportation management system to,

for a particular transport request from a particular requesting user:

determine an aggregate risk quantity for transporting the particular requesting user from-a particular pick-up location to-a particular drop-off location; and

filter a candidate set of vehicles for servicing the transport request, wherein the candidate set of vehicles is filtered based on the aggregate risk quantity exceeding a risk threshold corresponding to each of the candidate set of vehicles.

15. The transportation management system of claim 14 , wherein the executed instructions cause the transportation management system to filter out a vehicle type from the candidate set of vehicles, wherein the vehicle type is selected from safety driven autonomous vehicles and fully autonomous vehicles.

16. A non-transitory computer readable medium storing:

a machine-learned risk regressor, wherein the machine-learned risk regressor is trained using log data from vehicles operated throughout a given region to determine fractional risk values for path segments of the given region, the fractional risk values being specific to at least one of an autonomous vehicle (AV) type or AV software version and indicative of a probability that a harmful event will occur for traversal of the path segments; and

instructions that, when executed by one or more processors, cause the one or more processors to:

receive live data from a plurality of vehicles in the given region; determine a fractional risk value for traversing each path segment of the given region, wherein the fractional risk value is determined using the machine-learned risk regressor using the live data as input; and

route vehicles operating within the given region along lowest risk route options based on the fractional risk value for each path segment.

17. The non-transitory computer readable medium of claim 16 , wherein the executed instructions cause the one or more processors to:

for each determination of a fractional risk value for a particular path segment, temporally correlate the fractional risk value to a current set of conditions based on sensor data from the vehicles operating within the given region, wherein the fractional risk value is temporally correlated using the machine-learned risk regressor.

18. The non-transitory computer readable medium of claim 17 , wherein the current set of conditions comprises at least one of weather conditions or road conditions.

19. The non-transitory computer readable medium of claim 17 , wherein the executed instructions cause the one or more processors to:

receive a plurality of transport requests, wherein each of the plurality of transport requests indicates a pick-up location and a drop-off location; and, for each of the plurality of transport requests:

determine a plurality of route options between the pick-up location and the drop-off location associated with the respective transport request; and

determine an aggregate risk value for each route option of the plurality of route options based at least in part on the current set of conditions and the correlated fractional risk value of each path segment along the respective route option.

20. A computer-implemented method of facilitating an on-demand transportation service, the method being performed by a computer system comprising one or more processors, the method comprising:

collecting, by the computer system, log data from vehicles operating throughout a given region, the log data comprising historical data of harmful events and close calls;

receiving live data from a plurality of vehicles in the given region;

determining, by the computer system, a fractional risk value for traversing each path segment of the given region, wherein the fractional risk value is determined using a machine-learned risk regressor with live data as input, wherein the machine-learned risk regressor comprises a neural network trained using the log data to determine fractional risk values for path segments of the given region, the fractional risk values being specific to at least one of an autonomous vehicle (AV) type or AV software version and indicative of a probability that a harmful event will occur for traversal of the path segments; and

routing, by the computer system, vehicles within the given region along lowest risk route options based on the fractional risk value for each path segment.