Machine Learning for Predicting Locations of Objects Perceived by Autonomous Vehicles

• US 20190025841A1
• Filed: 08/23/2017
• Published: 01/24/2019
• Est. Priority Date: 07/21/2017
• Status: Active Grant

• Current Assignee: Uatc, LLC (Uber Technologies, Inc.)
• Original Assignee: Uber Technologies, Inc.
• Inventors: Haynes, Clark, Dewancker, Ian, Djuric, Nemanja, Huang, Tzu-Kuo, Lan, Tian, Lin, Hank, Marchetti-Bowick, Micol, Radosavljevic, Vladan, Schneider, Jeff, Styler, Alex, Traft, Neil, Wang, Huahua, Stentz, Tony
• Granted Patent: US 10,579,063 B2
• Time in Patent Office: Days
• CPC Class Codes: B60W 30/00 Purposes of road vehicle drive control systems not related to the control of a particular sub-unit, e.g. of systems using conjoint control of vehicle sub-units
• CPC Class Codes: G01C 21/20 Instruments for performing navigational calculations G01C21/24, G01C21/26 take precedence
• CPC Class Codes: G01C 21/3461 Preferred or disfavoured areas, e.g. dangerous zones, toll or emission zones, intersections, manoeuvre types, segments such as motorways, toll roads, ferries
• CPC Class Codes: G01C 21/3492 employing speed data or traffic data, e.g. real-time or historical traffic control systems for road vehicles involving transmission of navigation instructions to the vehicle G08G1/0968
• CPC Class Codes: G01S 13/726 Multiple target tracking
• CPC Class Codes: G01S 13/931 of land vehicles
• CPC Class Codes: G01S 17/931 of land vehicles
• CPC Class Codes: G01S 2013/9323 Alternative operation using light waves
• CPC Class Codes: G05D 1/0214 in accordance with safety or protection criteria, e.g. avoiding hazardous areas monitoring the location of vehicles within a certain area, e.g. forbidden or allowed areas, in traffic control systems for road vehicles G08G1/13
• CPC Class Codes: G05D 1/0221 involving a learning process
• CPC Class Codes: G08G 1/00 Traffic control systems for road vehicles arrangement of road signs or traffic signals E01F9/00 ; automatic vehicle control B62D
• CPC Class Codes: G08G 1/165 for passive traffic, e.g. including static obstacles, trees
• CPC Class Codes: G08G 1/166 for active traffic, e.g. moving vehicles, pedestrians, bikes

Abstract

The present disclosure provides systems and methods for predicting the future locations of objects that are perceived by autonomous vehicles. An autonomous vehicle can include a prediction system that, for each object perceived by the autonomous vehicle, generates one or more potential goals, selects one or more of the potential goals, and develops one or more trajectories by which the object can achieve the one or more selected goals. The prediction systems and methods described herein can include or leverage one or more machine-learned models that assist in predicting the future locations of the objects. As an example, in some implementations, the prediction system can include a machine-learned static object classifier, a machine-learned goal scoring model, a machine-learned trajectory development model, a machine-learned ballistic quality classifier, and/or other machine-learned models. The use of machine-learned models can improve the speed, quality, and/or accuracy of the generated predictions.

Claims

1. A computer system, comprising:

one or more processors;

a scenario generation system implemented by the one or more processors, wherein the scenario generation system is configured to;

receive state data descriptive of at least a current or past state of an object that is perceived by an autonomous vehicle; and

generate one or more goals for the object based at least in part on the state data, wherein each of the one or more goals corresponds to a set of one or more decisions that result in the object achieving a goal location;

wherein the scenario generation system comprises a machine-learned goal scoring model configured to generate a score for each of the one or more goals generated for the object; and

wherein the score generated by the machine-learned goal scoring model for each goal is indicative of a predicted likelihood that the object will pursue the set of one or more decisions associated with such goal to achieve the goal location associated with such goal; and

a scenario development system implemented by the one or more processors,

wherein at least one of the scenario generation system and the scenario development system is further configured to select at least one of the one or more goals based at least in part on the scores generated for the one or more goals by the machine-learned goal scoring model; and

wherein the scenario development system is configured to;

receive data indicative of the at least one of the one or more goals selected by the at least one of the scenario generation system and the scenario development system; and

determine at least one predicted trajectory by which the object achieves the at least one goal location associated with the at least one of the one or more goals selected by the at least one of the scenario generation system and the scenario development system.

2. The computer system of claim 1, further comprising:

a motion planning system implemented by the one or more processors, wherein the motion planning system is configured to generate a motion plan for the autonomous vehicle based at least in part on the at least one predicted trajectory determined for the object by the scenario development system.

3. The computer system of claim 1, wherein the scenario generation system comprises a machine-learned static object classifier configured to classify the object as either static or not static.

4. The computer system of claim 3, wherein the machine-learned static object classifier comprises a machine-learned random forest model.

6. The computer system of claim 1, wherein the goal location for each goal comprises one or more nominal pathways that correspond to common paths of travel, and wherein the machine-learned goal scoring model comprises a machine-learned binary classifier model that has been trained to score the object relative to the one or more nominal pathways respectively associated with each goal.

8. The computer system of claim 1, wherein:

the scenario generation system comprises a machine-learned static object classifier configured to classify the object as either static or not static; and

the one or more processors implement the machine-learned goal scoring model to generate the score for each of the one or more goals when the machine-learned static object classifier classifies the object as not static.

9. The computer system of claim 1, wherein the machine-learned goal scoring model comprises a machine-learned random forest model.

10. The computer system of claim 1, wherein the scenario development system comprises a machine-learned trajectory development model configured to determine the at least one predicted trajectory by which the object achieves the at least one of the one or more goals.

11. The computer system of claim 10, wherein:

the at least one of the one or more goals comprises continued travel of the object along a nominal pathway; and

the machine-learned trajectory development model provides the at least one predicted trajectory in coordinates defined relative to the nominal pathway.

12. The computer system of claim 10, wherein the machine-learned trajectory development model comprises a machine-learned regression forest model or a deep neural network.

13. The computer system of claim 12, wherein:

the machine-learned regression forest model comprises a plurality of decision trees that each output a plurality of sequences of future states of the object; and

the at least one predicted trajectory comprises a weighted sum of the plurality of sequences of future states of the object.

14. The computer system of claim 10, wherein the scenario generation system comprises a machine-learned goal scoring model configured to generate scores for goals generated by the scenario generation system, and wherein the scenario development system further comprises:

a trajectory scoring model configured to provide scores for predicted trajectories generated by the machine-learned trajectory development model;

a ballistics motion model configured to determine the at least one predicted trajectory for the object when at least one of;

the scenario generation system fails to generate one or more goals for the object that receive a first score from the machine-learned goal scoring model greater than a first threshold score; and

the machine-learned trajectory development model fails to generate one or more predicted trajectories for the object that receive a second score from the trajectory scoring model greater than a second threshold score; and

a ballistic quality classifier configured to classify the at least one predicted trajectory determined by the ballistics motion model as satisfactory or unsatisfactory.

21. The computer system of claim 1, wherein the at least one predicted trajectory comprises a sequence of future states for the object.

23. The computer system of claim 1, wherein to generate the one or more goals for the object the scenario generation system is configured to generate a plurality of lane sequences by rolling out a current lane of the object respectively into a plurality of successor lanes.

5. (canceled)

7. (canceled)

15. An autonomous vehicle, comprising:

one or more processors; and

one or more non-transitory computer-readable media that collectively store instructions that, when executed by the one or more processors, cause the one or more processors to perform operations, the operations comprising;

obtaining state data descriptive of at least one current or past state of an object that is perceived by an autonomous vehicle;

generating one or more goals for the object based at least in part on the state data, wherein each of the one or more goals corresponds to a set of one or more decisions that result in the object achieving a goal location;

selecting at least a first goal of the one or more goals generated for the object, wherein selecting at least the first goal of the one or more of the goals generated for the object comprises;

inputting at least the state data descriptive of the at least one current or past state of the object and data descriptive of the one or more goals generated for the object into a machine-learned goal scoring model implemented by the one or more processors;

receiving one or more scores for the object respectively relative to the one or more goals as an output of the machine-learned goal scoring model, wherein the score output by the machine-learned goal scoring model for each goal is indicative of a predicted likelihood that the object will pursue the set of one or more decisions associated with such goal to achieve the goal location associated with such goal; and

selecting at least the first goal from the one or more goals based at least in part on the one or more scores; and

determining at least a first predicted trajectory by which the object achieves a first goal location associated with the first goal.

16. The autonomous vehicle of claim 15, wherein generating the one or more goals for the object comprises:

inputting at least the state data descriptive of the at least one current or past state of the object into a machine-learned static object classifier; and

receiving a classification of the object as either static or not static as an output of the machine-learned static object classifier;

wherein the first goal selected for the object comprises a remaining stationary goal when the machine-learned static object classifier classifies the object as static.

18. The autonomous vehicle of claim 15, wherein determining at least the first predicted trajectory by which the object achieves the first goal comprises:

inputting at least the state data descriptive of the at least one current or past state of the object and data descriptive of at least the first goal into a machine-learned trajectory development model; and

receiving at least the first predicted trajectory as an output of the machine-learned trajectory development model.

19. The autonomous vehicle of claim 15, wherein determining at least the first predicted trajectory by which the object achieves the first goal comprises:

inputting at least the state data descriptive of the at least one current or past state of the object into a ballistics motion model; and

receiving at least the first predicted trajectory as an output of the ballistics motion model.

17. (canceled)

20. A computer-implemented method, comprising:

obtaining, by a computing system comprising one or more computing devices, state data descriptive of at least one current or past state of an object that is perceived by an autonomous vehicle;

generating, by the computing system, one or more goals for the object based at least in part on the state data, wherein each of the one or more goals corresponds to a set of one or more decisions that result in the object achieving a goal location;

selecting, by the computing system, at least a first goal of the one or more goals generated for the object, wherein selecting, by the computing system, at least the first goal of the one or more of the goals generated for the object comprises;

inputting, by the computing system, at least the state data descriptive of the at least one current or past state of the object and data descriptive of the one or more goals generated for the object into a machine-learned goal scoring model implemented by the one or more processors;

receiving, by the computing system, one or more scores for the object respectively relative to the one or more goals as an output of the machine-learned goal scoring model, wherein the score generated by the machine-learned goal scoring model for each goal is indicative of a predicted likelihood that the object will pursue the set of one or more decisions associated with such goal to achieve the goal location associated with such goal; and

selecting, by the computing system, at least the first goal from the one or more goals based at least in part on the one or more scores; and

determining, by the computing system, at least a first predicted trajectory by which the object achieves the first goal.

22. (canceled)