Granted Patent B2

US 12,354,415 · App. 17/586,389 · Granted Jul 8, 2025

Inventors: Blake Wulfe (San Francisco, CA); Adrien Gaidon (San Jose, CA)

Assignee: Toyota Research Institute, Inc.

G07C5/008G06N20/00G07C5/02B60W60/001B60W2552/00B60W2554/4046B60W2555/00

Abstract

Claims (69)

1. A method comprising:

receiving first driving data associated with a first vehicle;

receiving second driving data associated with one or more vehicles around the first vehicle;

creating training data comprising the first driving data labeled as positive data and the second driving data as unlabeled data;

using the training data to train a classifier to predict whether driving data input to the classifier is positive or unlabeled;

receiving third driving data associated with a second vehicle;

inputting the third driving data to the classifier after the classifier has been trained;

determining a reward function based on an output of the classifier;

determining a driving policy based on the reward function; and

causing the second vehicle to drive autonomously based on the driving policy.

2. The method of claim 1 , further comprising:

using the training data to train the classifier to receive a set of driving data as an input; and

outputting a probability that the set of driving data is associated with an expert driver.

3. The method of claim 1 , wherein the classifier comprises a neural network.

4. The method of claim 1 , wherein the first vehicle is driven by an expert driver.

5. The method of claim 1 , further comprising:

extracting features from the training data; and

using the features extracted from the training data to train the classifier.

6. The method of claim 1 , further comprising:

receiving roadway data associated with a road on which the first vehicle drives; and

creating the training data by associating the first driving data and the second driving data with the roadway data.

7. The method of claim 1 , further comprising:

receiving environment data associated with an environment through which the first vehicle drives; and

creating the training data by associating the first driving data and the second driving data with the roadway data.

8. The method of claim 1 , further comprising:

receiving a plurality of sets of driving data associated with a plurality of vehicle routes;

extracting features from each set of driving data associated with each vehicle route;

inputting the features extracted from each set of driving data into the trained classifier;

determining a reward function associated with each vehicle route based on an output of the trained classifier; and

selecting the vehicle route having the reward function with a largest value.

9. The method of claim 1 , further comprising:

receiving driving data associated with a plurality of unknown vehicles;

extracting features from the driving data associated with each of the unknown vehicles;

inputting the features associated with each of the unknown vehicles into the trained classifier;

determining a reward function for the driving data associated with each of the unknown vehicles based on an output of the trained classifier; and

determining second training data by labeling the driving data associated with each of the unknown vehicles based on the respective reward functions.

10. The method of claim 9 , further comprising:

labeling the driving data associated with each of the unknown vehicles having a reward function with a value greater than a predetermined threshold as positive; and

labeling the driving data associated with each of the unknown vehicles having a reward function with a value less than or equal to the predetermined threshold as negative.

11. The method of claim 9 , further comprising:

determining a driving policy based on the second training data using reinforcement learning.

12. A remote computing device comprising a controller programmed to:

receive first driving data associated with a first vehicle driven by an expert driver;

receive second driving data associated with one or more vehicles around the first vehicle;

create training data comprising the first driving data labeled as positive data and the second driving data as unlabeled data;

use the training data to train a classifier to predict whether driving data input to the classifier is positive or unlabeled;

receive third driving data associated with a second vehicle;

input the third driving data to the classifier after the classifier has been trained;

determine a reward function based on an output of the classifier;

determine a driving policy based on the reward function; and

cause the second vehicle to drive autonomously based on the driving policy.

13. The remote computing device of claim 12 , wherein the controller is further programmed to:

use the training data to train the classifier to receive a set of driving data as an input; and

output a probability that the set of driving data is associated with the expert driver.

14. The remote computing device of claim 12 , wherein the controller is further programmed to:

extract features from the training data; and

use the features extracted from the training data to train the classifier.

15. The remote computing device of claim 12 , wherein the controller is further programmed to:

receive a plurality of sets of driving data associated with a plurality of vehicle routes;

extract features from each set of driving data associated with each vehicle route;

input the features extracted from each set of driving data into the trained classifier;

determine a reward function associated with each vehicle route based on an output of the trained classifier; and

select the vehicle route having the reward function with a largest value.

16. The remote computing device of claim 12 , wherein the controller is further programmed to:

receive driving data associated with a plurality of unknown vehicles;

extract features from the driving data associated with each of the unknown vehicles;

input the features associated with each of the unknown vehicles into the trained classifier;

determine a reward function for the driving data associated with each of the unknown vehicles based on an output of the trained classifier; and

determine second training data by labeling the driving data associated with each of the unknown vehicles based on the respective reward functions.

Assignments (2)

Continuity (1)

References Cited (9)