Filtering user responses for generating training data for machine learning based models for navigation of autonomous vehicles
An autonomous vehicle uses machine learning based models such as neural networks to predict hidden context attributes associated with traffic entities. The hidden context represents behavior of the traffic entities in the traffic. The machine learning based model is configured to receive a video frame as input and output likelihoods of receiving user responses having particular ordinal values. The system uses a loss function based on cumulative histogram of user responses corresponding to various ordinal values. The system identifies user responses that are unlikely to be valid user responses to generate training data for training the machine learning mode. The system identifies invalid user responses based on response time of the user responses.
1. A method comprising:
for each of a plurality of images:
sending the image for presentation to a set of users, and for each of the set of users, receiving a user response describing a hidden context attribute for a traffic entity displayed in the image,
wherein each user response is associated with a user response time and each user response is an ordinal value selected from a plurality of ordinal values;
determining a threshold value for user response times by:
determining a plurality of threshold values for user response times;
for each of the plurality of threshold values:
determining a statistical distribution of user responses having a user response time above the threshold value represented by a first histogram and a statistical distribution of user responses having a user response time below the threshold value represented by a second histogram, and
determining a measure of difference between the statistical distribution of user responses having a user response time above the threshold value and the statistical distribution of user responses having a user response time below the threshold value;
selecting the threshold value based on the determined measure of differences for the plurality of threshold values, wherein the measure of difference is an aggregate of the differences of the rate of occurrences of the ordinal values between the first histogram and the second histogram;
selecting a subset of user responses having a user response time above the threshold value;
generating training data set using the subset of user responses;
training a neural network using the training data set, the neural network configured to receive an input image and predict the hidden context attribute for the input image; and
navigating an autonomous vehicle, based on the neural network.
2. The method of claim 1 , wherein the hidden context attribute represents a state of mind of a user represented by the traffic entity.
3. The method of claim 1 , wherein the hidden context attribute represents a task that a user represented by the traffic entity is planning on accomplishing.
4. The method of claim 1 , wherein the hidden context attribute represents a degree of awareness of a user represented by the traffic entity about the autonomous vehicle.
5. The method of claim 1 , wherein the hidden context attribute represents a goal of a user represented by the traffic entity, wherein the user expects to achieve the goal within a threshold time interval.
6. The method of claim 1 , wherein navigating the autonomous vehicle comprises:
capturing an image;
providing the image as input to the neural network; and
determining signals sent to controls of the autonomous vehicle based on an output of the neural network.
7. A non-transitory computer readable storage medium storing instructions that when executed by a computer processor, cause the computer processor to perform steps comprising:
for each of a plurality of images:
sending the image for presentation to a set of users, and for each of the set of users, receiving a user response describing a hidden context attribute for a traffic entity displayed in the image,
wherein each user response is associated with a user response time and each user response is an ordinal value selected from a plurality of ordinal values;
determining a threshold value for user response times by:
determining a plurality of threshold values for user response times;
for each of the plurality of threshold values:
determining a statistical distribution of user responses having a user response time above the threshold value represented by a first histogram and a statistical distribution of user responses having a user response time below the threshold value represented by a second histogram, and
determining a measure of difference between the statistical distribution of user responses having a user response time above the threshold value and the statistical distribution of user responses having a user response time below the threshold value;
selecting the threshold value based on the determined measure of differences for the plurality of threshold values, wherein the measure of difference is an aggregate of the differences of the rate of occurrences of the ordinal values between the first histogram and the second histogram;
selecting a subset of user responses having a user response time above the threshold value;
generating training data set using the subset of user responses;
training a neural network using the training data set, the neural network configured to receive an input image and predict the hidden context attribute for the input image; and
navigating an autonomous vehicle, based on the neural network.
8. The non-transitory computer readable storage medium of claim 7 , wherein the instructions further cause the computer processor to perform steps comprising:
capturing an image by the autonomous vehicle;
providing the image as input to the neural network; and
determining signals sent to controls of the autonomous vehicle based on an output of the neural network.
9. The non-transitory computer readable storage medium of claim 7 , wherein the hidden context attribute represents a state of mind of a user represented by the traffic entity.
10. The non-transitory computer readable storage medium of claim 7 , wherein the hidden context attribute represents a degree of awareness of a user represented by the traffic entity about the autonomous vehicle.
11. A computer system comprising:
a computer processor; and
a non-transitory computer readable storage medium storing instructions that when executed by the computer processor, cause the computer processor to perform steps comprising:
for each of a plurality of images:
sending the image for presentation to a set of users, and for each of the set of users, receiving a user response describing a hidden context attribute for a traffic entity displayed in the image,
wherein each user response is associated with a user response time and each user response is an ordinal value selected from a plurality of ordinal values;
determining a threshold value for user response times by:
determining a plurality of threshold values for user response times;
for each of the plurality of threshold values:
determining a statistical distribution of user responses having a user response time above the threshold value represented by a first histogram and a statistical distribution of user responses having a user response time below the threshold value represented by a second histogram, and
determining a measure of difference between the statistical distribution of user responses having a user response time above the threshold value and the statistical distribution of user responses having a user response time below the threshold value;
selecting the threshold value based on the determined measure of differences for the plurality of threshold values, wherein the measure of difference is an aggregate of the differences of the rate of occurrences of the ordinal values between the first histogram and the second histogram;
selecting a subset of user responses having a user response time above the threshold value;
generating training data set using the subset of user responses;
training a neural network using the training data set, the neural network configured to receive an input image and predict the hidden context attribute for the input image; and
navigating an autonomous vehicle, based on the neural network.
12. The computer system of claim 11 , wherein the instructions further cause the computer processor to perform steps comprising:
capturing an image by the autonomous vehicle;
providing the image as input to the neural network; and
determining signals sent to controls of the autonomous vehicle based on an output of the neural network.
13. The computer system of claim 11 , wherein the hidden context attribute represents a state of mind of a user represented by the traffic entity.
14. The computer system of claim 11 , wherein the hidden context attribute represents a degree of awareness of a user represented by the traffic entity about the autonomous vehicle.