Data synthesis for autonomous control systems
An autonomous control system generates synthetic data that reflect simulated environments. Specifically, the synthetic data is a representation of sensor data of the simulated environment from the perspective of one or more sensors. The system generates synthetic data by introducing one or more simulated modifications to sensor data captured by the sensors or by simulating the sensor data for a virtual environment. The autonomous control system uses the synthetic data to train computer models for various detection and control algorithms. In general, this allows autonomous control systems to augment training data to improve performance of computer models, simulate scenarios that are not included in existing training data, and/or train computer models that remove unwanted effects or occlusions from sensor data of the environment.
1. A method implemented by a system of one or more processors, the method comprising:
accessing vehicle sensor data from a plurality of sensors during operation of one or more vehicles, the sensors being configured to detect information related to a real-world environment proximate to the vehicle;
training a first machine learning model to output quality information assigned to input of the vehicle sensor data, wherein the quality information includes a category, of a plurality of categories, which is associated with the vehicle sensor data;
generating synthesized sensor data which adjusts sensor data to simulate at least one of the categories in the sensor data; and
training a second machine learning model based on the generated synthesized sensor data, wherein the second machine learning model is configured to provide autonomous or semi-autonomous control of a particular vehicle.
2. The method of claim 1 , wherein the quality information indicates a value selected from a range of values.
3. The method of claim 2 , wherein the synthesized data is generated to correspond with a particular value of the range of values.
4. The method of claim 2 , wherein based on an unplugging of the sensors, the quality information is assigned a value of zero.
5. The method of claim 1 , wherein interface indicates active sensors operating on a wavelength associated with at least one of the sensors.
6. The method of claim 1 , wherein the category indicates one or more of rain, ice, light-conditions, or interference.
7. The method of claim 1 , wherein the quality information is specific to individual sensors.
8. A system comprising one or more processors and non-transitory computer storage media storing instructions that when executed by the one or more processors, cause the one or more processors to:
access vehicle sensor data from a plurality of sensors during operation of one or more vehicles, the sensors being configured to detect information related to a real-world environment proximate to the vehicle;
train a first machine learning model to output quality information assigned to input of the vehicle sensor data, wherein the quality information includes a category, of a plurality of categories, which is associated with the vehicle sensor data;
generate synthesized sensor data which adjusts sensor data to simulate at least one of the categories in the sensor data; and
train a second machine learning model based on the generated synthesized sensor data, wherein the second machine learning model is configured to provide autonomous or semi-autonomous control of a particular vehicle.
9. The system of claim 8 , wherein the quality information indicates a value selected from a range of values.
10. The system of claim 9 , wherein the synthesized data is generated to correspond with a particular value of the range of values.
11. The system of claim 9 , wherein based on an unplugging of the sensors, the quality information is assigned a value of zero.
12. The system of claim 8 , wherein interface indicates active sensors operating on a wavelength associated with at least one of the sensors.
13. The system of claim 8 , wherein the category indicates one or more of rain, ice, light-conditions, or interference.
14. The system of claim 8 , wherein the quality information is specific to individual sensors.
15. Non-transitory computer storage media storing instructions that when executed by a system of one r more processors, cause the one or more processors to:
access vehicle sensor data from a plurality of sensors during operation of one or more vehicles, the sensors being configured to detect information related to a real-world environment proximate to the vehicle;
train a first machine learning model to output quality information assigned to input of the vehicle sensor data, wherein the quality information includes a category, of a plurality of categories, which is associated with the vehicle sensor data;
generate synthesized sensor data which adjusts sensor data to simulate at least one of the categories in the sensor data; and
train a second machine learning model based on the generated synthesized sensor data, wherein the second machine learning model is configured to provide autonomous or semi-autonomous control of a particular vehicle.
16. The computer storage media of claim 15 , wherein the quality information indicates a value selected from a range of values.
17. The computer storage media of claim 16 , wherein based on an unplugging of the sensors, the quality information is assigned a value of zero.
18. The computer storage media of claim 15 , wherein interface indicates active sensors operating on a wavelength associated with at least one of the sensors.
19. The computer storage media of claim 15 , wherein the category indicates one or more of rain, ice, light-conditions, or interference.
20. The computer storage media of claim 15 , wherein the quality information is specific to individual sensors.