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 system providing autonomous or semi-autonomous control of a vehicle, wherein the method comprises: obtaining vehicle sensor data from a plurality of sensors during operation of the vehicle, the sensors being located on the vehicle and configured to detect information related to a real-world environment proximate to the vehicle; providing the vehicle sensor data to one or more computer models, the one or more computer models comprising model data that was trained using synthetic sensor data modified to simulate one or more types of artifacts; and applying the one or more computer models to the vehicle sensor data to provide autonomous or semi-autonomous control of the vehicle based on output from the one or more computer models.
2. The method of claim 1 , wherein the vehicle sensor data is synthesized from the plurality of sensors comprising one or more of a camera, a microphone, a radar, or a LIDAR.
3. The method of claim 1 , wherein the synthetic sensor data comprises sensor data from a particular sensor, wherein the sensor data was modified to simulate a particular type of artifact, and wherein the one or more computer models were trained to remove the particular type of artifact.
4. The method of claim 3 , wherein the one or more computer models were trained to output sensor data with the particular type of artifact removed based on input of sensor data from the particular sensor.
5. The method of claim 1 , wherein the one or more types of artifacts include missing data included in sensor data from a particular sensor of the plurality of sensors.
6. The method of claim 5 , wherein the one or more computer models output a reconstruction of the missing data based on sensor data from one or more remaining sensors of the plurality of sensors.
7. The method of claim 6 , wherein the particular sensor is a LIDAR, and wherein the missing data is associated with a dark-colored vehicle in the real-world environment.
8. The method of claim 6 , wherein the particular sensor is a LIDAR, wherein the missing data is associated with attenuation of LIDAR signals, and wherein the remaining sensors comprise one or more cameras.
9. The method of claim 1 , wherein a type of artifact comprises atmospheric conditions, lighting conditions, and/or objects.
10. The method of claim 9 , wherein atmospheric conditions comprise one or more of precipitation, iciness or snow.
11. The method of claim 9 , wherein lighting conditions comprise one or more of sunlight or darkness.
12. The method of claim 9 , wherein objects comprise pedestrians or puddles.
13. The method of claim 1 , wherein the computer models determine quality scores associated with sensor data from the plurality of sensors, and wherein the computer models use the quality scores to prioritize individual sensors of the plurality of sensors.
14. The method of claim 13 , wherein the quality score for a particular sensor is based on lighting conditions and/or interface with one or more remaining sensors.
15. A system comprising a processor and non-transitory computer storage media storing instructions that when executed by the processor, cause the processor to perform operations, wherein the system is configured to guide a vehicle via autonomous or semi-autonomous control, and wherein the operations comprising: obtaining vehicle sensor data from a plurality of sensors during operation of the vehicle, the sensors being located on the vehicle and configured to detect information related to a real-world environment proximate to the vehicle; providing the vehicle sensor data to one or more computer models, the one or more computer models comprising model data that was trained using synthetic sensor data modified to simulate one or more types of artifacts; and applying the one or more computer models to the vehicle sensor data to provide autonomous or semi-autonomous control of the vehicle based on output from the one or more computer models.
16. The system of claim 15 , wherein the vehicle sensor data is synthesized from the plurality of sensors comprising one or more of a camera, a microphone, a radar, or a LIDAR.
17. The system of claim 15 , wherein the one or more types of artifacts include missing data included in sensor data from a particular sensor of the plurality of sensors.
18. The system of claim 17 , wherein the particular sensor comprises a LIDAR, wherein the one or more computer models output a reconstruction of the missing data based on sensor data from one or more remaining sensors of the plurality of sensors, and wherein the remaining sensors comprise one or more cameras.
19. Non-transitory computer storage media storing instructions that when executed by a processor cause the process to perform operations, the processor being configured to guide a vehicle via autonomous or semi-autonomous control, and wherein the operations comprising: obtaining vehicle sensor data from a plurality of sensors during operation of the vehicle, the sensors being located on the vehicle and configured to detect information related to a real-world environment proximate to the vehicle; providing the vehicle sensor data to one or more computer models, the one or more computer models comprising model data that was trained using synthetic sensor data modified to simulate one or more types of artifacts; and applying the one or more computer models to the vehicle sensor data to provide autonomous or semi-autonomous control of the vehicle based on output from the one or more computer models.
20. The computer-storage media of claim 19 , wherein the vehicle sensor data is synthesized from the plurality of sensors comprising one or more of a camera, a microphone, a radar, or a LIDAR.