Systems and methods for training a driving agent based on real-world driving data

1. A method, comprising:

receiving, by a device, video data from a video camera of a vehicle, the video data including a plurality of video frames and corresponding global positioning system (GPS) data of the vehicle and inertial measurement unit (IMU) data associated with the vehicle;

removing, by the device, video frames from the video data to generate modified video data;

selecting, by the device, objects and image regions of video frames of the modified video data;

determining, by the device, a current speed and a current turn angle of the vehicle based on the GPS data, the IMU data, and the modified video data;

masking, by the device, the objects of the video frames of the modified video data to learn first features of the video frames of the modified video data;

masking, by the device, the image regions of the video frames of the modified video data to learn second features of the video frames of the modified video data;

generating, by the device and based on the current speed, the current turn angle, the first features, and the second features, a trained neural network for controlling the vehicle; and

performing, by the device, one or more actions based on the trained neural network model.

2. The method of claim 1 , further comprising:

aligning GPS speed, using the IMU data, to compensate for GPS speed delay and to generate modified GPS data,

wherein the GPS speed and the GPS speed delay are associated with the vehicle.

3. The method of claim 2 , wherein aligning the GPS speed, using the IMU data, to compensate for the GPS speed delay and to generate the modified GPS data comprises:

integrating a GPS acceleration to generate a first GPS speed signal;

interpolating a second GPS speed signal based on the first GPS speed signal;

normalizing the first GPS speed signal and the second GPS speed signal to generate a normalized first GPS speed signal and a normalized second GPS speed signal;

estimating the GPS speed delay based on the normalized first GPS speed signal and the normalized second GPS speed signal; and

generating the modified GPS data based on the GPS speed delay.

4. The method of claim 1 , wherein the trained neural network model is configured to predict a future speed and a future turn angle of the vehicle based on additional video data.

5. The method of claim 1 , wherein the trained neural network model is one of a convolutional neural network (CNN) model, an attention CNN model, or a vision transformer model.

6. The method of claim 1 , wherein selecting the objects of the video frames of the modified video data comprises:

utilizing an object detection model to select the objects of the video frames of the modified video data.

7. The method of claim 1 , wherein selecting the image regions of the video frames of the modified video data comprises:

utilizing a semantic segmentation model to select the image regions of the video frames of the modified video data.

8. A device, comprising:

one or more processors configured to:

receive video data that includes a plurality of video frames and corresponding global positioning system (GPS) data and inertial measurement unit (IMU) data associated with a vehicle;

align GPS speed, using the IMU data, to compensate for GPS speed delay and to generate modified GPS data,

wherein the GPS speed and the GPS speed delay are associated with the vehicle;

remove video frames from the video data to generate modified video data;

select objects and image regions of video frames of the modified video data;

determine a current speed and a current turn angle of the vehicle based on the modified GPS data, the IMU data, and the modified video data;

mask the objects of the video frames of the modified video data to learn first features of the video frames of the modified video data;

mask the image regions of the video frames of the modified video data to learn second features of the video frames of the modified video data;

generate a trained neural network model based on the current speed, the current turn angle, the first features, and the second features; and

perform one or more actions based on the trained neural network model.

9. The device of claim 8 , wherein the vehicle is a self-driving vehicle or includes an advanced driver assistance system.

10. The device of claim 8 , wherein the one or more processors, to perform the one or more actions based on the trained neural network model, are configured to:

implement the trained neural network model in the vehicle to enable the vehicle to predict a future speed and a future turn angle of the vehicle based on additional video data.

11. The device of claim 8 , wherein the one or more processors, to perform the one or more actions based on the trained neural network model, are configured to:

utilize the trained neural network model to calculate an expected maneuver of the vehicle when the vehicle is stopped at a traffic light.

12. The device of claim 8 , wherein the one or more processors, to perform the one or more actions based on the trained neural network model, are configured to:

utilize the trained neural network model to identify an incorrect driving maneuver of the vehicle and to generate an alert of the incorrect driving maneuver.

13. The device of claim 8 , wherein the one or more processors, to perform the one or more actions based on the trained neural network model, are configured to:

utilize the trained neural network model to identify aggressive driving behavior associated with the vehicle and to generate an alert of the aggressive driving behavior.

14. The device of claim 8 , wherein the one or more processors, to perform the one or more actions based on the trained neural network model, are configured to one or more of:

utilize the trained neural network model to identify a defective sensor of the vehicle to generate an alert of the defective sensor; or

process additional video data, with the trained neural network model, to predict a future driving behavior of the vehicle.

15. A non-transitory computer-readable medium storing a set of instructions, the set of instructions comprising:

one or more instructions that, when executed by one or more processors of a device, cause the device to:

receive video data that includes a plurality of video frames and corresponding global positioning system (GPS) data and inertial measurement unit (IMU) data associated with a vehicle;

remove video frames from the video data to generate modified video data;

select objects and image regions of video frames of the modified video data;

determine a current speed and a current turn angle of the vehicle based on the GPS data, the IMU data, and the modified video data;

mask the objects of the video frames of the modified video data to learn first features of the video frames of the modified video data;

mask the image regions of the video frames of the modified video data to learn second features of the video frames of the modified video data;

generate a trained neural network model based on the current speed, the current turn angle, the first features, and the second features; and

implement the trained neural network model in the vehicle to enable the vehicle to predict a future speed and a future turn angle of the vehicle based on additional video data.

16. The non-transitory computer-readable medium of claim 15 , wherein the one or more instructions further cause the device to:

align GPS speed, using the IMU data, to compensate for GPS speed delay and to generate modified GPS data,

wherein the GPS speed and the GPS speed delay are associated with the vehicle.

17. The non-transitory computer-readable medium of claim 16 , wherein the one or more instructions, that cause the device to align the GPS speed, using the IMU data, to compensate for the GPS speed delay and to generate the modified GPS data, cause the device to:

integrate a GPS acceleration to generate a first GPS speed signal;

interpolate a second GPS speed signal based on the first GPS speed signal;

normalize the first GPS speed signal and the second GPS speed signal to generate a normalized first GPS speed signal and a normalized second GPS speed signal;

estimate the GPS speed delay based on the normalized first GPS speed signal and the normalized second GPS speed signal; and

generate the modified GPS data based on the GPS speed delay.

18. The non-transitory computer-readable medium of claim 15 , wherein the trained neural network model is configured to predict a future speed and a future turn angle of the vehicle based on additional video data.

19. The non-transitory computer-readable medium of claim 15 , wherein the one or more instructions, that cause the device to select the objects of the video frames of the modified video data, cause the device to:

utilize an object detection model to select the objects of the video frames of the modified video data.

20. The non-transitory computer-readable medium of claim 15 , wherein the one or more instructions, that cause the device to select the image regions of the video frames of the modified video data, cause the device to:

utilize a semantic segmentation model to select the image regions of the video frames of the modified video data.