Delay decision making for autonomous driving vehicles in response to obstacles based on confidence level and distance

1. A method of operating an autonomous driving vehicle (ADV), the method comprising:

in a first driving cycle, performing operations including:

determining an identification of an obstacle by analyzing sensor data captured by one or more sensors of the ADV using a neural network to implement computer vision;

determining a confidence score for the obstacle that represents a confidence level of how reliable the computer vision is in correctly identifying the obstacle with the computer vision, wherein the confidence score is determined at least based on a detection metric obtained from the neural network through an application programming interface (API) of the neural network, and wherein the identification of the obstacle is lower in response to an increase in a distance between the ADV and the obstacle, and wherein the confidence score is reduced in response to a line of sight between the one or more sensors and the obstacle being at least partially blocked;

generating a driving decision for the ADV based on a prediction of potential movement of the obstacle;

in response to determining that the confidence score representing the confidence level of how reliable the computer vision is in correctly identifying the obstacle is below a predetermined confidence threshold and a distance between the ADV and a potential point of contact with the obstacle is greater than a predetermined distance threshold, delaying executing the driving decision for a period of time; and

otherwise planning a path to drive the ADV based on the driving decision and providing a command to a vehicle control system which drives the ADV according to the path without delay; and

in a second driving cycle, repeating the operations performed in the first driving cycle, wherein in response to determining that a second confidence score representing the confidence level of how reliable the computer vision is in correctly identifying the obstacle during the second driving cycle is not below the predetermined confidence threshold, planning the path to drive the ADV based on the driving decision and providing the command to the vehicle control system which drives the ADV according to the path without additional delay.

2. The method of claim 1 , further comprising:

determining that the obstacle is a static obstacle based on the sensor data;

determining a current status of the obstacle; and

calculating the confidence score further based on the current status of the obstacle.

3. The method of claim 2 , wherein the current status of the obstacle comprises one or more of a speed of the obstacle or a type of the obstacle.

4. The method of claim 1 , further comprising:

determining that the obstacle is a dynamic obstacle based on the sensor data;

predicting a moving trajectory of the obstacle based on moving history of the obstacle; and

calculating the confidence score further based on the moving trajectory of the obstacle.

5. The method of claim 4 , wherein the confidence score is calculated further based on a current status of the obstacle.

6. The method of claim 1 , wherein the driving decision is a yield decision to slow down the ADV.

7. The method of claim 1 , wherein the driving decision is a stop decision to stop the ADV.

8. A non-transitory machine-readable medium having instructions stored therein, which when executed by a processor, cause the processor to perform operations of operating an autonomous driving vehicle (ADV), the operations comprising:

in a first driving cycle,

determining an identification of an obstacle by analyzing sensor data captured by one or more sensors of the ADV using a neural network to implement computer vision;

determining a confidence score for the obstacle that represents a confidence level of how reliable the computer vision is in correctly identifying the obstacle with the computer vision, wherein the confidence score is determined at least based on a detection metric obtained from the neural network through an application programming interface (API) of the neural network, and wherein the identification of the obstacle is lower in response to an increase in a distance between the ADV and the obstacle, and wherein the confidence score is reduced in response to a line of sight between the one or more sensors and the obstacle being at least partially blocked;

generating a driving decision for the ADV based on a prediction of potential movement of the obstacle;

in response to determining that the confidence score representing the confidence level of how reliable the computer vision is in correctly identifying the obstacle is below a predetermined confidence threshold and a distance between the ADV and a potential point of contact with the obstacle is greater than a predetermined distance threshold, delaying executing the driving decision for a period of time;

otherwise planning a path to drive the ADV based on the driving decision and providing a command to a vehicle control system which drives the ADV according to the path without delay; and

in a second driving cycle, repeating the operations performed in the first driving cycle, wherein in response to determining that a second confidence score representing the confidence level of how reliable the computer vision is in correctly identifying the obstacle during the second driving cycle is not below the predetermined confidence threshold, planning the path to drive the ADV based on the driving decision and providing the command to the vehicle control system which drives the ADV according to the path without additional delay.

9. The machine-readable medium of claim 8 , wherein the operations further comprise:

determining that the obstacle is a static obstacle based on the sensor data;

determining a current status of the obstacle; and

calculating the confidence score further based on the current status of the obstacle.

10. The machine-readable medium of claim 9 , wherein the current status of the obstacle comprises one or more of a speed of the obstacle or a type of the obstacle.

11. The machine-readable medium of claim 8 , wherein the operations further comprise:

determining that the obstacle is a dynamic obstacle based on the sensor data;

predicting a moving trajectory of the obstacle based on moving history of the obstacle; and

calculating the confidence score further based on the moving trajectory of the obstacle.

12. The machine-readable medium of claim 11 , wherein the confidence score is calculated further based on a current status of the obstacle.

13. The machine-readable medium of claim 8 , wherein the driving decision is a yield decision to slow down the ADV.

14. The machine-readable medium of claim 8 , wherein the driving decision is a stop decision to stop the ADV.

15. A data processing system, comprising:

a processor; and

a memory coupled to the processor to store instructions, which when executed by the processor, cause the processor to perform operations of operating an autonomous driving vehicle (ADV), the operations including:

in a first driving cycle,

determining an identification of an obstacle by analyzing sensor data captured by one or more sensors of the ADV using a neural network to implement computer vision;

determining a confidence score for the obstacle that represents a confidence level of how reliable the computer vision is in correctly identifying the obstacle with the computer vision, wherein the confidence score is determined at least based on a detection metric obtained from the neural network through an application programming interface (API) of the neural network, and wherein the identification of the obstacle is lower in response to an increase in a distance between the ADV and the obstacle, and wherein the confidence score is reduced in response to a line of sight between the one or more sensors and the obstacle being at least partially blocked,

generating a driving decision for the ADV based on a prediction of potential movement of the obstacle,

in response to determining that the confidence score representing the confidence level of how reliable the computer vision is in correctly identifying the obstacle is below a predetermined confidence threshold and a distance between the ADV and a potential point of contact with the obstacle is greater than a predetermined distance threshold, delaying executing the driving decision for a period of time,

otherwise planning a path to drive the ADV based on the driving decision and providing a command to a vehicle control system which drives the ADV according to the path without delay, and

in a second driving cycle, repeating the operations performed in the first driving cycle, wherein in response to determining that a second confidence score representing the confidence level of how reliable the computer vision is in correctly identifying the obstacle during the second driving cycle is not below the predetermined confidence threshold, planning the path to drive the ADV based on the driving decision and providing the command to the vehicle control system which drives the ADV according to the path without additional delay.

16. The system of claim 15 , wherein the operations further comprise:

determining that the obstacle is a static obstacle based on the sensor data;

determining a current status of the obstacle; and

calculating the confidence score further based on the current status of the obstacle.

17. The system of claim 16 , wherein the current status of the obstacle comprises one or more of a speed of the obstacle or a type of the obstacle.

18. The system of claim 15 , wherein the operations further comprise:

determining that the obstacle is a dynamic obstacle based on the sensor data;

predicting a moving trajectory of the obstacle based on moving history of the obstacle; and

calculating the confidence score further based on the moving trajectory of the obstacle.

19. The system of claim 18 , wherein the confidence score is calculated further based on a current status of the obstacle.

20. The system of claim 15 , wherein the driving decision is a yield decision to slow down the ADV.

21. The system of claim 15 , wherein the driving decision is a stop decision to stop the ADV.