Verification of vehicle prediction function

Patent metadata

Field	Value
Publication number	US-2024067216-A1
Application number	US-202217894006-A
Country	US
Kind code	A1
Filing date	Aug 23, 2022
Priority date	Aug 23, 2022
Publication date	Feb 29, 2024
Grant date	—

Abstract

In some examples, a system receives sensor data from sensors on board a vehicle. Based at least on the sensor data indicating an anomaly has occurred and is affecting a travel path of the vehicle, an uncertainty threshold is compared with an uncertainty value associated with a prediction model. For instance, the uncertainty value may be indicative of a difference between a predicted value predicted by the prediction model and an actual measured value related to at least one operation parameter of the vehicle. Based at least on determining that the anomaly has occurred and that the uncertainty value is within the uncertainty threshold, sending, by the system, a communication for obtaining an updated prediction model.

First claim

What is claimed:

1 . A system for evaluating or updating a prediction model of autonomous vehicle behavior, the system comprising:
one or more processors configured by executable instructions to perform operations comprising:
receiving sensor data from sensors on board a vehicle;
based at least on the sensor data indicating an anomaly has occurred and is affecting a travel path of the vehicle, comparing an uncertainty threshold with an uncertainty value associated with a prediction model, wherein the uncertainty value is indicative of a difference between a predicted value predicted by the prediction model and an actual measured value related to at least one operation parameter of the vehicle; and
based at least on determining that the anomaly has occurred and that the uncertainty value is within the uncertainty threshold, sending a communication for obtaining an updated prediction model.

2 . The system as recited in claim 1 , wherein the anomaly includes at least one of:
construction, an accident, a road blockage that is different from information on a map being used by the vehicle for navigation, a change in weather that affects a sensing ability of one or more of the sensors on board the vehicle, or a change in a driving condition of the vehicle detected by one or more of the sensors on board the vehicle.

3 . The system as recited in claim 1 , wherein the one or more processors comprise:
at least one integrated electronic control unit (ECU) and a plurality of zone ECUs, wherein the integrated ECU receives the sensor data, wherein the plurality of zone ECUs receive recognition information processed by the integrated ECU, and wherein one or more of the zone ECUs control the at least one operation parameter of the vehicle base at least in part on an output of the prediction model, wherein the prediction model is configured to at least predict behavior of the vehicle.

4 . The system as recited in claim 3 , wherein the at least one integrated ECU performs the comparing the uncertainty threshold with the uncertainty value to determine whether to request an updated prediction model based on a result of the comparing.

5 . The system as recited in claim 1 , wherein requesting the updated prediction model comprises at least one of:
sending data to a computing device remote from the vehicle to request that the computing device retrain the prediction model to obtain an updated prediction model; or sending data to the computing device to request that the computing device provide a trained model from a model library maintained by the computing device.

6 . The system as recited in claim 1 , the operations further comprising:
subsequently, based at least on determining that the anomaly does not exist and that the uncertainty value from the prediction model exceeds the threshold, determining that a false negative is detected, and based at least on identifying an error in model input information, updating input data for the prediction model.

7 . The system as recited in claim 1 , the operations further comprising:
based at least on determining that another anomaly is detected, determining that the model is working correctly based on the uncertainty value exceeding the uncertainty threshold; and transmitting information related to the anomaly to one or more other vehicles in a vicinity of the vehicle.

8 . The system as recited in claim 1 , the operations further comprising:
based at least on determining that the anomaly has occurred and that the uncertainty value is within the uncertainty threshold, detecting a false positive; sending information related to the anomaly and actual operation parameters of the vehicle to the remote computing device for retraining of the model, wherein the retraining of the model includes updating model parameters based on the information related to the anomaly and the actual operation parameters of the vehicle; and receiving the retrained model at the vehicle as an updated prediction model.

9 . The system as recited in claim 1 , wherein:
the computing device maintains a model library including a plurality of models trained based on anomaly types; and based on the detection of the anomaly, the computing device selects an updated model from the model library to provide to the vehicle as an updated prediction model.

10 . A method comprising:
receiving, by one or more processors, sensor data from a plurality of sensors on board a vehicle; based at least on the sensor data indicating an anomaly has occurred and is affecting a travel path of the vehicle, comparing, by the one or more processors, an uncertainty threshold with an uncertainty value associated with a prediction model, wherein the uncertainty value is indicative of a difference between a predicted value predicted by the prediction model and an actual measured value related to at least one operation parameter of the vehicle; and based at least on determining that the anomaly has occurred and that the uncertainty value is within the uncertainty threshold, sending, by the one or more processors, a communication for obtaining an updated prediction model.

11 . The method as recited in claim 10 , wherein the anomaly includes at least one of:
construction, an accident, a road blockage that is different from information on a map being used by the vehicle for navigation, a change in weather that affects a sensing ability of one or more of the sensors on board the vehicle, or a change in a driving condition of the vehicle detected by one or more of the sensors on board the vehicle.

12 . The method as recited in claim 10 , wherein:
the computing device maintains a model library including a plurality of models trained based on anomaly types; and based on the detection of the anomaly, the computing device selects an updated model from the model library to provide to the vehicle as an updated prediction model.

13 . The method as recited in claim 8 , further comprising:
subsequently, based at least on determining that the anomaly does not exist and that the uncertainty value from the prediction model exceeds the threshold, determining that a false negative is detected, and based at least on identifying an error in model input information, updating input data for the prediction model.

14 . The method as recited in claim 10 , further comprising:
based at least on determining that another anomaly is detected, determining that the model is working correctly based on the uncertainty value exceeding the uncertainty threshold; and transmitting information related to the other anomaly to one or more other vehicles in a vicinity of the vehicle.

15 . The method as recited in claim 10 , further comprising:
based at least on determining that the anomaly has occurred and that the uncertainty value is within the uncertainty threshold, detecting a false positive; sending information related to the anomaly and actual operation parameters of the vehicle to the remote computing device for retraining of the model, wherein the retraining of the model includes updating model parameters based on the information related to the anomaly and the actual operation parameters of the vehicle; and receiving the retrained model at the vehicle as an updated prediction model.

16 . A non-transitory computer-readable medium including executable instructions, which, when executed by one or more processors, configure the one or more processors to perform operations comprising:
receiving sensor data from sensors on board a vehicle; based at least on the sensor data indicating an anomaly has occur

Assignees

• Hitachi Astemo Ltd

Inventors

• Gokhale Akshay Vijay
• Kundu Subrata Kumar
• Lonari Yashodeep Dilip

Classifications

• B60W60/0015 Primary

  specially adapted for safety · CPC title

• B60W40/09

  Driving style or behaviour · CPC title

• B60W2555/20

  Ambient conditions, e.g. wind or rain · CPC title

• B60W2556/50

  of positioning data, e.g. GPS [Global Positioning System] data · CPC title

• B60W2556/40

  High definition maps · CPC title