KR-20260064510-A - Method and Apparatus for Generating Training Dataset for Driving Path Generation of Autonomous Vehicles in Edge Case Scenarios

KR20260064510AKR 20260064510 AKR20260064510 AKR 20260064510AKR-20260064510-A

Abstract

A method and apparatus for generating a training dataset for generating a driving path of an autonomous vehicle in an edge case situation are disclosed. According to one aspect of the present disclosure, a method for generating a learning dataset for generating a driving path of an autonomous vehicle in an edge case situation is provided, comprising: generating a trigger signal indicating the detection of the edge case when the edge case is detected; acquiring sensor data for a certain period of time based on the trigger signal; and constructing a learning dataset based on the sensor data and a correct path generated based on the sensor data, wherein the edge case consists of a total of 118 scenarios, and the 118 scenarios consist of a plurality of scenarios set for each of six types of situations: perception error, software error, hardware error, weather limit, road driving limit, and risk judgment limit.

Inventors

김준영
성재호
김태형
김봉섭
윤경수

Assignees

재단법인 지능형자동차부품진흥원

Dates

Publication Date: 20260507
Application Date: 20250912
Priority Date: 20241030

Claims (13)

As a method for generating a training dataset for generating a driving path of an autonomous vehicle in edge case situations, A process of generating a trigger signal indicating the detection of the edge case when the edge case is detected; A process of acquiring sensor data for a certain period of time based on the above trigger signal; and The process includes constructing a training dataset based on the sensor data and the correct answer path generated based on the sensor data, wherein The above edge cases consist of a total of 118 scenarios, and A method characterized in that the above 118 scenarios consist of multiple scenarios set for each of six types of situations: cognitive error, software error, hardware error, weather limit, road driving limit, and risk judgment limit.
In paragraph 1, The above correct path is a method that is either a semi-automatic path or a manual path.
In paragraph 2, The above semi-automatic path is a method, which is an estimated driving path generated by applying a rule algorithm to the sensor data.
In paragraph 3, The above rule algorithm is a method that is Dijkstra's algorithm or A-Star's algorithm.
In paragraph 2, The above manual path is the actual driving path of the autonomous vehicle recorded in the sensor data.
In paragraph 1, The above training dataset includes a plurality of sensor data and a plurality of correct paths for a plurality of edge cases, and A method comprising the process of constructing the above-mentioned training dataset, wherein the process of labeling the above-mentioned correct path for a specific edge case onto the sensor data for the specific edge case.
As a device for generating a training dataset for generating a driving path of an autonomous vehicle in edge case situations, Memory for storing instructions; and at least one processor, comprising The above processor, by executing the above instructions, A process of generating a trigger signal indicating the detection of the edge case when the edge case is detected; A process of acquiring sensor data for a certain period of time based on the above trigger signal; and Based on the sensor data and the correct answer path generated based on the sensor data, the process of constructing a training dataset is performed, The above edge cases consist of a total of 118 scenarios, and The device is characterized in that the above 118 scenarios consist of multiple scenarios set for each of six types of situations: cognitive error, software error, hardware error, weather limit, road driving limit, and risk judgment limit.
In Paragraph 7, The above correct path is a device that is either a semi-automatic path or a manual path.
In paragraph 8, The above semi-automatic path is a device, which is an estimated driving path generated by applying a rule algorithm to the sensor data.
In Paragraph 9, The above rule algorithm is a device that is Dijkstra's algorithm or ASTAR's algorithm.
In paragraph 8, The above manual path is a device that is the actual driving path of the autonomous vehicle recorded in the sensor data.
In Paragraph 7, The above training dataset includes a plurality of sensor data and a plurality of correct paths for a plurality of edge cases, and A device comprising a process for constructing the above-mentioned training dataset, wherein the process includes labeling the above-mentioned correct path for a specific edge case onto the sensor data for the specific edge case.
A computer program stored on a computer-readable recording medium to execute each process included in the method according to any one of paragraphs 1 through 6.

Description

Method and Apparatus for Generating Training Dataset for Driving Path Generation of Autonomous Vehicles in Edge Case Scenarios The present disclosure relates to a method and apparatus for generating a training dataset for generating a driving path of an autonomous vehicle in edge-case situations. The following description merely provides background information related to the present embodiment and does not constitute prior art. Collecting edge-case data is necessary to enhance the safety of autonomous vehicles. Edge cases refer to situations where an autonomous driving system encounters exceptional circumstances, and autonomous vehicles must learn from this data to respond appropriately to dangerous situations. Conventionally, devices that collect data based on impact, such as black boxes, have been used for edge case data collection. Existing devices store data even from minor impacts, leading to the accumulation of unnecessary data. Consequently, there was a limitation in that data had to be manually selected by humans. This method of manual data selection by humans can cause errors and may lead to data bias and overfitting. Therefore, there is a need for a method and device that can automatically collect data before and after the occurrence of an edge case using a trigger signal when an edge case occurs, and configure this data as training data for an artificial intelligence model. FIG. 1 is a block diagram showing a device and an artificial intelligence model according to one embodiment of the present disclosure. FIG. 2 is a diagram illustrating the process of a learning data generation module generating a learning dataset according to one embodiment of the present disclosure. FIG. 3a is a diagram illustrating the process of a learning data generation module generating a semi-automatic path according to one embodiment of the present disclosure. FIG. 3b is a diagram illustrating the process of a learning data generation module generating a manual path according to one embodiment of the present disclosure. FIG. 4 is a diagram illustrating the process of a training module training an artificial intelligence model according to one embodiment of the present disclosure. FIG. 5 is a diagram illustrating the process of an evaluation module evaluating an artificial intelligence model according to one embodiment of the present disclosure. FIG. 6 is a flowchart schematically illustrating a method for generating a training dataset according to one embodiment of the present disclosure. FIG. 7 is a schematic block diagram of an exemplary computing device that can be used to implement the devices and methods described in the present disclosure. Some embodiments of the present disclosure are described in detail below with reference to exemplary drawings. It should be noted that in assigning reference numerals to the components of each drawing, the same components are given the same reference numeral whenever possible, even if they are shown in different drawings. Furthermore, in describing the present disclosure, if it is determined that a detailed description of related known components or functions could obscure the essence of the present disclosure, such detailed description is omitted. In describing the components of the embodiments according to the present disclosure, symbols such as first, second, i), ii), a), b), etc., may be used. These symbols are intended only to distinguish the components from other components, and the essence, order, or sequence of the components is not limited by the symbols. When a part in the specification is described as 'comprising' or 'having' a component, this means that, unless explicitly stated otherwise, it does not exclude other components but may include additional components. The detailed description set forth below, together with the accompanying drawings, is intended to describe exemplary embodiments of the present disclosure and is not intended to represent the only embodiment in which the present disclosure can be practiced. In this disclosure, 'learning' and 'training' are used interchangeably with the same meaning. In this disclosure, 'inference' and 'prediction' are used interchangeably with the same meaning. FIG. 1 is a block diagram showing a device and an artificial intelligence model according to one embodiment of the present disclosure. Referring to FIG. 1, a device (10) and an artificial intelligence model (11) are illustrated. The device (10) may be implemented using one or more computing devices (70). The device (10) may include a monitoring module (100), a learning data generation module (102), a training module (104), and an evaluation module (106). The monitoring module (100) can acquire sensor data collected using sensors attached to the autonomous vehicle. The sensors may include a camera, lidar, radar, speed sensor, acceleration sensor, angular velocity sensor, or geomagnetic sensor. Sensor data may be collected at regular time intervals. The monitoring mod