CN-122020614-A - Action space dictionary generation method, intelligent device and readable storage medium

CN122020614ACN 122020614 ACN122020614 ACN 122020614ACN-122020614-A

Abstract

The application relates to the technical field of computers, in particular to a method for generating an action space dictionary, intelligent equipment and a readable storage medium, aiming at solving the technical problem of how to ensure that the coverage, robustness and utilization rate of the generated action space dictionary are considered. Therefore, the method and the system acquire the track data of multiple intelligent agents, carry out meshing division according to the track data, acquire the grid map corresponding to the track data, carry out grid screening on each grid of the grid map, acquire the action space dictionary according to the grid screening result, effectively improve the theoretical coverage of the action space, effectively remove invalid grids, and reserve high-probability tracks, thereby effectively improving the utilization rate of the acquired action space, further effectively improving the robustness of the acquired action space dictionary, and further providing an effective data basis for automatic driving model training, simulation, algorithm evaluation and other scenes.

Inventors

REN SHAOQING
WU KUANG
CHENG ZHENGXIN
LIU GUOYI
LIN WEI
WANG CHENGFA

Assignees

安徽蔚来智驾科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260415

Claims (10)

1. A method of generating an action space dictionary, the method comprising: Acquiring track data of multiple intelligent agents, wherein the track data comprises multiple tracks; According to the track data, meshing division is carried out, and a grid map corresponding to the track data is generated; aiming at each grid of the grid map, carrying out grid screening on the grids according to track conditions in surrounding grids of the grids to obtain grid screening results; And acquiring the action space dictionary according to the grid corresponding to the grid screening result and track data contained in the grid.
2. The method of generating an action space dictionary according to claim 1, wherein, The meshing division is performed according to the track data, including: acquiring an action set corresponding to each track according to the track data; And according to the action set, meshing division is carried out, and a meshing map corresponding to the track data is obtained.
3. The method of generating an action space dictionary according to claim 2, wherein, The step of obtaining the action set corresponding to each track according to the track data comprises the following steps: For each track in the track data, extracting track fragments with equal length according to the track, and converting the track fragments into a coordinate system taking an intelligent body as a center; and differentiating the track segment in the coordinate system taking the intelligent body as the center to acquire the action of the track segment at each moment, thereby acquiring an action set corresponding to the track.
4. The method of generating an action space dictionary according to claim 2, wherein, And according to the action set, performing meshing division to obtain a meshing map corresponding to the track data, including: Constructing a grid with a preset size; And determining the classifying grids of the action set according to the grids of the end points of the tracks corresponding to the action set, thereby obtaining the gridding map of the track data.
5. The method of generating an action space dictionary according to claim 1, wherein, And performing grid screening on the grids according to track conditions in surrounding grids of the grids to obtain grid screening results, wherein the grid screening results comprise: Judging whether tracks do not exist in surrounding grids of the grid if tracks exist in the grids aiming at each grid of the grid map; if yes, eliminating the grids and the tracks in the grids.
6. The method for generating an action space dictionary according to claim 1 or 5, wherein, And performing grid screening on the grids according to track conditions in surrounding grids of the grids to obtain grid screening results, wherein the grid screening results comprise: Judging whether at least one grid track exists in surrounding grids of the grid map or not according to each grid of the grid map if no track exists in the grid; if yes, reserving the grid.
7. The method of generating an action space dictionary according to claim 1, wherein, The step of obtaining the action space dictionary according to the grid corresponding to the grid screening result and the track contained in the grid comprises the following steps: Acquiring an average track of the grid according to the track included in the grid corresponding to the grid screening result; Acquiring a yaw angle of the average track corresponding to the grid according to the average track; and acquiring the action space dictionary according to the average track and the yaw angle.
8. The method of generating an action space dictionary according to claim 7, wherein, The obtaining the average track of the grid according to the track included in the grid corresponding to the grid screening result includes: Acquiring an action set corresponding to a track included in the grid; And acquiring the average track of the grid according to the action set.
9. An intelligent device, comprising: At least one processor; And a memory communicatively coupled to the at least one processor; Wherein the memory has stored therein a computer program which, when executed by the at least one processor, implements the method of action space dictionary generation of any one of claims 1 to 8.
10. A computer readable storage medium, in which a plurality of program codes are stored, characterized in that the program codes are adapted to be loaded and executed by a processor to perform the method of generating an action space dictionary according to any one of claims 1 to 8.

Description

Action space dictionary generation method, intelligent device and readable storage medium Technical Field The application relates to the technical field of computers, in particular to an action space dictionary generating method, intelligent equipment and a readable storage medium. Background In the autopilot trajectory generation direction, the action space dictionary (i.e., the trajectory vocabulary, trajectory Vocabulary) is the core component that supports the autoregressive generation paradigm. The method has the core functions of discretizing continuous tracks (namely corresponding to 'actions') of multiple intelligent agents (such as vehicles, bicycles and pedestrians) in an automatic driving scene into track action units with a limited set, so that complex continuous track distribution is converted into a discrete space capable of being modeled efficiently, generation and fitting of multi-mode driving behaviors are realized, and the method is finally applied to key scenes such as data acquisition, algorithm evaluation and the like of an automatic driving simulator. The current mainstream action space dictionary construction methods in the industry can be divided into two types of data driving type and rule driving type, but both types of methods have obvious technical defects. The data-driven action space dictionary has high utilization rate, but has the defects of robustness and coverage. The rule-driven action space dictionary has wide coverage, but extremely low utilization rate. Accordingly, there is a need in the art for a new action space dictionary generation scheme to address the above-described problems. Disclosure of Invention The present application is proposed to overcome the above drawbacks, and to solve or at least partially solve the technical problem of how to achieve ensuring that the generation of the motion space dictionary is compatible with coverage, robustness and utilization. In a first aspect, there is provided a method of generating an action space dictionary, the method comprising: Acquiring track data of multiple intelligent agents, wherein the track data comprises multiple tracks; According to the track data, meshing division is carried out, and a grid map corresponding to the track data is generated; aiming at each grid of the grid map, carrying out grid screening on the grids according to track conditions in surrounding grids of the grids to obtain grid screening results; And acquiring the action space dictionary according to the grid corresponding to the grid screening result and track data contained in the grid. In one technical scheme of the above method for generating an action space dictionary, the performing meshing division according to the trajectory data includes: acquiring an action set corresponding to each track according to the track data; And according to the action set, meshing division is carried out, and a meshing map corresponding to the track data is obtained. In one technical scheme of the above method for generating an action space dictionary, the acquiring, according to the track data, an action set corresponding to each track includes: For each track in the track data, extracting track fragments with equal length according to the track, and converting the track fragments into a coordinate system taking an intelligent body as a center; and differentiating the track segment in the coordinate system taking the intelligent body as the center to acquire the action of the track segment at each moment, thereby acquiring an action set corresponding to the track. In one technical scheme of the above method for generating an action space dictionary, the performing meshing division according to the action set to obtain a meshing map corresponding to the track data includes: Constructing a grid with a preset size; And determining the classifying grids of the action set according to the grids of the end points of the tracks corresponding to the action set, thereby obtaining the gridding map of the track data. In one technical scheme of the above method for generating an action space dictionary, the step of performing grid screening on the grid according to the track condition in the surrounding grids of the grid to obtain a grid screening result includes: Judging whether tracks do not exist in surrounding grids of the grid if tracks exist in the grids aiming at each grid of the grid map; if yes, eliminating the grids and the tracks in the grids. In one technical scheme of the above method for generating an action space dictionary, the step of performing grid screening on the grid according to the track condition in the surrounding grids of the grid to obtain a grid screening result includes: Judging whether at least one grid track exists in surrounding grids of the grid map or not according to each grid of the grid map if no track exists in the grid; if yes, reserving the grid. In one technical scheme of the above method for generating an action space dictiona