CN-120027805-B - Lane attribute creation system, lane attribute creation method, and computer program product

CN120027805BCN 120027805 BCN120027805 BCN 120027805BCN-120027805-B

Abstract

The present disclosure discloses a lane attribute making system, a making method and a computer program product. The lane attribute making system comprises a map element encoder and a large language model, wherein an output layer of the map element encoder is connected to an input layer of the large language model, the map element encoder is used for processing a vectorized map to output vector coding results, the vectorized map uses vector features to represent map elements, the map elements comprise lanes, the input layer of the large language model also receives at least one of image coding results and text coding results, and the large language model is configured to generate lane attribute data according to the vector coding results and at least one of the image coding results and the text coding results. The scheme of the embodiment of the disclosure can process data of multiple modes including vector modes so as to match the driving rules to corresponding lanes, so as to provide accurate and detailed lane attribute data for constructing a traffic rule layer.

Inventors

CHANG XINYUAN
LIU XINRAN
Xue Maixuan

Assignees

北京高德云图科技有限公司

Dates

Publication Date: 20260508
Application Date: 20241227

Claims (8)

1. A lane attribute making system is characterized by comprising a map element encoder and a large language model; an output layer of the map element encoder is connected to an input layer of the large language model, the map element encoder is used for processing a vectorized map to output a vector coding result, the vectorized map uses vector characteristics to represent map elements, and the map elements comprise lanes; The input layer of the large language model also receives at least one of image encoding results and text encoding results, the large language model configured to: Extracting driving rule features according to at least one of the image coding result and the text coding result, extracting lane information features according to the vector coding result, and matching the driving rule features with the lane information features to generate lane attribute data, wherein the lane attribute data comprises driving rules of vehicles and corresponding relations between the driving rules and lanes.
2. The lane attribute making system of claim 1 wherein the map element encoder includes an embedding layer, a first vector encoding block and a second vector encoding block connected in sequence, the embedding layer for converting vector features of lanes in the vectorized map from vector representations to embedded representations, the first vector encoding block and the second vector encoding block for encoding vector features of the embedded representations into vector encoding results containing lane information.
3. The lane attribute making system of claim 2 wherein the embedding layer is configured to perform an embedding transformation on points on the vector features of each lane in the vectorized map to generate a vector embedding, type embedding, instance embedding, and location embedding describing the vector features of each lane, and to set blank marker embedding at the beginning of the vector features of each lane in the vectorized map, the marker embedding, the vector embedding, the type embedding, the instance embedding, and the location embedding being aggregated to form an embedded representation of the vector features of the lane; wherein the vector embeddings are used for representing positions of points on the vector features, the type embeddings are used for representing map elements represented by the vector features where the points are located, the instance embeddings are used for indicating the vector features where the points are located, the position embeddings are used for representing relative positions between the points, and the marker embeddings are used as indexes of each vector feature.
4. The lane attribute making system of claim 3 wherein the first vector encoding block includes M first fransformer layers in series, M being a positive integer, the first fransformer layers employing an intra-instance attention mechanism, the M first fransformer layers in series for receiving an embedded representation of vector features of a lane and assigning values to blank tag embeddings in accordance with the vector embeddings, the type embeddings, the instance embeddings and the location embeddings; The second vector coding block comprises N serially connected second converter layers, N is a positive integer, the second converter layers adopt an inter-instance attention mechanism, and the N serially connected second converter layers are used for carrying out secondary assignment on the mark embedding according to the vector embedding, the type embedding, the instance embedding and the position embedding to form mark embedding which corresponds to the vector features one by one and serve as indexes of each vector feature so as to obtain a vector coding result containing lane information.
5. The lane attribute making system according to claim 1, wherein the lane attribute data output by the large language model is a running rule in a text format; The lane attribute making system further comprises a JSON decoder, wherein an input layer is connected to an output layer of the large language model so as to restore the lane attribute data from a text format to a JSON format.
6. A lane attribute making method is characterized by being applied to a lane attribute making system, wherein the lane attribute making system comprises a map element encoder and a large language model, an output layer of the map element encoder is connected to an input layer of the large language model, and the lane attribute making method comprises the following steps: receiving multi-modal data, wherein the multi-modal data comprises a vectorized map and at least one of an image coding result and a text coding result; Processing the vectorized map with the map element encoder to obtain a vector encoding result, wherein the vectorized map uses vector features to represent map elements including lanes, and And processing the lane information characteristics extracted according to at least one of the image coding result and the text coding result, the extracted driving rule characteristics and the vector coding result by utilizing the large language model, and matching the driving rule characteristics with the lane information characteristics to generate lane attribute data, wherein the lane attribute data comprises driving rules of vehicles and corresponding relations between the driving rules and lanes.
7. The lane attribute making method of claim 6 wherein the map element encoder includes an embedded layer, a first vector encoding block, and a second vector encoding block connected in sequence, wherein processing the vectorized map with the map element encoder to obtain a vector encoding result including lane information includes: Converting vector features of lanes in a vectorized map into an embedded representation using the embedding layer, and Encoding the embedded representation of the vector features with the first vector encoding block and the second vector encoding block to form a vector encoding result comprising lane information.
8. A computer program product comprising a computer program, characterized in that the computer program realizes the steps of the method of claim 6 or 7 when being executed by a processor.

Description

Lane attribute creation system, lane attribute creation method, and computer program product Technical Field The present disclosure relates generally to the field of map technology. More particularly, the present disclosure relates to a lane attribute making system, a lane attribute making method, and a computer program product. Background With the explosive development of artificial intelligence technology, 5G communication technology and the like, automated driving and intelligent transportation systems have also progressed to the application stage. The rapid development of autopilot and intelligent transportation systems places higher demands on the reliability and accuracy of navigation data, which requires accurate navigation data to achieve perception, positioning, path planning and decision control of the vehicle. The High Definition (HD) map is an important component for supporting the systems by the detailed representation of road elements, the geometric layer of the HD map provides information such as lane separation lines, lane center lines and the like for the systems, the connection layer provides lane relations for the systems so as to facilitate the systems to make path planning, and the traffic rule layer provides rule information related to lanes for the systems to make decision control. However, the HD map currently applied to the automatic driving system performs well on the geometric layer and the connection layer, but has the defects that firstly, the constructed traffic rule layer only describes the direction marks of the lanes, such as straight lanes, left-turn lanes and/or right-turn lanes, but more types of traffic marks and corresponding driving rules thereof, such as bus lanes and/or speed limit areas, exist in the actual driving scene of the vehicle, and secondly, even some related technologies can recognize the traffic marks except the direction marks, can only recognize the types of the traffic marks, but cannot form detailed rules required by automatic driving according to the traffic marks and match the traffic marks to the corresponding lanes, so as to be used as decision basis of automatic driving. In view of this, there is a need to provide a lane attribute making scheme to meet the requirements of the automatic driving scene on the map data. Disclosure of Invention To address at least one or more of the technical problems mentioned above, the present disclosure proposes lane attribute making schemes in various aspects. In a first aspect, the present disclosure provides a lane attribute making system including a map element encoder and a large language model, an output layer of the map element encoder being connected to an input layer of the large language model, the map element encoder for processing a vectorized map to output a vector encoding result, the vectorized map representing map elements using vector features, the map elements including lanes, the input layer of the large language model further receiving at least one of an image encoding result and a text encoding result, the large language model being configured to generate lane attribute data from the vector encoding result and at least one of the image encoding result and the text encoding result, the lane attribute data including a driving rule of a vehicle and a correspondence of the driving rule to the lanes. In a second aspect, the present disclosure provides a lane attribute making method applied to a lane attribute making system including a map element encoder and a large language model, an output layer of the map element encoder being connected to an input layer of the large language model, the lane attribute making method including receiving multi-modal data including a vectorized map and at least one of an image encoding result and a text encoding result, processing the vectorized map with the map element encoder to obtain the vector encoding result, wherein the vectorized map represents map elements using vector features, the map elements including lanes, and processing the vector encoding result and at least one of the image encoding result and the text encoding result with the large language model to generate lane attribute data, wherein the lane attribute data includes a driving rule of a vehicle and a correspondence of the driving rule and the lanes. In a third aspect, the present disclosure provides a computer program product comprising a computer program which, when executed by a processor, implements the method steps as in the second aspect. Through the lane attribute making system provided by the above, the embodiment of the disclosure encodes the map data of the vector mode provided by the vector map through the map element encoder, for example, encodes the vector features of the lanes, so as to obtain a vector encoding result, and processes the vector encoding result and the encoding result of other mode data through the large language model with excellent performance in acquiring implicit