SIGHT-LANGUAGE-ACTION COMPATIBLE SEMANTIC HD MAP LAYER FOR PREFERENCE-RESPONSIVE AUTONOMOUS ROUTE PLANNING
Patent Information
- Application Number
- TR202614549
- Authority / Receiving Office
- TR · TR
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2026-08-26
- Publication Date
- 2026-09-21
Smart Images
Figure 00000020_0000 
Figure 00000021_0000
Abstract
Description
1 TARIFF PREFERENCE-RESPONSIVE AUTONOMOUS ROUTE PLANNING FOR VISUAL-LANGUAGE- ACTION-COMPATIBLE SEMANTIC HD MAP LAYER Technical Area The invention relates to autonomous driving systems, semi-autonomous driving systems, and intelligent transportation. systems, navigation systems, high-resolution (HD) maps and artificial intelligence It falls under the field of intelligence-assisted route planning techniques. 10 More specifically, the invention involves the semantic properties of HD map data. enrichment, a graphic-based approach to the semantic map data in question. structuring it within a framework, expressing driver preferences in natural language Converting it into a machine-processable preference vector, preference vector 15 Creating candidate routes using semantic map data and candidate the combined evaluation of visual and structured route data of routes using a Vision-Language-Action (VLA) model It is related to the route selection process. The invention has implications for automotive, autonomous vehicles, intelligent mobility, navigation, and transportation systems. various land vehicles used primarily for route planning It can be applied in various applications. State of the Art 25 Route planning processes in autonomous and semi-autonomous land vehicles generally using digital map data, HD maps, or road network data This is implemented in systems where the path network consists of nodes and edges. It is represented as a graphic-based structure, and the starting point and the target point are 30. One or more routes are established between them. 2 The routes created were determined by factors such as distance, estimated travel time, traffic density, and energy. based on criteria such as consumption, safety or road accessibility This is being evaluated. In this context, chart-based routes such as Dijkstra, A* and similar are considered. Planning algorithms can be used. Known HD map structures include road geometry, lane information, speed limits, and traffic. physical and structural information such as signs, road connections and topological relationships This information is available. Route planning is mostly done based on this information. It is based on optimizing the defined cost function. However, current route planning approaches prioritize driving comfort and driving experience. stress, intersection complexity, pedestrian exposure, environmental context, specific areas semantic factors such as associated risks and the driver's personal expectations regarding the route Its direct and integrated inclusion in the cost is limited. For example, a user might use natural language to find a more comfortable route with fewer intersections. a route that includes, a lower-stress route or away from high-pedestrian areas They may request a route that is as long as possible. However, the classic route In planning systems, these types of natural language inputs can be directly incorporated into graphic-based routing systems. Converting them to costs is not possible or remains limited. 20 In addition, current systems process map data using user-provided information. Natural language inputs and the visual / semantic representation of the environment are mostly presented as separate data. It considers them as resources. Therefore, graphic-based route data and visual information are used. Elements in the map representation include a machine-trackable identifier. 25 the creation of matching and the use of multimodal artificial intelligence models for this matching. technical limitations regarding its use in route evaluation It is located. Semantic map generation systems provide navigation that takes driver preferences into account. 30 AI-based solutions that can interact with systems and natural language are known. 3 However, semantic HD map data offers a graphic-based semantic path. structuring it as a graph, static semantic HD map from this data. Creating tiles, generating preference vectors from natural language-based preferences, route-centric visual and structured data relating to candidate routes creation, identity mapping between visual elements and graphic data 5 and the data in question are analyzed together by the VLA model. integrated by being evaluated and subjected to rule-based validation. The structure is not included in existing technical solutions. Therefore, driver preferences, semantic HD map information, and multi-mode artificial intelligence (IMI) are taken into account. integrating intelligence-based route evaluation into a common technical processing chain An improved route planning system is needed. Purpose of the Invention The purpose of the invention is to provide a solution to the technical problems mentioned above. The goal is to provide a preference-sensitive route planning system and method. One aim of the invention is to enable the machine to interpret driver preferences expressed in natural language. route planning that enables conversion into a processable preference vector 20 to provide the infrastructure. Another aim of the invention is to interpret classical geometric and topological map data semantically. The aim is to create a semantic pathway by enriching it with features. Another aim of the invention is to analyze semantics at the node, edge, and region levels. The goal is to ensure that features are included in the route cost function. Another aim of the invention is to integrate the route planning process with pre-established semantics. Route evaluation 30 is enabled by using HD map tiles. The goal is to reduce the workload at that stage. 4 Another aim of the invention is to provide visual and structured data on candidate routes. The goal is to ensure they are created simultaneously. Another aim of the invention is the visual within the route-centered semantic visual. Elements and node, edge, region, and route data in the semantic path 5 The goal is to perform identity matching between them. Another aim of the invention is to structure routes with semantic visualizations centered on routes. The aim is to enable the data to be evaluated together by a VLA model. Another objective of the invention is to ensure that the route result suggested by the VLA model is used in traffic. rules, access restrictions, security conditions and road validity rules The aim is to ensure verification based on a solid foundation. Another aim of the invention is to make the route selection decision understandable to the driver and 15 allowing them to be created in a way that can be associated with their preferences The aim is to provide a verifiable route evaluation infrastructure. The structural and characteristic features and all the advantages of the invention are given below. detailed explanation written with figures and references to these figures 20 This will make it clearer, and therefore the evaluation will also be based on this. This should be done taking into account the figures and detailed explanations. Explaining the Figures The best way to structure the current invention and its advantages with additional elements. In order for it to be understood, the figures explained below are included. It needs to be evaluated. Figure 1. General system architecture of the preference-sensitive autonomous route planning system. It shows. Figure 2 shows a schematic representation of the semantic path graph data structure. The parts in the figures are individually numbered, and these numbers correspond to: given below. 5 The drawings do not necessarily need to be scaled, and the existing invention... Details that are not necessary for understanding may have been omitted. Therefore... other, at least substantially identical or at least substantially identical Elements with functions are indicated by the same number. 10 Reference numbers 1. OSM / HD map data source 2. Semantic path generation module 15 3. Semantic path diagram 4. Semantic HD map tiles 5. Driver preferences 6. Preference vector 7. Preference-sensitive route planning module 20 8. Route-centered semantic visualization 9. Structured route data 10. VLA model 11. Rule-based validation module 12. Final route selection 25 13. Node data 14. Edge data Region 15 data 16. Route data 17. Image matching data 30 6 Node number 18. 19. Node location information Edges attached to the 20th knot 21. Return complexity 22. Pedestrian exposure 5 23. Node stress score 24. Node security score 25. Edge number 26. Edge start node 27. Edge end node 10 28. Side length 29. Edge speed limit 30. Road type 31. Edge comfort score 32. Edge stress score 15 33. Edge safety score 34. Fuel / efficiency score District number 35 Area 36 type 37. Geometric area 20 38. Related nodes 39. Associated edges 40. Risk / importance intensity 41. Environmental context 42. Time-dependent effect 25 43. Route tag 44. Candidate route sequence 45. Edge array 46. Start-destination nodes 47. Total distance 30 48. Estimated time 7 49. Average comfort score 50. Average stress score 51. Average security score 52. Fuel / efficiency score 53. Visual object number 5 54. Visual object type 55. Associated node number 56. Associated edge number 57. Associated region number 58. Associated route number 10 59. Visual meaning explanation Detailed Description of the Invention The invention utilizes semantically enriched HD map data and natural language-based 15 driver preferences and the Vision-Language-Action (VLA) model a system and method that performs preference-sensitive route planning using It is related to the system. The general architecture of the system is shown in Figure 1, and the system is OSM / HD. map data source (1), semantic path generation module (2), semantic path graph (3), semantic HD map tiles (4), driver preferences (5), preference vector 20 (6), preference-sensitive route planning module (7), route-centric semantic visualization (8), Structured route data (9), VLA model (10), rule-based validation module It includes components (11) and final route selection (12). In one application of the invention, the first route 25 is obtained from the OSM / HD map data source (1). Map data relating to the network is being collected. This map data includes: There is node data (13) and edge data (14). Node data (13) is the node number (18), node location information (19) and edges connected to the node (20) Edge data (14) includes edge number (25), starting node (26), end node (27), length (28), speed limit (29) and road type (30) information 30 It includes. 8 The received map data is entered into the semantic path graph creation module (2) It is transferred. The semantic path graph creation module (2), geometric and By combining topological map data with semantic features, a semantic path graph can be created. (3) Thus, the road network represents only physical connections. 5 semantic factors that can influence route planning decisions by moving away from being purely graphic. It is transformed into a data structure that also possesses these characteristics. In the semantic path diagram (3), node data (13), edge data (14), region data (15), route data (16) and visual matching data (17) are available. Node data (13), return complexity in addition to classical node information (21), pedestrian 10 semantic features such as exposure (22), stress score (23) and safety score (24) It includes. Return complexity (21), number of edges connected to the node, input and output Directions, turning angles, sharp turns, intersection geometry and intersection control 15 It can be derived from features such as elements. Pedestrian exposure (22), knot pedestrian crossings, sidewalk connections, schools, hospitals, bus stops, parks in the surrounding area, taking into account factors such as shopping areas or pedestrian zones Node stress score (23) and node safety score (24) can be determined. the subject is the predetermined evaluation of semantic and geometric features 20 by combining them through rules or learned models can be created. Edge data (14), in addition to basic road characteristics, comfort score (31), stress This includes the score (32), safety score (33) and fuel / efficiency score (34). These 25 Scores are based on factors such as road width, number of lanes, road type, road gradient, speed limit, and intersection density. traffic control elements, road surface, stop-and-go traffic conditions, and similar features. It can be calculated using [method]. Regional data (15) is directly represented by a single node or edge on the road network. 30 It defines the environmental semantic areas that cannot be identified. Regional data (15); region 9 number (35), region type (36), geometric area (37), associated nodes (38), associated edges (39), risk / importance intensity (40), environmental context (41) and time dependence It includes the impact (42) information. District type (36), e.g. school district, hospital district, park district, shopping district 5 refers to an area, pedestrian zone, construction zone, or similar environmental zone. Time-dependent effect (42) is able to affect a particular region at certain times of the day. differentiating route costs during specific hours or time intervals It provides opportunities. Route data (16) represents the characteristics of candidate routes generated by the system. Route data (16); route label (43), candidate route sequence (44), edge sequence (45), starting-destination nodes (46), total distance (47), estimated time (48), average comfort score (49), average stress score (50), average safety score (51) and fuel / efficiency score (52). 15 Visual matching data (17) is structural within the semantic path diagram (3). data and their visual representations in route-centered semantic visual (8) It enables the establishment of a relationship between them. In this context, visual matching data (17); visual object number (53), visual object type (54), associated node number 20 (55), associated edge number (56), associated zone number (57), associated route It may include the number (58) and visual meaning description (59). After the semantic path diagram (3) is created, the following are located in this diagram Semantic HD map tiles (4) are created using semantic information. 25 The map area is converted to metric coordinates by performing an appropriate coordinate transformation. It is converted into a system and divided into square tiles of specific dimensions. Each A tile has nodes (13), edges (14) and within its geographical boundaries. It is created as an independent map unit containing regions (15). 10 A two-dimensional visual of road geometries on semantic HD map tiles (4). The properties of nodes, edges, and regions are transferred to a plane and visually represented. For example, different safety, stress, or comfort scores are visualized differently. They can be represented by their display characteristics. Thus, the map tiles in question... (4), 5 in advance which can be used in subsequent route evaluation processes. It provides a processed semantic visual data layer. The next stage of the invention is driver expression in natural language by the driver. preferences (5) are processed. Driver preference (5), for example a more comfortable route, A route with fewer junctions, a less stressful route, and lower energy consumption. expressed as a route or a route that avoids certain environmental areas It is possible. Driver preferences expressed in natural language (5), a natural language processing mechanism It is analyzed by and converted into a preference vector (6). The preference vector is 15 (6), weighting coefficients of the criteria to be used in route evaluation These criteria include comfort, travel time, energy efficiency, and traffic. traffic density, intersection complexity, safety, stress, or a combination of these It may include. Preference vector (6) and semantic HD map tiles (4), preference-sensitive route planning It is transferred to module (7). Preference-sensitive route planning module (7), Dijkstra, Using one of the A* or similar graphical route planning algorithms one or more candidate routes between the starting and destination points It constitutes. 25 A key feature of the invention is the route planning algorithm itself. Rather than changing the cost structure presented to the algorithm, the semantic path... using the node, edge and region properties in graph (3) It is the creation of. 30 11 In this context, the total transit cost for one edge is as follows: It can be calculated as follows: M(eᵢⱼ) = M_basic(eᵢⱼ) + M_semantic(eᵢⱼ) + M_node(vⱼ) + M_region(eᵢⱼ) Here, M(eᵢⱼ) is the total transit cost, and M_basic(eᵢⱼ) is the basic edge cost. M_semantic(eᵢⱼ) represents the semantic edge cost, and M_node(vⱼ) represents the cost related to the target node. M_region(eᵢⱼ) represents the transit cost and M_region(eᵢⱼ) represents the regional impact cost. Basic edge cost; edge length (28), estimated transition time and accessibility 10 It can be determined depending on the situation. The semantic edge cost, on the other hand, is comfort. score (31), stress score (32), safety score (33) and fuel / efficiency score (34) It can be calculated using [method]. Node crossing cost; turn complexity (21), pedestrian exposure (22), node 15 It can be calculated via stress score (23) and node security score (24). The area impact cost (M_area) is the cost incurred by the area through which the edge passes or within a certain distance. risk / importance density of the semantic regions in which it is located (40), environmental context (41) and its time-dependent effect (42) can be determined by taking into account. The weights of these cost components in the route planning process are preferred. The vector is determined by (6). Thus, the same starting and target points different candidate route sets in line with different driver preferences (5) can be created. Route-centered semantic visualization of the generated candidate routes (8) This is being done by creating starting points, destination points, and candidate routes. The geographical area it covers is determined, and the semantic HD that intersects with that area is identified. Map tiles (4) are selected. The selected tiles are combined to form candidate routes. A route-centered visual map area suitable for evaluation is being created. 30 12 Route-centered semantic visual (8), pre-generated static semantic HD map tiles (4), visual overlays of candidate routes and driver preferences visual information regarding the semantic features highlighted in this context It may include. Structured route data (9) is generated for the same candidate routes. Structured route data (9), route IDs, node arrays, edge arrays, JSON, XML, or similar machine learning formats can include semantic scores and route metrics. It can be created in a data format that can be processed by [the system / organization]. At this stage, route-centric visual matching data (17) is also used. semantic visual (8) with structured route data (9) linking node, edge, region, and route objects within it. This is done. Thus, a visual element of the VLA model (10) is only not through its visual content, but through the semantic path of the element in question (3) 15 It is also evaluated based on its structural equivalent. Route-centric semantic visualization (8), structured route data (9) and visualization Matching data (17) is transferred to the VLA model (10). The VLA model (10) is used to define the word The subject is to evaluate multimodal inputs together to determine candidate routes for drivers. 20 It analyzes its compatibility with their preferences. VLA model (10) candidate routes only such as total distance or estimated time not through numerical metrics, but through the semantic environment represented visually. It can evaluate structured route features by considering them together. 25 The route result determined by the VLA model (10) is then rule-based It is transferred to the validation module (11). Rule-based validation module (11), The designated route must be followed considering traffic regulations, access restrictions, road validity, and safety. It checks against requirements and other predefined constraints. 30 13 The route found to be compliant with the rules as a result of the verification is the final route selection (12). route validation recommended by the VLA model (10) is determined. The route may be rejected if it does not meet one or more of the rules. It can be reconsidered or a suitable alternative candidate route can be chosen. A route can be selected. 5 Thanks to this structure, the route planning process includes; acquiring raw map data, semantic analysis route plot creation, static semantic HD map layer creation, Converting natural language-based preferences to numerical preferences, preference-sensitive candidates. route creation, creation of visual and structured route data, 10 visual-structural identity mapping, VLA-based multimodal evaluation, and rule-making. an integrated technical process chain that includes verification stages based on data processing. is being brought. In this way, driver preferences directly contribute to the route planning cost function. transfer, inclusion of semantic environmental information in route cost, candidate Evaluation of routes by multimodal artificial intelligence and artificial intelligence Rule-based verification of route selection performed by is provided. 25
Claims
14 REQUESTS 1. Preference-aware route planning in autonomous or semi-autonomous land vehicles. It is a route planning system for implementation, and its feature is; OSM Map 5 taken from (OpenStreetMap) and / or HD map data source (1) by processing data to extract semantic features at the node, edge, and region levels. semantic path that forms a semantic path containing (3) creation module (2); pre-processed from semantic path diagram (3) Semantic HD map tiles (4) containing semantic map data a map tile structure; driver preferences expressed in natural language (5) route 10 preference including weighting coefficients for evaluation criteria a preference processing structure that converts to its vector (6); preference vector (6) and by using the semantic features of the semantic path (3) one or more Preference-sensitive route planning module that creates multiple candidate routes (7); word Route-centered semantic visuals (8), 15 regarding candidate routes structured route data (9) and visual elements and route diagram objects visual matching data that provides identity matching between (17) a data production structure that creates a route-centric semantic visualization (8), structured route data (9) and visual matching data (17) together by processing candidate routes in terms of suitability for driver preferences 20 a VLA (Vision-Language-Action) model (10) that evaluates VLA predefined route result determined by model (10) rule-based systems that monitor traffic, access, and security regulations. verification module (11) and final route selection based on the verification result (12) is characterized by containing a selection structure that constitutes 25 2. According to Claim 1, it is a route planning system, the characteristic of which is; a semantic path diagram. (3), node data (13), edge data (14), region data (15), route data (16) and includes visual matching data (17). 15 3. Route planning system according to claim 2, and its feature is; node data (13), node number (18), location information (19), edges connected to the node (20), turn complexity (21), pedestrian exposure (22), node stress score (23) and node security score (24) is included.
4. It is a route planning system according to claim 2, and its feature is; edge data (14), edge number (25), start node (26), end node (27), length (28), speed limit (29), road type (30), comfort score (31), stress score (32), It includes a safety score (33) and a fuel / efficiency score (34).
5. Route planning system according to claim 2, its feature is; regional data (15), Zone number (35), Zone type (36), Geometric area (37), Associated nodes (38), related edges (39), risk / importance density (40), environmental context (41) and time-dependent effect (42).
6. According to claim 2, it is a route planning system and its feature is that route data (16), route label (43), candidate route sequence (44), edge sequence (45), start-target nodes (46), total distance (47), estimated time (48), average comfort score (49), mean stress score (50), mean safety score (51) and It includes a fuel / efficiency score (52). 20 7. Route planning system according to Claim 2, its feature is visual matching. data (17), visual object number (53), visual object type (54), related node number (55), associated edge number (56), associated region number (57), associated route number (58) and visual meaning description (59) 25 It includes.
8. Route planning system according to Claim 1, its feature is preference-sensitive routing. planning module (7), semantic path during candidate route creation node data (13), edge data (14) and region 30 in the graph (3) Using the data (15), the total transition cost for each edge is calculated. Calculation and calculated total toll cost on a graphical route. It provides input to the planning algorithm. 16 9. Route planning system according to claim 8, its feature is; total crossings. cost of basic edge cost, semantic edge cost, node transition by considering the cost and the regional impact cost together It is the calculation. 5 10. According to claim 9, it is a route planning system, the characteristic of which is; the cost in question. obtained from the driver preferences expressed in natural language of the components (5) It is weighting with the weighting coefficients in the preference vector (6).
11. It is a route planning system according to claim 1, and its feature is that the preference vector (6), comfort, travel time, energy efficiency, traffic congestion, intersection weighting for at least one of the following criteria: complexity, safety, and stress. It includes a coefficient.
12. It is a route planning system according to Claim 1, and its feature is preference-sensitive routing. planning module (7), Dijkstra, A* or another graphic-based route It involves generating candidate routes using a planning algorithm. Route planning system according to Claim 13, its feature is; semantic HD map 20 The nodes, edges and tiles (4) obtained from the semantic path diagram (3) Pre-processed visual map of semantic features at the regional level It is created in the form of layers.
14. It is a route planning system according to Claim 1, and its characteristic is; route-centered semantic 25 The semantic HD of the image (8) intersects with the geographic area covered by the candidate routes. Combining the map tiles (4) and candidate route information It is created by overlaying a combined map onto another map.
15. Route planning system according to Claim 1, its feature is; structured route 30 (9) of the data, route IDs, node sequences, edge sequences, semantic machine-readable data including scores and route metrics It is created in the format. 17 According to claim 16, it is a route planning system whose feature is visual matching data. (17) through a route-centered semantic visual (8) located within (17) related nodes, edges, within the semantic path of the visual element (3) It is associated with the region and / or route object. 5 17. Route planning system according to claim 16, its feature is; VLA model (10), visual elements within the route-centered semantic visual (8), associated with visual elements via visual matching data (17) It is an evaluation together with structured route data (9). 10 According to claim 18, it is a route planning system, and its feature is the VLA model (10), candidate routes driver preference vector (6), route metrics and route-centric in line with the semantic environment information in the semantic visual (8) It is a comparative evaluation. 15 According to Claim 19, it is a route planning system, characterized by rule-based validation. the route result determined by the module (11) and the VLA model (10) traffic regulations, access restrictions, safety requirements, and road validity It is an audit in terms of at least one of the criteria. 20 20. Route planning system according to claim 19, its feature is; VLA model (10) at least one of the verification criteria for the route result determined by If it does not meet the requirements, the route in question will be rejected. reassessment or alternative route from the candidate routes 25 It is the selection process.
21. Preference-aware route planning in autonomous or semi-autonomous land vehicles. It is a computer-led method for accomplishing something, Feature; Retrieval of map data from OSM / HD map data source (1); 30 Semantic path creation module for acquired map data (2) processed semantically at the node, edge, and region levels. creation of a semantic pathway diagram (3) containing features; semantic pathway 18 from the (3) line, the creation of semantic HD map tiles (4); natural by processing the driver preferences expressed in language (5) and the preference vector (6) creation; using preference vector (6) and semantic path graph (3) Creation of candidate routes; route-centered semantic analysis of candidate routes. image (8), structured route data (9) and image matching 5 creation of data (17); route-centered semantic visual (8), VLA of structured route data (9) and visual matching data (17) joint evaluation by model (10); VLA model (10) Rule-based validation module of route result determined by (11) Verification by and according to the verified route result, the final route is 10 (12) includes the steps of forming the selection.
22. According to claim 21, it is a route planning method, characterized by its semantic path. During the creation of the graph (3), node data (13), edge data (14), region data (15), route data (16) and visual matching data (17) 15 is the creation of.
23. Route planning method according to claim 21 or 22, characterized by; candidate route node data (13), edge data (14) and region data during creation (15) The total transition cost for each edge is 20 It is the calculation.
24. Route planning method according to claim 23, characterized by; total crossings. cost of basic edge cost, semantic edge cost, node transition by considering the cost and regional impact cost together 25 It is the calculation.
25. This is a route planning method according to claim 24, and its characteristic feature is; basic edge cost, semantic edge cost, node transition cost, and region impact cost with the weighting coefficients in the preference vector (6) 30 It is weighting. 19 26. This is a route planning method according to claim 21, characterized by its route-centric nature. semantic visual (8) intersecting with the geographic area covered by the candidate routes Selection, combination and candidate of semantic HD map tiles (4) The routes are created by overlaying them onto the map.
27. This is a route planning method according to claim 21, and its characteristic is visual matching. Using data (17), route-centered semantic visualization (8) node in route data structured with visual elements (9), It involves matching edge, region, and route information.
28. Route planning method according to claim 21, its feature is; VLA model (10) route-centered semantic visualization of candidate routes (8), structured route data (9), visual matching data (17) and preference vector (6) together It is evaluated using [method / technology].
29. Route planning method according to claim 21, characterized by its rule-based nature. VLA model (10) determined by the verification module (11) the route outcome traffic regulations, access restrictions, safety requirements and the route is checked for validity and the route result is found to be acceptable. The final route selection is determined as (12). 20