Vehicle positioning based on radar detection

Abstract

This document describes radar detection based vehicle positioning methods and systems. Radar positioning starts with constructing a radar reference map. The radar reference map can be generated and updated using different techniques as described herein. Once the radar reference map is available, real-time positioning can be achieved using inexpensive radar sensors and navigation systems. Using the techniques described in this document, data from radar sensors and navigation systems can be processed to identify stationary positioning objects or landmarks in the vicinity of the vehicle. Comparing landmark data originating from the vehicle's on-board sensors and systems with the detailed landmark data in the radar reference map can generate an accurate pose of the vehicle in its environment. By using inexpensive radar systems and lower quality navigation systems, a highly accurate vehicle pose can be obtained in a cost-effective manner.

Description

[0001] Cross-referencing related applications

[0002] This application claims the benefit of U.S. Provisional Application No. 63 / 146,483, filed February 5, 2021, and U.S. Provisional Application No. 63 / 127,049, filed December 17, 2020, pursuant to 35U.SC119(e), the disclosures of which are incorporated herein by reference in their entirety. Background Technology

[0003] Vehicle localization is a technique that uses sensor data to locate a vehicle on a map (e.g., determine the vehicle's position on the map). Vehicle localization can be used to support autonomous vehicle operation (e.g., navigation, route planning, lane determination and centering, and curve execution in the absence of lane markings). To accurately locate a vehicle relative to its environment, vehicle localization involves acquiring the positions of stationary objects (e.g., signs, poles, obstacles) from various sensors and navigation systems on the vehicle and associating these positions with known positions on a map (e.g., positions within a Uniform Transverse Mercator or UTM frame). The vehicle's position or attitude can then be determined relative to these objects.

[0004] Some autonomous vehicles perform driving maneuvers that depend on positioning accuracy approaching sub-meter levels. Sub-meter accuracy can be achieved through expensive sensors and navigation systems, including optical cameras, light-detection-based ranging systems (such as LiDAR), and high-quality Global Navigation Satellite System (GNSS) receivers. However, cameras and LiDAR may experience reduced performance when operating in less than ideal lighting and weather conditions. High-quality GNSS systems may be prohibitively expensive for most consumer vehicles, which may be integrated with less robust GPS capabilities. Summary of the Invention

[0005] The aspects described below include a vehicle localization method based on radar detection. The method includes: receiving radar detection from one or more radar sensors of the vehicle by at least one processor. The method further includes: receiving navigation data from one or more navigation units of the vehicle by at least one processor. The method further includes: outputting ego trajectory information about the current dynamic state of the vehicle by at least one processor. The method further includes: extracting landmark data from the radar detection and ego trajectory information by at least one processor. The method further includes: determining a normal distribution transformation grid by at least one processor from the extracted landmark data and a radar reference map. The method further includes: in response to determining the normal distribution transformation grid, correcting the vehicle attitude according to the normal distribution transformation grid by at least one processor to localize the vehicle.

[0006] Aspects described below also include a system for enabling vehicle localization based on radar detections. The system includes at least one processor and at least one computer-readable storage medium including instructions that, when executed by the processor, cause the system to receive radar detections from one or more radar sensors. The instructions also cause the system to receive navigation data from one or more navigation units. The instructions further cause the system to output ego trajectory information about a current dynamic state of the vehicle. The instructions further cause the system to extract landmark data from the radar detections and the ego trajectory information. The instructions further cause the system to determine a normal distribution transform grid from the extracted landmark data and a radar reference map. The instructions further cause the system to, in response to determining the normal distribution transform grid, correct a vehicle pose according to the normal distribution transform grid to localize the vehicle.

[0007] BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Systems and techniques for enabling vehicle localization based on radar detections are described with reference to the following drawings. Like numbers may refer to like features throughout the drawings:

[0009] Figure 1 is an example illustration of an environment in which vehicle localization based on radar detections can be implemented according to techniques of this disclosure;

[0010] Figure 2-1 is an example illustration of a system that can be used to implement vehicle localization based on radar detections according to techniques of this disclosure;

[0011] Figure 2-2 is an example illustration of a radar localization module that can be used to implement vehicle localization based on radar detections;

[0012] Figure 2-3 is another example illustration of a radar localization module that can be used to implement vehicle localization based on radar detections;

[0013] Figure 3 is an example illustration of generating a radar reference map according to techniques of this disclosure;

[0014] Figure 4 is an example illustration of generating a radar occupancy grid according to techniques of this disclosure;

[0015] Figure 5 is an example illustration of generating a radar reference map from radar landmarks according to techniques of this disclosure;

[0016] Figure 6 is an example illustration of a process of generating a radar reference map according to techniques of this disclosure;

[0017] Figure 7is an example illustration of generating a radar occupancy grid based on multiple vehicle traversals with low-precision location data, according to the techniques of this disclosure;

[0018] Figure 8 is another example illustration of generating a radar occupancy grid based on multiple vehicle traversals with low-precision location data, according to the techniques of this disclosure;

[0019] Figure 9 is an example illustration of a process of generating a radar occupancy grid based on multiple vehicle traversals with low-precision location data, according to the techniques of this disclosure;

[0020] Figure 10 is an example illustration of generating a radar reference map using low-precision location data and a high-definition (HD) map, according to the techniques of this disclosure;

[0021] Figure 11 is an example illustration of a process of generating a radar reference map using low-precision location data and a HD map, according to the techniques of this disclosure;

[0022] Figure 12 is another example illustration of generating a radar reference map using low-precision location data and a HD map, according to the techniques of this disclosure;

[0023] Figure 13 is another example illustration of a process of generating a radar reference map using low-precision location data and a HD map, according to the techniques of this disclosure;

[0024] Figure 14 is an example illustration of determining a radar occupancy grid using a HD map, according to the techniques of this disclosure;

[0025] Figure 15 is another example illustration of determining a radar occupancy grid using a HD map, according to the techniques of this disclosure;

[0026] Figure 16 is an example illustration of a process of determining a radar occupancy grid using a HD map, according to the techniques of this disclosure;

[0027] Figure 17 illustrates a flowchart of an example process for updating a radar reference map for radar detection based vehicle localization through multiple iterations, according to the techniques of this disclosure;

[0028] Figure 18 illustrates an example implementation 1800 configured for updating a radar reference map for radar detection based vehicle localization through multiple iterations, according to the techniques of this disclosure;

[0029] Figure 19FIG. illustrates a pipeline for updating a radar reference map for radar detection based vehicle localization through multiple iterations, in accordance with the techniques of the disclosure;

[0030] Figures 20-1 to 20-3 FIG. illustrates an example implementation of hindsight for updating a radar reference map for radar detection based vehicle localization through multiple iterations, in accordance with the techniques of the disclosure;

[0031] Figure 21 FIG. illustrates an example process for determining a hindsight maximum bound for determining radar coordinates when updating a radar reference map for radar detection based vehicle localization through multiple iterations, in accordance with the techniques of the disclosure;

[0032] Figures 22-1 to 22-2 FIG. illustrates a flowchart of an example of a radar detection based vehicle localization process, in accordance with the techniques of the disclosure; and

[0033] Figure 23 FIG. illustrates a flowchart of an example process for radar detection based vehicle localization. DETAILED DESCRIPTION

[0034] SUMMARY

[0035] This document describes radar detection based vehicle localization methods and systems. Radar localization begins with the construction of a radar reference map. The radar reference map can be generated and updated using different techniques as described herein. Once the radar reference map is available, real-time localization can be achieved using inexpensive radar sensors and navigation systems. Using the techniques described in this document, data from radar sensors and navigation systems can be processed to identify stationary localization objects or landmarks in the vicinity of the vehicle. Comparing landmark data originating from the vehicle’s on-board sensors and systems to landmark data detailed in the radar reference map can generate an accurate pose of the vehicle in its environment. By using inexpensive radar systems and lower quality navigation systems, highly accurate vehicle poses can be obtained in a cost effective manner.

[0036] Example Environment

[0037] Figure 1 is an example illustration 100 of an environment in which a radar reference map can be generated, updated, or used. In the example illustration 100, a system 102 is disposed in a vehicle 104 (e.g., an ego vehicle) traveling along a road 106.

[0038] The system 102 utilizes a radar system (not shown) to transmit radar signals (not shown). The radar system receives radar reflections 108 of the radar signals from the objects 110 (radar detections). In the example illustration 100, the radar reflections 108-1 correspond to the objects 110-1 (e.g., a sign), the radar reflections 108-2 correspond to the objects 110-2 (e.g., a building), and the radar reflections 108-3 correspond to the objects 110-3 (e.g., a guardrail).

[0039] The radar reflections 108 can be used to generate a radar reference map, as discussed with reference to Figures 3 to 13 The radar reflections 108 can be used to update an existing radar reference map, as discussed with reference to Figures 17 to 21 The radar reflections 108 can also be used in conjunction with an existing radar reference map to radar localize the vehicle 104, as discussed with reference to Figure 23

[0040] Example system

[0041] Figure 2-1 is an example illustration 200-1 of a system that can be used to generate, update, or use a radar reference map. The example illustration 200-1 includes the system 102 of the vehicle 104 and a cloud system 202. Although the vehicle 104 is illustrated as a car, the vehicle 104 can include any vehicle (e.g., a truck, a bus, a boat, an airplane, etc.) without departing from the scope of the present disclosure. The system 102 and the cloud system 202 can be connected by a communication link 204. One or both of the system 102 and the cloud system 202 can be used to perform the techniques described herein.

[0042] As shown below the respective systems, each of the systems includes at least one processor 206 (e.g., the processor 206-1 and the processor 206-2), at least one computer-readable storage medium 208 (e.g., the computer-readable storage medium 208-1 and 208-2), a radar localization module 210 (e.g., the radar localization module 210-1 and 210-2), and a communication system 212 (e.g., the communication system 212-1 and 212-2). The communication system 212 facilitates the communication link 204.

[0043] ​The system 102 additionally includes a navigation system 214 and a radar system 216. The navigation system 214 can include a geospatial positioning system (e.g., a GPS, GNSS, or GLONASS sensor), an inertial measurement system (e.g., a gyroscope or accelerometer), or other sensors (e.g., a magnetometer, a software positioning engine, a wheel tick sensor, a lidar odometer, a visual odometer, a radar odometer, or other sensor odometer). The navigation system 214 can provide high-precision position data (e.g., within a meter) or low-precision position data (e.g., within a few meters). The radar system 216 indicates radar hardware used to transmit and receive radar signals (e.g., the radar reflections 108). In some implementations, the radar system 216 provides static detections to the radar positioning module 210 (e.g., filtering can be performed within the radar system 216).

[0044] The processor 206 (e.g., an application processor, microprocessor, digital signal processor (DSP), or controller) executes instructions 218 (e.g., instructions 218-1 and 218-2) stored in the computer-readable storage medium 208 (e.g., a non-transitory storage device, such as a hard drive, SSD, flash memory, read only memory (ROM), EPROM, or EEPROM) to cause the system 102 and the cloud system 202 to perform the techniques described herein. The instructions 218 can be part of an operating system and / or one or more applications of the system 102 and the cloud system 202.

[0045] The instructions 218 cause the system 102 and the cloud system 202 to take actions (e.g., create, receive, modify, delete, send, or display) with respect to data 220 (e.g., 220-1 and 220-2). The data 220 can include application data, module data, sensor data, or I / O data. Although illustrated as being within the computer-readable storage medium 208, portions of the data 220 can be within random access memory (RAM) or cache (not illustrated) of the system 102 and the cloud system 202. Moreover, the instructions 218 and / or the data 220 can be located remotely from the system 102 and the cloud system 202.

[0046] The radar positioning module 210 (or portions thereof) can be comprised of the computer-readable storage medium 208 or can be a separate component (e.g., implemented in dedicated hardware in communication with the processor 206 and the computer-readable storage medium 208). For example, the instructions 218 can cause the processor 206 to implement or otherwise cause the system 102 or the cloud system 202 to implement the techniques described herein.

[0047] Figure 2-2is an example diagram 200-2 of a radar localization module 210 that can be used to implement radar detection-based vehicle localization. In example diagram 200-2, the radar localization module 210 is configured in a reference mode. The reference mode is used when the radar localization module 210 is used to construct a radar reference map. The radar localization module 210 includes two sub-modules, a vehicle state estimator 222, a scan matcher 224, and two optional sub-modules, a static object identifier 226 and an occupancy grid generator 228. One or both of the static object identifier 226 and the occupancy grid generator 228 can or can not be present.

[0048] The vehicle state estimator 222 receives navigation data 230 from a navigation system (e.g., from the navigation system 214 of the vehicle 104). Generally, in the reference mode, the navigation data 230 can come from a high-quality navigation system that provides a higher degree of accuracy than a commercial or consumer-grade navigation system (e.g., a navigation system used for mass production). From the navigation data 230, the vehicle state estimator 222 determines ego trajectory information about the current dynamic state (e.g., velocity, yaw rate) of the vehicle 104, and can provide the state estimate and other navigation data 230 (e.g., vehicle 104 latitude and longitude) to any other sub-modules present in the radar localization module 210. The ego trajectory information includes information derived from the vehicle system that can be used to predict the direction and velocity of the vehicle. Figure 2-1

[0049] The static object identifier 226 receives radar detections 232 from one or more radar sensor locations around the vehicle 104. If the static object identifier 226 is not being used, the radar detections 232 can be received by the occupancy grid generator 228, the scan matcher 224, or another sub-module designed to accept radar detections 232 and distribute radar data to other modules and sub-modules of the vehicle system. The static object identifier 226 determines whether the radar detections 232 are static detections based on the ego trajectory information from the vehicle state estimator 222, and outputs any identified static detections to the occupancy grid generator 228 (if it is being used) or the scan matcher 224.

[0050] The occupancy grid generator 228 can receive as input the radar detections 232 (if the static object identifier 226 is not being used by the radar localization module 210), or the static radar detections output by the static object identifier 226, as well as the ego trajectory information and navigation data 230 output from the vehicle state estimator 222. The occupancy grid generator 228 uses this input to determine a statistical probability of occupancy (e.g., an occupancy grid) at any given location in the environment of the vehicle 104, as discussed in other parts of this document.​

[0051] The scan matcher 224 can receive ego trajectory information and landmark data as input. The landmark data can be radar detections 232, static radar detections from the static object identifier 226, or an occupancy grid output by the occupancy grid generator 228, depending on which optional sub-module is being used. As described elsewhere in this document, the scan matcher 224 finds the best Normal Distribution Transform (NDT) between the landmark data and the high-quality navigation data 230 and outputs an NDT radar reference map 234.

[0052] Figure 2-3 is another example illustration 200-3 of a radar localization module that can be used to implement radar-detection-based vehicle localization. In the example illustration 200-3, the radar localization module 210 is configured in a real-time localization mode. The main difference between the real-time localization mode and the reference mode of the radar localization module 210 includes the navigation data 230 sourced from a low-quality navigation system and the additional input to the scan matcher module 224. The output of the radar localization module 210 in the real-time localization mode is an updated vehicle pose 236 of the vehicle 104.

[0053] In the real-time localization mode, the scan matcher 224 receives the NDT radar reference map 234 as input in addition to the landmark data and ego trajectory information. The scan matcher uses this input to determine an NDT grid. The NDT grid is compared to the NDT radar reference map to determine the updated vehicle pose 236.

[0054] In one non-limiting example, the radar localization module 210 can be used in the reference mode in a vehicle 104 equipped with a high-quality GNSS system that is specifically configured to create or assist in creating an NDT radar reference map. The real-time localization mode can be considered the normal mode of operation of the radar localization module 210; that is, a vehicle that is not specifically configured to create an NDT radar reference map can operate normally with the radar localization module 210 in the real-time localization mode.

[0055] Constructing a Radar Reference Map

[0056] Figure 3is an example illustration 300 of generating a radar reference map from radar detections. The example illustration 300 can be performed by the system 102 and / or the cloud system 202. At 302, radar detections 304 are received. The radar detections 304 include stationary radar detections (e.g., detections of stationary objects from the radar system 216, including signs, poles, obstacles, landmarks, buildings, overpasses, curbs, road adjacent objects such as fences, trees, foliage, leaves, or spatial statistical patterns) with corresponding global coordinates (e.g., from the navigation system 214) for each time / location. The global coordinates can include high precision location data (e.g., when the navigation system 214 is a high precision navigation system). The radar detections 304 can include point clouds, with corresponding uncertainty, and / or include various radar data or sensor measurements.

[0057] At 306, a radar occupancy grid 308 is determined from the radar detections 302. The radar occupancy grid 308 is a grid-based representation of the environment. For example, the radar occupancy grid 308 can be a Bayesian, Dempster Shafer, or other type of occupancy grid. Each cell of the radar occupancy grid 308 represents an independent portion of space, and each cell value of the radar occupancy grid 308 represents a probability (e.g., 0-100%) that the corresponding portion of space is occupied. A probability of approximately 0% for a cell can indicate that the corresponding portion of space is free, while a probability close to 100% can indicate that the corresponding portion of space is occupied and thus not free space. Techniques for determining the radar occupancy grid 308 will be discussed further around Figure 4 、 Figures 7-9 and Figures 14-17 are discussed further.

[0058] At 310, radar landmarks 312 are determined from the radar occupancy grid 308. The radar landmarks 312 are center coordinates of groups of individual cells of the radar occupancy grid 308 having probabilities greater than a threshold value. The radar landmarks 312 include clusters, contours, or bounding boxes of cells of the radar occupancy grid 308. The radar landmarks 312 have weights based on one or more of the probabilities, classifications, or cross-sectional values of the corresponding radar landmarks 312. The radar landmarks 312 can be determined using binarization, clustering algorithms, or machine learning on the radar occupancy grid 308. The determination of the radar landmarks 312 generally applies a threshold to the radar occupancy grid 308 and removes any noise from the radar occupancy grid 308.

[0059] At 314, a radar reference map 316 is generated from the radar landmarks 312. The radar reference map 316 can be a statistical reference map, such as a Gaussian representation. The radar reference map 316 is a collection of Gaussian distributions 318 corresponding to the occupied area. The Gaussian distributions 318 (or cells of the radar reference map 316) have associated location information (e.g., low or high quality location information, depending on how the radar reference map 316 is generated). Each cell of the radar reference map 316 can have a single Gaussian distribution 318 or be blank. Although not required, the radar reference map 316 has larger cells than the cells of the radar occupancy grid 308. The radar reference map 316 can be a standalone map or a layer in another map (e.g., a layer in a high definition (HD) map).

[0060] The radar reference map 316 can contain metadata associated with the respective Gaussian distributions 318. For example, the metadata can contain information about the shape or size of the clusters of Gaussian distributions 318. The metadata can also include object associations, such as certain Gaussian distributions 318 belonging to a sign or guardrail. Location data can also be contained within the metadata. Techniques for generating the radar reference map 316 will be discussed further with respect to Figure 5 、 Figure 6 and Figures 10-13 .

[0061] Figure 4 is an example illustration 400 of determining the radar occupancy grid 308 from the radar detections 304. The example illustration 400 is generally performed by the system 102, although some or all of the example illustration 400 can be performed by the cloud system 202. The example illustration 400 assumes that the location data associated with the radar detections 304 is high-precision location data (e.g., the navigation system 214 contains a high-precision GNSS).

[0062] At 402, a set (e.g., temporal) of radar detections 304 (e.g., radar detections 304 corresponding to time zero) is received and radar occupancy evidence 404 is determined from the set of radar detections 304. The radar occupancy evidence 404 corresponds to respective cells of the radar occupancy grid 308 and indicates occupied space within the radar occupancy grid 308. The radar occupancy evidence 404 is based on radar reflections 108 corresponding to the set of radar detections 304 as well as associated range and azimuth uncertainty.

[0063] At 406, radar occupancy probabilities 408 are determined from the radar occupancy evidence 404. For example, the radar occupancy probabilities 408 can be given by Equation 1:

[0064] p = 0.5 + 0.5 - e[1]

[0065] where p is the radar occupancy probability 408 and e is the occupancy evidence 404.

[0066] Steps 402 and 406 can be repeated for radar detections 304 of other groups corresponding to later times / locations. For each of the later times / locations, the radar occupancy probabilities 408 are fused at 410 with a decayed and shifted radar occupancy grid 412. The decayed and shifted radar occupancy grid 412 represents the current radar occupancy grid 414 (e.g., the radar occupancy grid 308 for the current time / location) with decayed probabilities and cells shifted due to movement of the vehicle between the previous time / location and the current time / location. The fusion serves to update (e.g., append) the radar occupancy grid 308 based on the subsequent radar detections 304 corresponding to the later times / locations.

[0067] To generate the decayed and shifted radar occupancy grid 412, at 416, the current radar occupancy grid 414 (e.g., at the respective locations) is decayed to form a decayed radar occupancy grid 418. Decaying includes forgetting, minimizing, or otherwise removing old evidence from the current radar occupancy grid 414. This ensures that only recently generated cells are used for the fusion. It should be noted that the radar occupancy grid 308 is not decayed; rather, the current radar occupancy grid 414, which is a snapshot of the radar occupancy grid 308, is decayed.

[0068] The decayed radar occupancy grid 418 is then shifted at 420 to form the decayed and shifted radar occupancy grid 412. Each cell of the radar occupancy grid 308 (and the current radar occupancy grid 414) represents an area. Thus, when the vehicle 104 moves, the grid must be shifted. To shift the grid, a vehicle position 422 at the time of the shift / decay is received. The decayed radar occupancy grid 418 is shifted by an integer number of cells corresponding to the vehicle position 422. For example, the integer number can be based on a change between the vehicle position 422 corresponding to the unshifted occupancy grid (e.g., the decayed radar occupancy grid 418 and the current radar occupancy grid 414) and the vehicle position 422.

[0069] As described above, the decayed and shifted radar occupancy grid 412 is fused at 410 with the radar occupancy probabilities 408 of the current group of radar detections 304. Over time, the fusion effectively accumulates the radar occupancy probabilities 408 to make the radar occupancy grid 308 more robust. Any fusion method can be used. For example, a Bayesian fusion method can be used according to Equation 2:

[0070]

[0071] where p 新 is the occupancy probability of the respective cell (e.g., in the radar occupancy grid 308), p 旧is the existing radar occupancy probability of the respective cell (e.g., in the decayed and shifted radar occupancy grid 412), and p 测量 is the radar occupancy probability 408 of the respective cell.

[0072] By using the example illustration 400, radar occupancy probabilities 408 from multiple times / locations can be fused. In doing so, the radar occupancy grid 308 becomes accurate and robust for the example illustration 300.

[0073] Figure 5 is an example illustration 500 of determining a radar reference map 316 from radar landmarks 312. At 502, an NDT cell 504 is established. The NDT cell 504 is a cell of the radar reference map 316. The NDT cell 504 is typically much larger (e.g., 15 times larger) than a cell of the radar occupancy grid 308.

[0074] For each NDT cell 504 with multiple radar landmarks 312, a Gaussian distribution 318 (e.g., a multivariate distribution with a mean and a covariance) is determined. To do so, at 506, a mean and a covariance 508 are determined for the respective NDT cell 504. The mean and the covariance 508 are based on the radar landmarks 312 identified within the respective NDT cell 504. The mean of the respective NDT cell 504 can be determined based on Equation 3:

[0075]

[0076] where p j is the occupancy probability of the radar occupancy grid 308 at a given cell of the radar occupancy grid 308, x j is the given cell location, and n is the number of cells within the radar landmarks 312 of the respective NDT cell 504.

[0077] The covariance (e.g., a 2x2 matrix) of the respective NDT cell 504 can be determined based on Equation 4:

[0078]

[0079] Advantageously, the mean and the covariance 508 are based on occupancy probabilities. At 510, the covariance of the respective NDT cell 504 can be manipulated such that the smallest eigenvalue of the covariance matrix is at least a certain multiple of the largest eigenvalue of the covariance matrix. The mean and the covariance 508 (or the manipulated covariance if step 510 is performed) constitute the Gaussian distribution 318 of the respective NDT cell 504. If there is one or fewer radar landmarks 312 within the respective NDT cell 504, the respective NDT cell 504 is indicated as unoccupied.

[0080] Steps 506 and 510 can then be performed for other NDT cells 504. At 512, the Gaussian distributions 318 of the respective NDT cells 504 are combined to form the radar reference map 316. Once combined, the NDT cells 504 of the radar reference map 316 can have a single Gaussian distribution 318 or nothing (e.g., indicating as unoccupied).

[0081] While steps 506 and 510 are discussed as being performed for respective NDT cells 504 and then for other NDT cells 504, in some implementations, steps 506 can be performed for a group of NDT cells 504 before performing step 510 for the group of NDT cells 504. For example, a mean and a covariance 508 can be determined for each NDT cell 504 of the group. Then, the covariance can be manipulated for each NDT cell 504 of the group as needed.

[0082] Example flow of constructing a map

[0083] Figure 6 is an example illustration 600 of a method of constructing a radar reference map 316. The example illustration 600 can be implemented with the examples previously described, such as the example illustrations 100, 300, 400, and 500. The order in which the operations 602-610 are shown and / or described is not intended to be construed as a limitation, and any number or combination of the operations can be combined in any order to implement the method 600 of the illustration or an alternative method.

[0084] At 602, a radar occupancy grid is received. For example, the radar localization module 210 can receive the radar occupancy grid 308.

[0085] At 604, radar landmarks are determined from the radar occupancy grid. For example, the radar localization module 210 can use a threshold on occupancy probabilities within the radar occupancy grid 308 to determine the radar landmarks 312. The radar landmarks 312 can include center coordinates of groups of respective cells of the radar occupancy grid having occupancy probabilities above or within a threshold value.

[0086] At 606, radar reference map cells are formed. For example, the radar localization module 210 can create the NDT cells 504 of the radar reference map 316.

[0087] At 608, a Gaussian distribution is determined for a radar reference map cell containing a plurality of radar landmarks. For example, the radar localization module 210 can determine a mean and a covariance 508 for each of the NDT cells 504 containing a plurality of radar landmarks 312.

[0088] At 610, a radar reference map is generated. The radar reference map includes radar reference map cells containing a Gaussian distribution and radar reference map cells indicated as unoccupied. The radar reference map cells indicated as unoccupied correspond to radar reference map cells that do not contain a plurality of radar landmarks. For example, the radar localization module 210 can combine the NDT cells 504 having the Gaussian distribution 318 with the NDT cells 504 indicated as unoccupied to form the radar reference map 316.

[0089] Figure 7 is an example illustration 700 of determining radar occupancy probabilities 408 using multiple vehicle trips with low-accuracy position data. Figure 8 is an example illustration 800 of a similar process. Thus, the following description describes both example illustrations 700 and 800. Example illustrations 700 and 800 are generally performed by the cloud system 202 based on radar detections 304 received from multiple vehicle trips, but one or more steps can be performed by the system 102 (e.g., collecting the radar detections 304 and sending the radar detections 304 to the cloud system 202 using the communication system 212). The radar occupancy probabilities 408 can then be fused at 410 to create the radar occupancy grid 308. The radar reference map 316 can then be generated from the radar occupancy grid 308, similar to example illustrations 100 and 300.

[0090] For large area regions, collecting high-accuracy position data is generally impractical or expensive. Example illustrations 700 and 800 determine radar occupancy probabilities 408 using multiple trips with low-accuracy position data, such as low-accuracy position data generated by most navigation systems (e.g., the navigation system 214) implemented within consumer and commercial vehicles. Because the position data is low-accuracy, multiple trips are needed to obtain accurate occupancy probabilities 408. Traditional techniques, such as multiple trip averaging, generally result in ambiguous and useless probability data.

[0091] Example illustrations 700 and 800 use statistical map fusion of multiple trips 702 (e.g., trip 702A and trip 702B) to correct for errors in the low-accuracy position data. Any number of trips 702 (but not just one) can be used, and the trips 702 can use the same vehicle or multiple vehicles and be created at different times. The statistical map fusion can be an extended particle filter simultaneous localization and mapping (SLAM) algorithm.

[0092] At 704, a particle 706 is created at a given location (e.g., at time t = 0). The particle 706 corresponds to a respective possible future location of the vehicle 104, but the specific further location is not yet determined. The particle 706 is based on a vehicle trajectory 708 corresponding to the given location (e.g., from the navigation system 214).

[0093] At 710, the future position of the particle is predicted to form a predicted position 712 (e.g., of the vehicle 104). The prediction is based on the vehicle trajectory 708. For example, the vehicle trajectory 708 can include velocity and yaw rate information. The velocity and yaw rate can be used to predict a new pose of the respective drive 702, and thus the predicted position 712 of the particle 706.

[0094] At 714, the particle weight 716 is updated. To this end, the particle 706 is projected onto the radar occupancy grid 308, where each particle 706 has a corresponding grid cell. The sum of all probability values in the corresponding grid cell (e.g., from multiple drives 702) is the weight of the particle 706. In other words, the weight of the particle 706 corresponds to how well the next radar detection 304 fits with the predicted position 712.

[0095] At 718, the existing probabilities are updated using the particle weights 716 to create the radar occupancy probability 408.

[0096] At 720, the particles are resampled to create resampled particles 722. Particles with high weights can be split, while particles with low weights can be eliminated. The resampled particles 722 become the particles at time t+1 for position prediction (step 710). The resampled particles 722 can also be used to correct the vehicle trajectory 708.

[0097] As time t increases, the radar occupancy probability 408 is updated, and the radar occupancy probability 408 is fused with previous radar occupancy probabilities according to 410.

[0098] One advantage of the example diagrams 700 and 800 is that they use data from the drives 702 simultaneously to build the radar occupancy probability 408. Thus, each set of particles 706 contains the same data from all drives 702. This means that the radar occupancy probability 408 of one particle 706 contains data from all drives 702. Furthermore, more than one particle 706 is used to update the radar occupancy probability 408 (e.g., at 718). Statistical map fusion also allows newer drives to be weighted more heavily than older drives, so that changing detections (seasonal vegetation changes, construction, etc.) can be compensated for at the cell level of the radar occupancy grid 308.

[0099] Figure 9is an example illustration 900 of a method of determining a radar occupancy probability 408. The example illustration 900 can be implemented with the examples previously described, such as the example illustrations 700 and 800. Operations 902-910 can be performed by one or more entities of the system 102 and / or the cloud system 202 (e.g., the radar localization module 210). The order in which the operations are illustrated and / or described is not intended to be a limitation, and any number or combination of operations can be combined in any order to implement the illustrated method or an alternative method.

[0100] At 902, a particle is created. For example, the radar localization module 210 can receive the radar detections 304 and create a particle 706 corresponding to a possible future location of the vehicle 104 (or vehicles, if the travel 702 corresponds to multiple vehicles).

[0101] At 904, a particle position is predicted for the particle. For example, the radar localization module 210 can receive the vehicle trajectory 708 and determine a predicted position 712.

[0102] At 906, a particle weight 716 of the particle is updated. For example, the radar localization module 210 can determine the particle weight 716 based on radar detections 304 corresponding to a later time.

[0103] At 908, a probability is updated based on the particle weight. For example, the radar localization module 210 can use the particle weight 716 to determine a radar occupancy probability 408 for fusing at 410.

[0104] At 910, the particle is resampled. For example, the radar localization module 210 can create a resampled particle 722 for predicting a future location at 710.

[0105] Figure 10 is an example illustration 1000 of generating a radar reference map 316 using low-precision location data and an HD map 1002. The example illustration 1000 is generally implemented by the system 102.

[0106] The HD map 1002 contains object properties 1004 determined at 1006 for HD map objects 1008 within the HD map 1002. The HD map objects 1008 can include street signs, overpasses, guardrails, traffic control devices, poles, buildings, k-rails, or other semi-permanent objects. The HD map 1002 contains information about each of the HD map objects 1008.

[0107] The object properties 1004 that can be determined at 1006 include aspects such as type 1010, size / orientation 1012, location 1014, links to roads 1016 of the corresponding HD map object 1008, and radar hardware information 1018. The type 1010 can define the corresponding HD map object 1008, such as a sign, an overpass, a guardrail, a traffic control device, a pole, a building, a k-track, or other semi-permanent object. The size / orientation 1012 can include the physical size and / or orientation of the corresponding HD map object 1008 (e.g., longitudinal vs. transverse, rotation relative to the ground, height relative to the ground). The location 1014 can include the UTM coordinates of the corresponding object, and the links to roads 1016 can include the specific location of the corresponding object relative to the corresponding road. For example, a guardrail can have some offset relative to its referenced location. In other words, the guardrail itself can not be exactly present at its location 1014. The links to roads 1016 can account for this. The radar hardware information 1018 can include any information that affects the radar reflections 108 from the corresponding HD map object 1008.

[0108] The misaligned radar detections 1020 are received at 1022 along with the object properties 1004. The misaligned radar detections 1020 are similar to the radar detections 304 with low-precision location data. The object properties 1004 are used to determine a vehicle pose 1024 of the vehicle 104 at the respective times of the misaligned radar detections 1020.

[0109] To do so, for each set of misaligned radar detections 1020, the vehicle can position itself relative to one or more of the HD map objects 1008. For example, a respective set of misaligned radar detections 1020 can contain detections of one or more HD map objects 1008. Since the locations 1014 (and other object properties 1004) of the one or more HD map objects 1008 are known, the radar localization module 210 can determine a vehicle pose 1024 at the respective set of misaligned radar detections 1020.

[0110] Once the vehicle pose 1024 is known for the respective misaligned radar detections 1020, the misaligned radar detections 1020 can be aligned at 1026. The alignment can include shifting or rotating the misaligned radar detections 1020 based on the respective vehicle pose 1024.

[0111] The aligned radar detections become the radar detections 304. The radar detections 304 can subsequently be used in the example illustrations 300, 400, and 500 to generate the radar reference map 316.

[0112] The radar reference map 316 can optionally be sent to the cloud system 202. There, at 1028, the radar reference map 316 can be updated from other radar reference maps based on other similar travels of the vehicle or other vehicles, or the radar reference map 316 is merged with the other radar reference maps. Figure 1 The operations 1102 to 1110 are generally performed by the system 102. The order in which the operations are illustrated and / or described is not intended to be limiting, and any number or combination of the operations can be combined in any order to implement the illustrated method or alternative methods.

[0113] Figure 11 The example illustration 1100 is an example illustration 1000 of a method to generate a radar reference map 316 using low-precision location data and an HD map 1100. The example illustration 1100 can be implemented with the examples described previously, such as the example illustration 1000. The operations 1102 to 1110 are generally performed by the system 102. The order in which the operations are illustrated and / or described is not intended to be limiting, and any number or combination of the operations can be combined in any order to implement the illustrated method or alternative methods.

[0114] At 1102, misaligned radar detections are received. For example, the radar localization module 210 can receive the misaligned radar detections 1020.

[0115] At 1104, HD map object properties are determined. For example, the radar localization module 210 can determine the object properties 1004 of the HD map objects 1008 of the HD map 1002.

[0116] At 1106, a vehicle pose is determined for each set of misaligned radar detections. For example, the radar localization module 210 can determine the vehicle pose 1024 based on the misaligned radar detections 1020 and the object properties 1004.

[0117] At 1108, the misaligned radar detections are aligned. For example, the radar localization module 210 can use the vehicle pose 1024 to un-misalign the misaligned radar detections 1020. The aligned radar detections essentially become the radar detections 304.

[0118] At 1110, a radar reference map is generated. For example, the radar localization module 210 can perform the example illustrations 300, 400, and 500 to generate the radar reference map 316 from the aligned radar detections (radar detections 304).

[0119] Optionally, at 1112, the radar reference map can be sent to a cloud system for updating. The updating can be based on similar reference maps generated by the vehicle or another vehicle. For example, the radar localization module 210 of the cloud system 202 can modify or update the radar reference map 316 based on other similar radar reference maps received from vehicles or other vehicles.

[0120] Figure 12is an example illustration 1000 of generating a radar reference map 316 using low-accuracy position data and an HD map 1200 (not shown). Example illustration 1200 is generally performed by cloud system 202 based on information received from system 102.

[0121] At system 102, misaligned radar detections 1020 run example illustrations 300, 400, and 500 to generate a misaligned radar reference map 1202. Misaligned radar reference map 1202 can be similar to radar reference map 316, except that the Gaussian distributions 318 can not be in the correct position (due to the low-accuracy position data).

[0122] In some implementations, only a portion of example illustration 400 can be performed. For example, the steps up to step 410 can be performed to form a separate occupancy grid for a respective set of misaligned radar detections 1020, as the low-accuracy position data can not be suitable by itself to be fused with other data to form a single radar occupancy grid (e.g., radar occupancy grid 308). Each misaligned radar occupancy grid can then be used to form misaligned radar reference map 1202.

[0123] Misaligned radar reference map 1202 (e.g., with misaligned Gaussian distributions similar to Gaussian distributions 318) is then sent to cloud system 202. At 1204, cloud system 202 uses object properties 1004 of HD map 1002 to align misaligned radar reference map 1202 to generate radar reference map 316.

[0124] To do so, similar to example illustration 1000, object properties 1004 can be used to align or change Gaussian 318 distributions within misaligned radar reference map 1202. For example, object properties 1004 can be used to determine Gaussian distributions 318 within misaligned radar reference map 1202 that correspond to corresponding HD map objects 1008. Since the locations of these objects are known, Gaussian distributions 318 can be shifted to the correct position.

[0125] If misaligned radar detections 1020 are continuous in space (e.g., they incrementally follow a path), misaligned radar reference map 1202 can have the exact positions of Gaussian distributions 318 relative to each other. In this case, misaligned radar reference map 1202 can only need to be globally shifted or rotated (rather than having to align each Gaussian distribution 318).

[0126] Misaligned radar detections 1020 can also be sent to cloud system 202 to be processed in a similar manner to example illustration 1000. However, misaligned radar reference map 1202 is much smaller and thus easier to transmit.

[0127] Figure 13 is an example illustration 1300 of a method of generating a radar reference map 316 using low-precision position data and an HD map 1002. The example illustration 1300 can be implemented with the examples described previously, such as the example illustration 1200. Operations 1302 through 1306 are generally performed by the cloud system 202. The order in which operations are shown and / or described is not intended to be limiting, and any number or combination of operations can be combined in any order to implement the illustrated method or alternative methods.

[0128] At 1302, an unaligned radar reference map is received. For example, the radar localization module 210 can receive the unaligned radar reference map 1202 from the system 102.

[0129] At 1304, HD map object properties are determined. For example, the radar localization module 210 can determine the object properties 1004 of the HD map objects 1008 of the HD map 1002.

[0130] At 1306, the unaligned radar reference map is aligned based on the HD map object properties. For example, the radar localization module 210 can use the object properties 1004 to determine a Gaussian distribution within the unaligned radar reference map 1202 that corresponds to the associated HD map object 1008. The unaligned radar reference map 1202 can then be corrected, adjusted, shifted, or otherwise corrected using the difference between the corresponding Gaussian distribution and the location of the HD map object 1008 to form the radar reference map 316.

[0131] Figure 14 is an example illustration 1400 of generating a radar occupancy grid 308 using an HD map 1002. The example illustration 1400 can be integrated with the example illustrations 300 and 500 to generate a radar reference map 316. The example illustration 1400 does not require radar (e.g., radar reflections 108, radar detections 304) to create the radar occupancy grid 308. However, it will be apparent that the example illustration 1400 does rely on the availability of HD map objects 1008 in the HD map 1002.

[0132] Similar to the example illustrations 1000 and 1200, the object properties 1004 of the HD map objects 1008 are determined. The object properties 1004 are used to determine a polyline 1404 at 1402. The polyline 1404 is a geometric representation of the HD map object 1008 relative to the radar occupancy grid 308. For example, the location, orientation, and details (e.g., offset) of a guardrail can be used to generate a polyline 1404 of the occupied space in the radar occupancy grid 308 that corresponds to the guardrail.

[0133] At 1406, the length of the respective polyline 1404 is compared to the grid size of the radar occupancy grid 308. If the polyline 1406 is no longer than a grid cell of the radar occupancy grid 308, the corresponding grid cell is marked as occupied.

[0134] However, if the polyline 1406 is longer than a grid cell, the polyline 1406 is oversampled at 1408 to create an oversampled polyline 1410. Oversampling includes adding more points along the respective polyline 1404 to model the radar occupancy grid output from the radar detections (e.g., from the radar detections 304).

[0135] The polyline 1404 or the oversampled polyline 1410 is then adjusted at 1412 based on known sensor perception offsets to form an adjusted polyline 1414. Continuing the guardrail example above, the system can know that a certain type of guardrail is always some distance further from the road edge than the locations contained within the object properties 1004. The adjusted polyline 1414 is used to mark the corresponding grid cell of the radar occupancy grid 308 as occupied.

[0136] As shown in the example radar occupancy grid 308, the HD map object 1008 (e.g., a guardrail) is represented as an occupancy space. In this way, the radar occupancy grid 308 can be generated without the vehicle having to travel through the corresponding area.

[0137] Figure 15 is an example illustration 1500 of generating the radar occupancy grid 308 using the HD map 1002 and a machine learning model. The example illustration 1500 can be integrated with the example illustration 1400 (e.g., to determine the offsets used to adjust at 1412). However, in some implementations, the machine learning model can be used to directly indicate the cells of the radar occupancy grid 308 (e.g., without the example illustration 1400).

[0138] In the example illustration 1500, the object properties 1004 are used to select and apply a model 1502 for adjusting the polylines at 1412. The model 1502 is based on the respective object properties 1004.

[0139] At 1504, the model 1502 is selected and applied to each HD map object 1008. The model 1502 is categorized by the respective object properties 1004. For example, there can be a model 1502 for each type of HD map object 1008 (e.g., a model 1502-1 for guardrails, a model 1502-2 for signs, a model 1502-3 for buildings, etc.). There can also be multiple models 1502 for a single type of HD map object 1008. For example, different types of guardrails can have different respective models 1502.

[0140] Model 1502 is previously generated and can be taught using machine learning on real-world occupancy grid data. For example, occupancy data (e.g., portions determined by example illustrations 300 and 400) can be fed into a model training program along with object attributes 1004 and HD map locations of corresponding HD map objects 1008. In doing so, the system is able to “learn” how to represent corresponding HD map objects 1008 in radar occupancy grid 308 (e.g., by polyline offset).

[0141] The output of corresponding model 1502 is occupancy grid data 1506 corresponding to polyline offset data. The polyline offset data can then be used to adjust the polylines of example illustration 1400.

[0142] In some implementations, occupancy grid data 1506 can include direct occupancy grid data. In such cases, no polylines are used and occupancy grid data 1506 is used as a direct input to radar occupancy grid 308 (e.g., occupancy grid data 1506 can be used to indicate cells of radar occupancy grid 308 as occupied).

[0143] As discussed above, radar occupancy grid 308 can be subsequently used to generate radar reference map 316. In this way, real-world radar occupancy data can be used to estimate offsets or occupancy for HD map objects 608 represented in radar occupancy grid 308.

[0144] Figure 16 is an example illustration 1600 of a method of generating radar occupancy grid 308 using HD map 1002. Example illustration 1600 can be implemented with the examples previously described, such as example illustrations 1400 and 1500. Operations 1602 through 1608 are generally performed by cloud system 202 as it is not necessary for vehicle 104. However, operations 1602 through 1608 (or portions thereof) can be performed by system 102. The order in which operations are shown and / or described is not intended to be limiting and any number or combination of operations can be combined in any order to implement the illustrated method or alternative methods.

[0145] At 1602, HD map object attributes are determined for HD map objects within the HD map. For example, radar localization module 210 can determine object attributes 1004 for HD map objects 1008 of HD map 1002.

[0146] At 1604, a polyline of the HD map object is determined. In some implementations, the polyline can be oversampled based on a size of the respective polyline and a grid size of the radar occupancy grid that is desired. For example, the radar localization module 210 can determine the polyline 1404 of the HD map object 1008, and if the polyline 1404 is longer than the grid size of the radar occupancy grid 308, the polyline 1404 is oversampled.

[0147] At 1606, an offset is applied to the polyline as needed. The offset can be based on the HD map object properties or a machine learning model used for the respective HD map object. For example, the radar localization module 210 can adjust the offset of the polyline 1404 based on the object properties 1004 or the model 1502.

[0148] At 1608, cells of the radar occupancy grid are indicated as occupied based on the polyline. For example, the radar localization module 210 can indicate cells of the radar occupancy grid 308 based on the polyline 1404 (after oversampling and offsetting according to 1406 and 1412).

[0149] Updating a Radar Reference Map

[0150] The following sections describe techniques for updating a radar reference map. Radar reference maps need to be continually improved as any particular environment that a vehicle travels through changes over time. Radar reference maps can include temporary obstacles that can not be considered landmarks, which can be added or removed. Additionally, radar reference maps can include erroneous landmark data or missing landmarks (e.g., occlusions in the radar reference map). Current techniques for updating radar reference maps typically use different sensors to collect landmark data in a single pass through an environment. The techniques described below use radar-centric data collected from multiple iterations of passing through an environment to update the quality of a radar reference map. These techniques use a process to ensure the most accurate and stable data (called post-observation); two non-limiting examples of this technique are illustrated. One example uses radar detections of objects and compares them to a HD map. The second example uses only radar detections when using post-observation as a way to ensure data accuracy and stability.

[0151] Figure 17 A flowchart 1700 is illustrated for updating a radar reference map for radar detection based vehicle localization through multiple iterations. The flowchart includes multiple passes 1702 (e.g., pass 1 through pass n, where n can be any integer greater than 1), where each pass is an iteration through an environment represented by the radar reference map. A radar localization module (e.g., radar localization module 210) is used to collect radar data for each pass. The radar localization module can use radar data to determine radar detections of objects in the environment. The radar localization module can also use a HD map (e.g., HD map 1002) to determine a location of each radar detection in the environment. The radar localization module can use the radar detections and the HD map to determine a quality of the radar reference map. The radar localization module can use the quality of the radar reference map to update the radar reference map. Figure 2-1The first step 1704 in the method 1700 is to receive radar detections and navigation data. In step 1706, the static object identifier can use the radar detections and navigation data to identify static objects from the raw radar detections and navigation data collected in step 1704. Step 1708 generates an occupancy grid based on the static objects identified in step 1706. In step 1710, a radar reference map is constructed from each occupancy grid generated in each trip 1702. The final step 1712 constructs (in initial trip 1) and updates (in each successive trip n) a relative NDT radar reference map (e.g., a map relative to the vehicle and using a relative coordinate system) based on the radar reference map generated in step 1710. The final step 1710 is conditioned on the HD map 1714 not being used in step 1710 in combination with the occupancy grid generated in step 1708. Otherwise, if the HD map 1714 is used in step 1710, the final step 1712 updates an absolute map (e.g., a universal map using a global coordinate system such as a UTM coordinate system) during each trip 1702.

[0152] Figure 18 An example illustration 1800 of updating a radar reference map for radar detection based vehicle localization through multiple iterations is illustrated. In the example implementation 1800, a vehicle 1802 equipped with a radar localization module (e.g., on-board, accessed through the cloud) uses a posteriori observations to accumulate accurate and stable radar data about dynamic objects 1804. The radar sensors on the vehicle 1802 have a radar scan 1806 that emits electromagnetic energy and receives reflections of the electromagnetic energy from objects. The radar scan 1806 can not be in range of the dynamic objects 1804-1, 1804-2, 1804-3 at all times. The radar scan 1806 can be in range of the dynamic objects 1804-1, 1804-2, 1804-3 at some times. The radar scan 1806 can be in range of the dynamic objects 1804-1, 1804-2, 1804-3 at all times. Figure 18 The dynamic objects 1804 are moving from in front of the vehicle 1802 (dynamic object 1804-1) to the side of the vehicle 1802 (dynamic object 1804-2) to the back of the vehicle 1802 (dynamic object 1804-3). The blind spot 1808 represents a range rate blind spot of one or more radar sensors on the vehicle 1802. Although in the example implementation 1800, the blind spot 1808 relates to the range rate state of the dynamic objects, any dynamic state of the dynamic objects 1804 can be used as an example.

[0153] The corner radar sensor mounted on the vehicle 1802 is configured such that the radar sensor's aperture angle is 45° with respect to the longitudinal axis of the vehicle 1802. This enables the corner radar sensor to have the same radar performance as the front and rear radar sensors of the vehicle 1802. The cumulative data from all radar sensors presents the most stable result of object detection at the back of the vehicle 1802, with the dynamic object 1804 reflecting several radar detections from different radar sensors. Since the occupancy grid exhibits cumulative data, all available sensor detections contribute to the radar reference map, even if the rear detections are the only ones taken into account. The detections from all radar sensors contribute to the occupancy probability, and no radar data is omitted. This process can be interpreted as applying binary weights to each cell of the occupancy grid, and the dynamic object 1804 can be excluded from the updated radar reference map.

[0154] Figure 19 A pipeline 1900 for updating a radar reference map for radar detection based vehicle localization through multiple iterations is illustrated. At a radar sensor stage 1902 of the pipeline 1900, a radar sensor 1904 receives raw radar detections 1906. At a static object identifier 1908 stage, the raw radar detections 1904 are classified as static or dynamic radar detections at 1910, and enhanced static radar detections 1912 are passed to an occupancy grid generator 1914 stage. At the occupancy grid generator 1914 stage, occupancy evidence from the enhanced static radar detections 1912 is extracted at 1916. The occupancy evidence extracted from 1916 is used to accumulate and filter static occupancy of an occupancy grid 1920 at 1918. A cumulator 1922 stage then extracts a posteriori observation information at 1924.

[0155] Figures 20-1 to 20-3 An example implementation of a posteriori observation for updating a radar reference map for radar detection based vehicle localization through multiple iterations is illustrated. In Figure 20-1 a vehicle 2002 equipped with a radar localization module (e.g., on-board, accessed through the cloud) uses a posteriori observation to accumulate precise and stable radar data about a static object 2004 (a street sign 2004). In a first time frame 2000-1 in FIG. 20, the street sign 2004 can be first detected by radar scans 2006-1 and 2006-2. Radar scans 2006-3 and 2006-4 have not yet detected the street sign 2004.

[0156] In Figure 20-2In the second time frame 2000-2, the vehicle 2002 has moved along the road, and the street sign 2004 is located to the side of the vehicle 2002. At least radar scans 2006-1, 2006-2, and possibly radar scan 2006-3 have detected the road sign 2004. In addition, radar scans 2006-1 and 2006-2 can have detected a second static object 2008, a tree 2008.

[0157] In the third time frame 2000-3, the vehicle 2002 has moved forward so that the road sign 2004 has been cumulatively detected by radar scans 2006-1, 2006-2, 2006-3, and 2006-4. At this point, the road sign is in hindsight of the radar scans, and the radar data relative to the road sign can be considered stable and accurate with high confidence. At least radar scans 2006-1, 2006-2, and possibly radar scan 2006-3 detected the tree in the time frame range 2000-3 and have moderate confidence, but higher than a third static object 2010, a guardrail 2010. Only scans 2006-1 and 2006-2 can have detected the guardrail 2010. Figure 20-3

[0158] Driving through the road in the depictions 2000-1 through 2000-3 over multiple iterations can increase the confidence level that the static objects 2004, 2008, and 2010 are permanent and can be considered landmarks. If any of the static objects 2004, 2008, and 2010 disappear during any of the multiple iterations of driving along the road, the confidence level of that object can decrease, and that object can be removed from the updated radar reference map. In this way, the radar reference map can be updated after each iteration to add or remove landmarks as they are detected or disappear, and to remove any false noise that can exist in the radar reference map.

[0159] Figure 21 ​An example process 2100 for determining and using a posterior observation maximum boundary for radar coordinates when updating a radar reference map for radar detection based vehicle localization through multiple iterations is illustrated. There are two options for radar coordinates, radar relative coordinates and radar absolute coordinates. At 2102, all radar reference maps are loaded. In addition, if radar absolute coordinates are used, at 2104, sample points are extracted from one or more HD maps and at 2106, the extracted sample points are transformed into a statistical distribution. At 2108, all posterior observation samples are collected and the minimum and maximum values for the coordinates (X and Y) are found. At 2110, the minimum and maximum values for the coordinates are used to create a maximum boundary for the new radar group. At 2112, a resolution is selected and the sample indices in the coordinate index resolution are checked. At 2114, based on the result of the checking process at 2112, if the sample is not new in the selected resolution, at 2116, only the log odds ratio (through the Bayesian inverse model or the max strategy) is merged into the radar reference map. If the sample is new in the selected resolution at 2114, at 2118, the original index of the new sample is added to the new map index.

[0160] Example Architecture

[0161] There are many advantages to applying the techniques discussed in this document to localize a vehicle based on radar detections. By using a radar-centric system (e.g., a radar system that includes four short-range radar sensors, each located at each of the four corners of the vehicle), adverse weather and lighting conditions that can reduce the effectiveness of other systems (e.g., cameras, LiDAR) can be overcome simply by using the radar system. In addition, the radar system used can be less expensive than some other sensor systems.

[0162] The techniques and systems described in this document enable a vehicle to determine its vehicle pose or position with sub-meter accuracy. To localize a vehicle based on radar detections, two steps can be performed in accordance with the techniques described herein, including the following steps: building an accurate radar reference map, and comparing the radar reference map to radar detections generated in real-time to accurately localize the vehicle. Figure 22-1 and 22-2 A detailed example of how these two steps can be implemented using a radar localization module (such as the radar localization module shown in Figure 2-2 and Figure 2-3 is described. Other examples can exclude some of the details in Figure 22-1 and Figure 22-2 (e.g., some of the sub-modules of the radar localization module are optional, as shown in Figure 2-2 and 2-3 ).

[0163] Figures 22-1 to 22-2An example flowchart 2200 illustrating a radar detection based vehicle localization process is shown. Figure 22-1 The first step is covered as flowchart 2200-1, Figure 22-2 The second step is covered as flowchart 2200-2. Figure 22-1 And Figure 22-2 The sub-steps within dashed box 2202 are the same in each step.

[0164] The first step of vehicle localization 2204 is to construct a precise radar reference map containing landmark information. Details of several different processes of constructing a radar reference map have been described above. Figure 22-1 The example flowchart shown details the architecture of constructing a radar reference map in a vehicle 2204-1 that is specially equipped with a high-quality navigation system. The radar localization module is in reference mode at this step.

[0165] In Figure 22-1 In the example flowchart shown, one or more radar sensors 2206 receive raw radar detections 2208. At the same time, a vehicle state estimator 2214 collects high-quality GNSS 2210-1 and inertial measurement unit (IMU) 2212-1 data to determine vehicle state and vehicle pose. Raw radar detections 2208 are collected at a certain rate, for example, every 50 milliseconds (ms). Raw radar detections 2208 are identified as static detections or dynamic detections by a static object identifier 2216. The static object identifier 2216 uses vehicle state information (e.g., rate of change of distance) provided by the vehicle state estimator 2214 to determine (e.g., by rate of change of distance dealiasing) the identification of raw radar detections 2208.

[0166] Static detections are output to an occupancy grid generator 2218. The occupancy grid generator 2218 estimates the occupancy probability of each cell (e.g., 20 centimeter (cm) by 20 cm cell) in the occupancy grid. The cell size can affect the processing time of this sub-step. Different processes (including Bayesian inference, Dempster-Shafer theory, or other processes) can be used to estimate the occupancy probability of each cell. The occupancy grid generated by the occupancy grid generator 2218 can be in relative coordinates (e.g., local frame of reference) at the vehicle 2204-1, as the occupancy grid generator 2218 receives vehicle state data from the vehicle state estimator 2214 that assists in creating the occupancy grid.

[0167] The scan matcher 2220 performs a series of sub-steps 2220-1 through 2220-4. Sub-step 2220-1 transforms the occupancy grid from a relative coordinate system to a UTM coordinate system. Vehicle state information from the vehicle state estimator 2214 is used during the transformation. Sub-step 2220-2 accumulates the occupancy grid output from sub-step 2220-1 at a particular rate (e.g., 10 Hz). This rate can be adjusted based on the driving scenario (e.g., the number of landmarks in the environment of the vehicle 2204-1) and the accumulated occupancy grid output to sub-step 2220-3. Sub-step 2220-3 selects occupancy grid cells based on a high occupancy probability (e.g., a probability equal to or greater than 0.7). The selected occupancy grid cells can be represented as a point cloud. Sub-step 2220-4 transforms the selected occupancy grid cells into a Gaussian representation to create a Gaussian or NDT radar reference map. The NDT radar reference map can be stored locally on the vehicle 2204-1 or uploaded to the cloud 2222.

[0168] The second step of vehicle localization 2204 is to determine an adjusted vehicle pose based on a comparison of landmark-based radar detections to the radar reference map. Figure 22-2 The example flowchart in FIG. 22A details an architecture for determining an adjusted vehicle pose for the vehicle 2204-2. It can be assumed in this example that the vehicle 2204-2 is configured as a non-luxury vehicle manufactured in large quantities and at a cost margin that makes the use of high-quality GNSS and sensor packages impractical. That is, the GNSS system 2210-2 and the IMU 2212-2 used in the vehicle 2204-2 can be considered as general (lower quality) commercial navigation systems. The radar localization module is in a real-time localization mode at the second step. All sub-steps within the dashed box 2202 are the same as those of the first step shown in FIG. 22A and are not repeated for the sake of brevity. Figure 22-1 The first step of vehicle localization 2204 is to determine a coarse vehicle pose based on GNSS and IMU data.

[0169] At step 2224, a radar reference map based on vehicle state determined by the vehicle state estimator 2214 is downloaded from the cloud 2222. At sub-step 2220-5, the radar reference map is compared to the selected occupancy grid cells and, based on the comparison, a vehicle pose is corrected for the vehicle 2204-2. A confidence level in the accuracy of the corrected pose can be used to determine the accuracy of the corrected pose. Furthermore, the corrected vehicle pose can be used to remove errors (e.g., drift) in the GNSS system 2210-2 and the IMU 2212-2.

[0170] The comparison process matches against a radar reference map, which is a set of Gaussian representations that can utilize a real-time "map" derived from real-time radar detections to minimize the memory size of the data and contain statistical information. The real-time map is a sub-portion of the area represented by the radar reference map and contains the same Gaussian-type statistical information as the radar reference map. In another implementation, the filtered output from the occupancy grid can be directly compared to the Gaussian distributions in the radar reference map.

[0171] The NDT process matches statistical probability distributions between the NDT process matching reference data (e.g., discrete cells with built-in statistical models). For any given transformation (e.g., x, y, and rotation) of a real-time point (e.g., occupancy grid output), the real-time point can be assigned to a discrete NDT cell that contains the statistical distribution from the model in the radar reference map. The real-time point is an occupancy grid cell that is considered to be occupied but is treated as a point with a probability value attached to the point. The probability distribution of the real-time point can thus be computed. The NDT process finds the best transformation that maximizes the probability distribution.

[0172] Figure 23 An example process 2300 for radar detection based vehicle localization is illustrated. At 2302, at least one or more processors of a vehicle receive radar detections. At 2304, at least one or more processors of the vehicle receive navigation data. At 2306, at least one or more processors of the vehicle output ego trajectory information about a current dynamic state of the vehicle. The ego trajectory information can include at least one of a travel direction, a speed, a distance rate of change, and a yaw rate as determined from the radar detections and the navigation data. At 2308, landmark data is extracted from the radar detections and the ego trajectory information. At 2310, a normal distribution transformation grid is determined from the extracted landmark data and a radar reference map. At 2312, the vehicle pose is corrected according to the normal distribution transformation grid to localize the vehicle.

[0173] In this way, the techniques and systems described herein use a cost effective system, ignore dynamic objects, maximize statistical distribution patterns, and effectively handle static noise to accurately adjust the pose of a vehicle.

[0174] Example

[0175] Example 1 : A radar detection based vehicle positioning method, the method comprising: receiving, by at least one processor of a vehicle, radar detections from one or more radar sensors of the vehicle; receiving, by the at least one processor, navigation data from one or more navigation units of the vehicle; outputting, by the at least one processor, ego-trajectory information about a current dynamic state of the vehicle; extracting, by the at least one processor, landmark data from the radar detections and the ego-trajectory information; determining, by the at least one processor, a normal distribution transform grid from the extracted landmark data and a radar reference map; and in response to determining the normal distribution transform grid, correcting, by the at least one processor, a vehicle pose according to the normal distribution transform grid to position the vehicle.

[0176] Example 2: The method of example 1, wherein positioning the vehicle based on the radar detections comprises: outputting the ego-trajectory information about the current dynamic state of the vehicle comprises executing a vehicle state estimator that outputs the ego-trajectory information about the current dynamic state of the vehicle based on the navigation data received from the one or more navigation units, the one or more navigation units comprising at least one or more of a global navigation satellite system, an inertial navigation system, or an odometer of the vehicle; and determining the normal distribution transform grid comprises executing a scan matcher configured to generate the normal distribution transform grid based on the extracted landmark data and the radar reference map.

[0177] Example 3: The method of example 2, further comprising: using, by the vehicle state estimator, a previously corrected pose as a radar positioning input to the vehicle state estimator for updating the ego-trajectory information about the current dynamic state of the vehicle; determining, based at least in part on the radar positioning input, updated ego-trajectory information about the current dynamic state of the vehicle; and outputting, by the vehicle state estimator, the updated ego-trajectory information about the current dynamic state of the vehicle.

[0178] Example 4: The method of example 3, further comprising: extracting, by the at least one processor, updated landmark data from the radar detections and the updated ego-trajectory information; determining, by the at least one processor, an updated normal distribution transform grid from the updated landmark data and the radar reference map; and in response to determining the updated normal distribution transform grid, correcting, by the at least one processor, an updated vehicle pose according to the updated normal distribution transform grid to position the vehicle.

[0179] Example 5: The method of any of the preceding examples, wherein extracting the landmark data from the radar detections and the ego-trajectory information further comprises: identifying, by a static object identifier, static radar detections from the received radar detections and the ego-trajectory information.

[0180] Example 6: The method of any of the preceding examples, wherein extracting landmark data from the radar detections and the ego-trajectory information further comprises generating, by the occupancy grid generator, an occupancy probability for any location in the vehicle’s surrounding environment, the probability being based on the radar detections and the ego-trajectory information.

[0181] Example 7: The method of any of the preceding examples, wherein determining the normal distribution transform grid from the prior landmark data and the radar reference map comprises adjusting the prior landmark data prior to determining the normal distribution transform grid, the adjusting being based on a number of landmark detections in the vehicle’s surrounding environment.

[0182] Example 8: The method of any of the preceding examples, the method further comprising determining a confidence level of an accuracy of the corrected pose; and using the confidence level in determining the vehicle pose from the normal distribution transform grid to localize the vehicle.

[0183] Example 9: The method of any of the preceding examples, wherein the radar reference map comprises a Gaussian distribution, the Gaussian distribution comprising a mean and a covariance of a radar landmark within a respective cell of the radar reference map.

[0184] Example 10: The method of any of the preceding examples, wherein correcting the vehicle pose to localize the vehicle from the normal distribution transform grid comprises comparing the normal distribution transform grid to the radar reference map; and generating an accurate location of the vehicle in the vehicle’s environment based on comparing the normal distribution transform grid to the radar reference map.

[0185] Example 11: The method of any of the preceding examples, wherein the ego-trajectory information comprises at least one of a heading, a velocity, a rate of change of distance, or a yaw rate.

[0186] Example 12: The method of any of the preceding examples, wherein the landmark data comprises at least one of dynamic radar detections, static radar detections, or an occupancy grid.

[0187] Example 13: The method of any of the preceding examples, further comprising operating the vehicle in an autonomous or semi-autonomous mode based on the corrected vehicle pose.

[0188] Example 14: A system for vehicle localization based on radar detections, the system comprising: one or more processors configured for performing the method of any of examples 1-13.

[0189] Example 15: A computer-readable storage medium comprising instructions that, when executed by a processor, perform the method of any of examples 1-13.

[0190] Example 16: A system for vehicle localization based on radar detections, the system comprising: one or more processors configured to: receive radar detections from one or more radar sensors; receive navigation data from one or more navigation units; output ego-trajectory information about a current dynamic state of a vehicle; extract landmark data from the radar detections and the ego-trajectory information; determine a normal-distribution transform grid from the extracted landmark data and a radar reference map; and responsive to determining the normal-distribution transform grid, correct a vehicle pose according to the normal-distribution transform grid to localize the vehicle.

[0191] Example 17: The system of example 16, wherein the one or more processors are further configured to: output the ego-trajectory information about the current dynamic state of the vehicle at least by: executing a vehicle state estimator that outputs the ego-trajectory information about the current dynamic state of the vehicle based on the navigation data received from the one or more navigation units, the one or more navigation units comprising at least one or more of a global navigation satellite system, an inertial navigation system, or an odometer of the vehicle; and determine the normal-distribution transform grid from the extracted landmark data and the radar reference map at least by: executing a scan matcher configured to generate the normal-distribution transform grid based on the extracted landmark data and the radar reference map.

[0192] Example 18: The system of example 17, wherein the one or more processors are further configured to: use, by the vehicle state estimator, a previously corrected pose as a radar localization input to the vehicle state estimator for updating the ego-trajectory information about the current dynamic state of the vehicle; determine updated ego-trajectory information about the current dynamic state of the vehicle based at least in part on the radar localization input; and output, by the vehicle state estimator, the updated ego-trajectory information about the current dynamic state of the vehicle.

[0193] Example 19: The system of example 18, wherein the one or more processors are further configured to: extract updated landmark data from the radar detections and the updated ego-trajectory information; determine an updated normal-distribution transform grid from the updated landmark data and the radar reference map; and responsive to determining the updated normal-distribution transform grid, correct an updated vehicle pose according to the updated normal-distribution transform grid to localize the vehicle.

[0194] Example 20: The system of any one of examples 16-19, wherein the one or more processors are configured to extract the landmark data from the radar detections and the ego-trajectory information by at least: executing a static object identifier configured to identify static radar detections from the received radar detections and the ego-trajectory information.

[0195] Example 21 : The system of any one of examples 16 to 20, wherein the one or more processors are configured to extract landmark data from the radar detections and the ego-trajectory information by at least: executing an occupancy grid generator configured to generate an occupancy probability of any location in the vehicle’s surrounding environment, the probability based on the radar detections and the ego-trajectory information.

[0196] Example 22: The system of any one of examples 16 to 21, wherein the one or more processors are configured to determine the normal distribution transform grid from the extracted landmark data and the radar reference map by at least: adjusting the extracted landmark data prior to determining the normal distribution transform grid, the adjustment based on a number of landmark detections in the vehicle’s surrounding environment.

[0197] Example 23: The system of any one of examples 16 to 22, wherein the one or more processors are further configured to: determine a confidence level of an accuracy of the corrected pose; and use the confidence level when determining the vehicle pose from the normal distribution transform grid to localize the vehicle.

[0198] Example 24: A non-transitory computer-readable medium storing instructions configured to cause a processing device to: receive radar detections from one or more radar sensors; receive navigation data from one or more navigation units; execute a vehicle state estimator that outputs ego-trajectory information about a current dynamic state of a vehicle based on the navigation data received from the one or more navigation units, the one or more navigation units comprising at least one or more of a global navigation satellite system, an inertial navigation system, or an odometer of the vehicle; extract landmark data from the radar detections and the ego-trajectory information; execute a scan matcher configured to generate a normal distribution transform grid based on the extracted landmark data and a radar reference map; in response to determining the normal distribution transform grid, correct a vehicle pose from the normal distribution transform grid to localize the vehicle.

[0199] Example 25: The non-transitory computer-readable medium of example 24, wherein the instructions are further configured to cause the processing device to: use, by the vehicle state estimator, a previously corrected pose as a radar localization input to the vehicle state estimator for updating the ego-trajectory information about the current dynamic state of the vehicle; determine, based at least in part on the radar localization input, updated ego-trajectory information about the current dynamic state of the vehicle; and output, by the vehicle state estimator, the updated ego-trajectory information about the current dynamic state of the vehicle.

[0200] Example 26: The non-transitory computer-readable medium of example 24 or 25, wherein the instructions configured to cause the processing device to extract landmark data from the radar detections and ego-trajectory information comprise at least: executing a static object identifier configured to identify static radar detections from the received radar detections and ego-trajectory information.

[0201] Example 27: The non-transitory computer-readable medium of any one of examples 24-26, wherein the instructions configured to cause the processing device to extract landmark data from the radar detections and ego-trajectory information comprise at least: executing an occupancy grid generator configured to generate a probability of occupancy of any location in the vehicle's surroundings based on the radar detections and ego-trajectory information.

[0202] CONCLUSION

[0203] Although various implementations of radar-detection-based vehicle localization have been described in feature- and / or method-specific language, subject matter of the appended claims is not necessarily limited to the specific features or methods described. Rather, the specific features and methods are disclosed as example implementations of radar-detection-based vehicle localization. Further, although each of the various examples has been described with reference to certain features, it will be understood that features of one example can be used in combination with features of another example. Furthermore, to the extent that terms have been used in the description, these terms are used in their dictionary senses, unless otherwise explicitly provided. Finally, it will be understood that the terms and expressions used herein have their ordinary meanings.

Claims

1. A vehicle positioning method based on radar detection, the method comprising: The vehicle receives radar detection from one or more radar sensors of the vehicle by at least one processor; The at least one processor receives navigation data from one or more navigation units of the vehicle; The at least one processor receives a radar reference map, the radar reference map comprising a Gaussian distribution for the corresponding cells of the radar reference map, the Gaussian distribution comprising the mean and covariance of radar landmarks within the corresponding cells of the radar reference map; The at least one processor determines ego trajectory information about the current dynamic state of the vehicle from the navigation data; The radar occupancy grid is determined from the radar detection by the at least one processor; The at least one processor transforms the radar occupancy grid from a relative coordinate system to a Universal Transverse Mercator (UTM) coordinate system; The accumulated radar occupancy grid output; Based on the high occupancy probability, the accumulated occupancy grid output is selected as the occupancy grid cell; and the selected occupancy grid cell is transformed into a Gaussian radar reference map in Gaussian representation; as well as In response to the transformation, the at least one processor corrects the vehicle attitude of the vehicle to locate the vehicle based on a comparison between the Gaussian radar reference map and the selected occupied grid cells.

2. The method of claim 1, further comprising: The static object identifier determines whether the received radar detection is a static detection based on the self-trajectory information; as well as Output any identified static detections to the scan matcher.

3. The method of claim 1, further comprising: The at least one processor determines a normal distribution transformation grid, the normal distribution transformation grid comprising a Gaussian distribution for the corresponding cells of the normal distribution transformation grid, the Gaussian distribution comprising the mean and covariance of the radar landmarks within the corresponding cells of the normal distribution transformation grid.

4. The method of claim 3, further comprising: The radar landmark is determined from the radar occupancy grid by the at least one processor.

5. The method of claim 4, further comprising: Before determining the normal distribution transformation grid, the radar landmarks are adjusted by the at least one processor, the adjustment being based on the number of landmarks detected in the environment surrounding the vehicle.

6. The method as described in claim 5, characterized in that, The radar landmarks include clusters of occupied cells in the radar occupied grid.

7. The method of claim 1, further comprising: Determine the confidence level of the accuracy of the corrected vehicle attitude.

8. The method as described in claim 1, characterized in that, The vehicle attitude includes at least one of the vehicle's direction of travel, speed, rate of change of distance, and yaw rate.

9. A system for locating vehicles based on radar detection, the system comprising: One or more processors are configured to: Receive radar detection from one or more radar sensors; Receive navigation data from one or more navigation units; Receive a radar reference map, the radar reference map including a Gaussian distribution for the corresponding cells of the radar reference map, the Gaussian distribution including the mean and covariance of radar landmarks within the corresponding cells of the radar reference map; The navigation data is used to determine the vehicle's ego trajectory information regarding its current dynamic state. The radar grid occupancy is determined from the radar detection; Transform the radar occupancy grid from the relative coordinate system to the Universal Transverse Mercator (UTM) coordinate system; The accumulated radar occupancy grid output; Based on the high occupancy probability, the accumulated occupancy grid output is selected as the occupancy grid cell; and the selected occupancy grid cell is transformed into a Gaussian radar reference map in Gaussian representation; as well as In response to the transformation, the vehicle attitude is corrected based on a comparison between the Gaussian radar reference map and the selected occupied grid cells to locate the vehicle.

10. The system as claimed in claim 9, characterized in that, The one or more processors are also configured to: Execute a static object identifier, the static object identifier being configured to determine whether the received radar detection is a static detection; and Output any identified static detections to the scan matcher.

11. The system as described in claim 9, characterized in that, The one or more processors are also configured to: A normal distribution transformation grid is determined, the normal distribution transformation grid comprising a Gaussian distribution for the corresponding cells of the normal distribution transformation grid, the Gaussian distribution comprising the mean and covariance of the radar landmarks within the corresponding cells of the normal distribution transformation grid.

12. The system as claimed in claim 11, characterized in that, The one or more processors are also configured to: The radar landmarks are determined from the radar occupancy grid.

13. The system as described in claim 12, characterized in that, The one or more processors are also configured to: The radar landmarks are adjusted before the normal distribution transformation grid is determined, and the adjustment is based on the number of landmarks detected in the environment surrounding the vehicle.

14. The system as described in claim 13, characterized in that, The radar landmarks include clusters of occupied cells in the radar occupied grid.

15. The system as described in claim 9, characterized in that, The one or more processors are also configured to: Determine the confidence level of the accuracy of the corrected vehicle attitude.

16. The system as described in claim 9, characterized in that, The vehicle attitude includes at least one of the vehicle's direction of travel, speed, rate of change of distance, and yaw rate.

17. A non-transient computer-readable medium comprising instructions configured to cause a processing device to: Receive radar detection from one or more radar sensors; Receive navigation data from one or more navigation units; Receive a radar reference map, the radar reference map including a Gaussian distribution for the corresponding cells of the radar reference map, the Gaussian distribution including the mean and covariance of radar landmarks within the corresponding cells of the radar reference map; A vehicle state estimator is executed, which outputs ego trajectory information about the current dynamic state of the vehicle based on navigation data received from the one or more navigation units, wherein the one or more navigation units include at least one or more of the vehicle's global navigation satellite system, inertial navigation system, or odometer. The navigation data is used to determine the vehicle's ego trajectory information regarding its current dynamic state. The radar grid occupancy is determined from the radar detection; Transform the radar occupancy grid from the relative coordinate system to the Universal Transverse Mercator (UTM) coordinate system; The accumulated radar occupancy grid output; Based on the high occupancy probability, the accumulated occupancy grid output is selected as the occupancy grid cell; and the selected occupancy grid cell is transformed into a Gaussian radar reference map in Gaussian representation; In response to the transformation, the vehicle attitude is corrected based on a comparison between the Gaussian radar reference map and the selected occupied grid cells to locate the vehicle.

18. The non-transient computer-readable medium as claimed in claim 17, characterized in that, The instructions are also configured to cause the processing device to: A normal distribution transformation grid is determined, the normal distribution transformation grid comprising a Gaussian distribution for the corresponding cells of the normal distribution transformation grid, the Gaussian distribution comprising the mean and covariance of the radar landmarks within the corresponding cells of the normal distribution transformation grid.

Images