阅读说明：本技术 使用高清地图为自主车辆生成路线 (Generating routes for autonomous vehicles using high-definition maps ) 是由 马克·达蒙·惠乐 于 2017-12-22 设计创作，主要内容包括：一种系统生成用于自主车辆从源位置行驶到目的地位置的高清地图。该系统确定低分辨率路线，并且接收覆盖低分辨率路线的一组地理区域的高清地图数据。该系统使用地理区域内的车道元素来形成一组潜在部分路线。该系统计算潜在部分路线与低分辨率路线之间的误差，并且移除误差高于阈值的潜在部分路线。一旦完成，该系统选择最终路线，并且将信号发送至自主车辆的控制器以遵循最终路线。该系统在紧急情况下确定不是道路的一部分的车道附近的表面区域对于车辆驾驶是否安全。该系统存储用车道的表示来描述可导航表面区域的信息。(A system generates a high-definition map for autonomous vehicle travel from a source location to a destination location. The system determines a low resolution route and receives high definition map data covering a set of geographic areas of the low resolution route. The system uses lane elements within a geographic area to form a set of potential partial routes. The system calculates an error between the potential partial route and the low resolution route, and removes the potential partial route having an error above a threshold. Once completed, the system selects a final route and sends a signal to the controller of the autonomous vehicle to follow the final route. The system determines in an emergency whether a surface area near a lane that is not part of the road is safe for vehicle driving. The system stores information describing the navigable surface area with a representation of a lane.)

1. A non-transitory computer readable storage medium having encoded thereon instructions for determining a route using a high definition map, the instructions when executed by a processor cause the processor to perform steps comprising:

receiving, by an autonomous vehicle, a low resolution route from a source address to a destination address, the low resolution route identifying a set of streets for traveling from the source address to the destination address, wherein each street includes one or more lanes;

receiving a representation of a lane element graph, the lane element graph including nodes representing lane elements and edges representing connections between the lane elements;

determining a set of potential partial routes to an intermediate point between the source address and the destination address, each partial route being represented as a sequence of connected lane elements, the determining comprising repeatedly performing the steps of:

selecting a potential partial route from the set of potential partial routes;

identifying additional lane elements from the lane element map that are connected to the lane elements of the selected potential partial route;

determining an error measure between the low resolution route and a modified partial route obtained by adding the supplementary lane element to the selected potential partial route; and

in response to the error measure being below a threshold, replacing the selected partial route with the modified partial route and updating the threshold to the error measure for the modified partial route;

determining a potential full route based on a particular potential partial route of the set of potential partial routes, connecting the source address with the destination address;

selecting the potential full route as a final route in response to the potential full route having a smaller error measure than any other potential full route or potential partial route; and

sending a signal to a controller of the autonomous vehicle to cause the autonomous vehicle to drive along the final route.

2. The non-transitory computer-readable medium of claim 1, wherein the low resolution map is represented as a navigation map generated by a third party map provider, the navigation map including geocoding instructions for a route from the source address to the destination address.

3. The non-transitory computer-readable medium of claim 2, wherein the received navigation map is based on latitude-longitude coordinates.

4. The non-transitory computer-readable medium of claim 1, wherein a side between a first lane element and a second lane element through which a vehicle may travel directly from a portion of a street corresponding to the first lane element to a portion of a street corresponding to the second lane element indicates physical proximity between the lane elements.

5. The non-transitory computer readable medium of claim 1, wherein lane elements comprise one or more of:

a geometric lane boundary;

a traffic direction within the lane compared to the HD map route;

static traffic flow restrictions; or

Non-traffic road characteristics.

6. The non-transitory computer-readable medium of claim 1, further comprising: retrieving, from an online server, representations of one or more geographical areas that overlap with the low resolution route, each representation of a geographical area comprising a plurality of lane elements and relationships between the lane elements.

7. The non-transitory computer-readable medium of claim 1, further comprising:

sending information describing geocoding instructions for the low resolution route to an online server; and is

One or more geographical areas through which the low resolution route passes are received.

8. The non-transitory computer-readable medium of claim 1, further comprising:

rejecting the additional lane element in response to the error measure being above the threshold, wherein the threshold is not updated; and is

Identifying from the lane element map an alternative supplementary lane element that is connected to the lane element of the partial route.

9. The non-transitory computer-readable medium of claim 1, further comprising:

removing the selected potential partial route from the plurality of potential partial routes in response to determining that no other supplementary lane elements connected to the potential partial route result in the modified partial route having an error below the threshold; and is

A second potential partial route is selected from the plurality of potential partial routes.

10. The non-transitory computer-readable medium of claim 1, wherein the error measure represents a vertical distance between lane elements along the length of the potential partial route and the low resolution route.

11. The non-transitory computer-readable medium of claim 1, wherein determining the error measure further comprises:

establishing the threshold value representing a maximum allowed vertical distance between a potential partial route and the low resolution route;

identifying additional lane elements that may be added to the partial route; and

the error measure is calculated for each possible additional lane element.

12. The non-transitory computer readable medium of claim 11, wherein the error measure is determined for each modified partial route and the threshold is updated to represent a lowest error measure for a potential full route.

13. The non-transitory computer-readable medium of claim 1, wherein, in response to identifying the plurality of additional lane elements for adding to the potential partial route:

determining a plurality of modified partial routes, wherein each modified partial route is obtained by adding additional lanes to the potential partial route; and is

In response to determining that the modified partial route has an error measure that is lower than a currently updated threshold, adding the modified partial route to the plurality of potential partial routes.

14. A method for determining a route using a high definition map, comprising:

receiving, by an autonomous vehicle, a low resolution route from a source address to a destination address, the low resolution route identifying a set of streets for traveling from the source address to the destination address, wherein each street includes one or more lanes;

receiving a representation of a lane element graph, the lane element graph including nodes representing lane elements and edges representing connections between the lane elements;

determining a set of potential partial routes to an intermediate point between the source address and the destination address, each partial route being represented as a sequence of connected lane elements, the determining comprising repeatedly performing the steps of:

selecting a potential partial route from the set of potential partial routes;

identifying additional lane elements from the lane element map that are connected to the lane elements of the selected potential partial route;

determining an error measure between the low resolution route and a modified partial route obtained by adding the supplementary lane element to the selected potential partial route; and

in response to the error measure being below a threshold, replacing the selected partial route with the modified partial route and updating the threshold to the error measure for the modified partial route;

determining a potential full route based on a particular potential partial route of the set of potential partial routes, connecting the source address with the destination address;

selecting the potential full route as a final route in response to the potential full route having a smaller error measure than any other potential full route or potential partial route; and

sending a signal to a controller of the autonomous vehicle to cause the autonomous vehicle to drive along the final route.

15. The method of claim 14, wherein the low resolution map is represented as a navigation map generated by a third party map provider, the navigation map including latitude and longitude coordinate-based geocoding instructions for a route from the source address to the destination address.

16. The method of claim 14, wherein a side between a first lane element and a second lane element through which a vehicle may travel directly from a portion of a street corresponding to the first lane element to a portion of a street corresponding to the second lane element indicates physical proximity between the lane elements.

17. The method of claim 14, further comprising: retrieving, from an online server, representations of one or more geographical areas that overlap with the low resolution route, each representation of a geographical area comprising a plurality of lane elements and relationships between the lane elements.

18. The method of claim 14, further comprising:

rejecting the additional lane elements in response to the error measure being above the threshold, wherein the error measure represents a vertical distance between lane elements along the length of the potential partial route and the low resolution route, and wherein the threshold is not updated; and

identifying from the map of lane elements, alternative additional lane elements that are connected to lane elements of the potential partial route.

19. The method of claim 14, wherein, in response to identifying the plurality of additional lane elements for adding to the potential partial route:

determining a plurality of modified partial routes, wherein each modified partial route is obtained by adding additional lanes to the potential partial route; and is

In response to determining that the modified partial route has an error measure below a threshold, adding the modified partial route to the plurality of potential partial routes.

20. A method for determining a route using a high definition map, the method comprising:

storing a lane element map, the lane element map comprising lane elements and edges/nodes;

storing a plurality of geographic regions, each geographic region describing a portion of a world map;

receiving a request from an autonomous vehicle, the request describing instructions in the form of a low resolution route from a source location to a destination;

identifying a geographic area through which the low resolution route passes;

determining a subset of a lane element map for the low resolution route based on the identified geographic area; and

transmitting a representation of a subset of the lane element map to the autonomous vehicle, wherein the autonomous vehicle uses the lane element map to construct a high definition route for driving from the source address to the destination address.

21. A non-transitory computer readable storage medium storing instructions for implementing navigable surface boundaries in a high definition map encoded on the non-transitory computer readable storage medium, the instructions when executed by a processor cause the processor to perform steps comprising:

receiving a high-definition map representation of a geographic area, the high-definition map representation comprising a three-dimensional representation of a structure in the geographic area, the three-dimensional representation comprising representations of one or more lanes, each lane having a pair of lane boundaries, each lane boundary representing an edge of the lane;

identifying one or more structures of the geographic area from the high-definition map, each structure located outside the lane boundary representing a potential obstacle for a vehicle;

for each of the identified structures, identifying a set of points on the structure based on its vertical distance from a lane boundary of a lane;

generating a polyline representation through the identified set of points;

storing the multi-segment line representation in a high definition map representation of the geographic area as a navigable surface boundary to the lane; and

providing a high-definition map representation of the geographic area to an autonomous vehicle driving within the geographic area, wherein the vehicle makes a determination whether to drive on a navigable surface of a lane outside of the lane boundary.

22. The non-transitory computer readable medium of claim 21, wherein a structure represents at least one of:

a fence;

a safety guard rail;

a series of columns;

a wall;

a curb;

a ditch or drain;

a hill;

a building; or

And (4) trees.

23. The non-transitory computer readable medium of claim 21, wherein, for each identified structure, a type of surface is determined, the type of surface comprising:

a pavement;

gravel; or

Dust.

24. The non-transitory computer readable medium of claim 21, wherein a structure is identified from the high definition map based on images captured by a sensor or one or more vehicles driving on the lane.

25. The non-transitory computer-readable medium of claim 21, wherein identifying one or more structures representing an obstacle further comprises:

receiving a point cloud representation of the geographic area;

identifying three-dimensional points from the point cloud representation having heights above a threshold level;

identifying the three-dimensional point in an image; and

for each structure, the image is used to classify the structure.

26. The non-transitory computer readable medium of claim 25, wherein the structure is classified according to a machine learning based object recognition method.

27. The non-transitory computer readable medium of claim 21, wherein generating a multi-segment line further comprises:

determining a vertical distance between data points corresponding to the structure and the lane boundary;

for each structure, selecting a data point having a minimum vertical distance from the lane boundary; and

a polyline is determined that passes through the selected data point and one or more neighboring points.

28. The non-transitory computer-readable medium of claim 21, wherein the multi-segment line representation can be generated in one or more of the following coordinate systems:

a two-dimensional coordinate system; or

A three-dimensional coordinate system.

29. The non-transitory computer-readable medium of claim 21, wherein determining a difficulty level of the vehicle traveling on the navigable surface comprises:

for each of a plurality of points along a polyline, determining a score representing a level of difficulty for a car to travel on the navigable surface associated with the polyline based on a classification of a structure associated with the navigable surface; and

for each of a plurality of points along the polyline, storing the score with a representation of the lane.

30. The non-transitory computer-readable medium of claim 21, wherein the level of difficulty of the vehicle traveling on the navigable surface is based on a type of navigable surface on which the vehicle will travel.

31. The non-transitory computer-readable medium of claim 21, wherein implementing the navigable surface boundary comprises:

for a lane element on which the vehicle is currently traveling, accessing a polyline corresponding to the navigable surface boundary;

receiving an indication of an emergency representing an event forcing the vehicle out of the lane;

determining, in response to the indication of the emergency, that the vehicle is capable of safely traveling on the navigable surface, the determination being based on a score representing the difficulty level; and

sending one or more signals to a controller of the vehicle that cause the vehicle to travel within the navigable surface boundary.

32. A computer-implemented method, comprising:

receiving a high-definition map representation of a geographic area, the high-definition map representation comprising a representation of a structure in the geographic area, the representation of the structure comprising a representation of one or more lanes, each lane having a pair of lane boundaries, each lane boundary representing an edge of the lane;

identifying one or more structures of the geographic area from the high-definition map, each structure located outside the lane boundary representing a potential obstacle for a vehicle;

for each of the identified structures, identifying a set of points on the structure based on its vertical distance from a lane boundary of a lane;

generating a polyline representation through the identified set of points;

storing the multi-segment line representation in a representation of a high definition map of the geographic area as a navigable surface boundary to the lane; and

providing a high-definition map representation of the geographic area to an autonomous vehicle driving within the geographic area, wherein the vehicle makes a determination whether to drive on a navigable surface of a lane outside of the lane boundary.

33. The computer-implemented method of claim 31, wherein the structure represents at least one of:

a fence;

a wall;

a curb;

a hill;

a building; or

And (4) trees.

34. The computer-implemented method of claim 31, wherein, for each identified structure, a type of surface is determined, the type of surface comprising:

a pavement;

gravel; or

Dust.

35. The computer-implemented method of claim 31, wherein a structure is identified from the high-definition map based on images captured by a sensor or one or more autonomous vehicles driving on the lane.

36. The computer-implemented method of claim 31, wherein identifying one or more structures representing an obstacle further comprises:

receiving a point cloud representation of the geographic area;

identifying three-dimensional points from the point cloud representation having heights above a threshold level;

identifying the three-dimensional point in an image; and

for each structure, the image is used to classify the structure.

37. The computer-implemented method of claim 31, wherein generating a multi-segment line further comprises:

determining a vertical distance between data points corresponding to the structure and the lane boundary;

for each structure, selecting a data point having a minimum vertical distance from the lane boundary; and

a polyline is determined that passes through the selected data point and one or more neighboring points.

38. The computer-implemented method of claim 31, wherein determining a difficulty level of the vehicle traveling on the navigable surface comprises:

for each of a plurality of points along a polyline, determining a score representing a level of difficulty for a car to travel on a navigable surface associated with the polyline based on a classification of a structure associated with the navigable surface; and

for each of a plurality of points along the polyline, storing the score with a representation of the lane.

39. The computer-implemented method of claim 31, wherein the difficulty level of the vehicle traveling on the navigable surface is based on one or more of:

the type of navigable surface on which the vehicle is to travel; and

the navigable surface boundary poses a risk of damage to the vehicle.

40. The computer-implemented method of claim 31, wherein implementing the navigable surface boundary comprises:

for a lane element on which the vehicle is currently traveling, accessing a polyline corresponding to the navigable surface boundary;

receiving an indication of an emergency representing an event forcing the vehicle out of the lane;

determining, in response to the indication of the emergency, that the vehicle is capable of safely traveling on the navigable surface, the determination being based on a score representing the difficulty level;

sending one or more signals to a controller of the vehicle that cause the vehicle to travel within the navigable surface boundary.

41. A method for implementing navigable surface boundaries in a high definition map, comprising:

storing a high-definition map representation of a geographic area, the high-definition map representation comprising a representation of a structure in the geographic area, the representation of the structure in the geographic area comprising a representation of one or more lanes, each lane having a pair of lane boundaries, each lane boundary representing an edge of the lane;

receiving a request from an autonomous vehicle, the request describing instructions in the form of a low resolution route from a source address;

identifying a geographic area through which the low resolution route passes;

identifying one or more structures of the geographic area from the high-definition map, each structure located outside the lane boundary representing a potential obstacle for a vehicle;

for each of the identified structures, identifying a set of points on the structure based on its vertical distance from a lane boundary of a lane;

generating a polyline representation through the identified set of points;

storing the multi-segment line representation in a high definition map representation of the geographic area as a navigable surface boundary to the lane; and

transmitting a high-definition map representation of the geographic area to a vehicle computing system of an autonomous vehicle driving within the geographic area, wherein the autonomous vehicle makes a determination whether to drive on a navigable surface of a lane outside of the lane boundary.

Background

The present disclosure relates generally to generating routes for vehicles and, more particularly, to generating accurate routes for safe navigation of autonomous vehicles based on high-definition maps with high precision.

Autonomous vehicles (also known as autodrive cars, unmanned cars, a33uto, or robotic cars) drive from a source location to a destination location without requiring a human driver to control and navigate the vehicle. Automation of driving is difficult for several reasons. For example, autonomous vehicles make driving decisions instantaneously using sensors, but vehicle sensors cannot always view all objects. Vehicle sensors may be obscured by corners, rolling hills and other vehicles. Vehicle sensors may not be able to observe something early to make a decision. In addition, lanes and signs may be missing on the road, or signs may be knocked down or hidden by jungles and therefore not detected by the sensors. Furthermore, road signs relating to right of way may not be readily visible for determining where a vehicle may be driven, or turned or moved out of a lane in an emergency, or there are stagnant obstacles that must pass.

Rather than relying on less reliable sensor data, autonomous vehicles may use map data to calculate some of the above information with high confidence. However, conventional maps have several drawbacks that make them difficult to use for autonomous vehicles. To be useful, the geometry of the map and the ability of the vehicle to determine its position in the map need to be highly accurate (e.g., 10cm or less). Conventional maps do not provide the precision required for secure navigation. GPS systems provide an accuracy of about 3-5 meters, but have large error conditions that result in an accuracy of over 100m, which often occurs depending on environmental conditions. This makes it challenging to accurately determine the location of the vehicle using conventional maps and GPS.

Furthermore, conventional maps are created by surveyors who use drivers and specially equipped cars with high resolution sensors that travel around a geographic area and make measurements. Measurements are reclaimed and a set of map editors aggregate maps from the measurements. This process is expensive and time consuming (e.g., it may take several months to complete a map). Thus, maps summarized using this technique have no new data. For example, roads are updated/modified at a frequency of about 5-10% per year. Survey vehicles, however, are expensive and limited in number and therefore cannot capture a large portion of the updates. For example, a survey fleet may include one thousand cars. Even for a state in the united states, a thousand cars cannot regularly keep the map up to date to allow safe autopilot. Thus, conventional techniques for maintaining maps fail to provide correct data that is sufficiently accurate and up-to-date for safe navigation of autonomous vehicles.

Disclosure of Invention

The vehicle computing system uses information from the online system to generate a high-definition map for the autonomous vehicle to travel from a source address to a destination address. The information provided by the online system may include geographic areas covering the route of the low resolution map and lane elements describing the transportation facilities within those geographic areas. The lane elements are created to form a connected graph that indicates a path through the map that is physically navigable. The vehicle computing system analyzes the lane element map and allows the resolution route to produce several potential partial routes or route hypotheses (e.g., partial routes from a starting point to some intermediate point before the ending point). The vehicle computing system develops the potential partial routes with the lowest error by iteratively adding additional lane elements to the potential partial routes until a complete potential partial route is found. The error may be calculated as the maximum deviation between the high-resolution route and the low-resolution route. This may provide an upper limit for error in the high resolution lane element route when the potential route ultimately includes a final destination or end point (e.g., a potential complete route). Other potential routes whose error exceeds the upper limit may be discarded. If the remaining potential routes have an error greater than the upper limit, the search is complete (i.e., this is the final route from the start point to the end point). If some incomplete potential routes have errors below the upper bound, the system may continue searching along those potential routes until the destination lane element is reached or the error exceeds the upper bound. Thus, the system finds the best possible complete route defined by the error measurements (e.g., the maximum distance between the low resolution route and the high resolution route).

In an embodiment, the vehicle computing system receives a low resolution route from a third party service and sends the route to the online system. The vehicle computing system receives, from the online system, a representation of a lane element map describing traffic characteristics within a set of geographic regions covering a low resolution route. Based on the lane element map, the vehicle computing system determines a set of potential partial routes from the source address to the destination address. For each potential partial route, additional connected lane elements are identified and an error measure between the potential partial route and the low resolution route is calculated. In an embodiment, the vehicle computing system does not consider potential partial routes having errors above a threshold.

In an embodiment, in a complete route from a source address to a destination address, the vehicle computing system selects a final route and sends a signal to the controller of the autonomous vehicle to travel the final route.

The system uses information from the online system to generate a high-definition map for use by the autonomous vehicle to travel from a source address to a destination address. The system uses the information collected by the sensors to generate high-definition maps of traffic lanes and features relative to the lanes, such as navigable surface boundaries of each lane. The system uses the detected information to generate a representation of a navigable surface boundary describing a lane that is located beyond the boundary of the lane but within which the vehicle can safely navigate without damaging the physical area of the vehicle (i.e. opening onto the shoulder). By analyzing the navigable surface boundaries, the system can signal the controller of the autonomous vehicle to travel on the navigable surface in an emergency situation, e.g., if an unexpected obstacle is encountered in the lane into which the autonomous vehicle is driving, the vehicle can safely avoid or bypass the obstacle by moving within the navigable surface boundaries of the lane.

In an embodiment, a vehicle computing system receives a high-definition map representation of a geographic area from an online system. A high definition map is composed of one or more geographical areas. Within each geographic region, the system identifies a lane and one or more structures outside the boundaries of the lane. These structures represent potential obstacles such as fences, walls, trees, or buildings. The vehicle computing system identifies a set of points within each structure that are proximate to the lane boundary. The system generates a 3D or 2D polyline representation of the navigable surface boundaries of the lane. The polylines represent attributes that are stored in the high definition map as lanes.

In an embodiment, the system determines a difficulty level of the vehicle traveling on the navigable surface. The difficulty level may be related to the type of navigable surface being considered. For example, a gravel shoulder may be associated with a higher level of difficulty than a paved shoulder. The system further determines the severity of the emergency situation encountered while driving in the lane. The system determines whether to drive on the navigable surface based on the difficulty level of driving on the navigable surface and the severity of the emergency. Thus, the system sends control signals to the vehicle controller of the autonomous vehicle, causing the autonomous vehicle to change lanes, stay in the current lane, or travel on a navigable surface.

Drawings

Fig. 1 illustrates an overall system environment of an HD mapping system interacting with multiple vehicle computing systems, according to an embodiment.

FIG. 2 illustrates a system architecture of a vehicle computing system, according to an embodiment.

FIG. 3 illustrates various instruction layers in the HD map API of the vehicle computing system according to an embodiment.

Fig. 4 shows a system architecture of an online HD mapping system according to an embodiment.

Fig. 5 shows components of an HD map according to an embodiment.

Fig. 6A to 6B illustrate geographical areas defined in an HD map according to an embodiment.

Fig. 7 shows a representation of a lane in an HD map according to an embodiment.

Fig. 8A to 8B show lane elements in the HD map and a relationship between the respective lane elements according to the embodiment.

Fig. 9A shows a representation of navigable surface boundaries in an HD map according to an embodiment.

Fig. 9B illustrates a representation of multiple navigable surface boundaries in an HD map according to an embodiment.

FIG. 10 illustrates a system architecture of a navigable surface module, according to an embodiment.

FIG. 11 shows a flowchart of the overall process for generating navigable surface boundaries, according to an embodiment.

Fig. 12 shows a flowchart of a process for obstacle detection according to an embodiment.

FIG. 13 shows a flowchart of a process for determining navigable surface boundaries, under an embodiment.

Fig. 14 shows a flowchart of a process for implementing navigable surface boundaries in a high definition map, according to an embodiment.

Fig. 15 illustrates a system architecture of a route generation module according to an embodiment.

Fig. 16 shows a flowchart of a route generation process according to an embodiment.

Fig. 17 illustrates an embodiment of a computing machine that may read instructions from a machine-readable medium and execute the instructions in a processor or controller.

The figures depict various embodiments of the present invention for purposes of illustration only. One skilled in the art will readily recognize from the following discussion that alternative embodiments of the structures and methods illustrated herein may be employed without departing from the principles of the invention described herein.

Detailed Description

SUMMARY

Embodiments of the present invention maintain High Definition (HD) maps containing up-to-date information with high precision. The autonomous vehicle may use the HD map to safely navigate to its destination without manual input or with limited manual input. An autonomous vehicle is a vehicle that is capable of sensing its environment and navigating without manual input. Autonomous vehicles may also be referred to herein as "driverless cars," autonomous cars, "or" robotic cars. High definition maps provide high geometric accuracy and additional information that allows a vehicle to identify its location in the map with similar accuracy. HD maps refer to maps that store data with very high accuracy (typically 5-10 cm). Embodiments generate an HD map that contains spatial geometry information about roads on which an autonomous vehicle may travel. Thus, the generated HD map includes information needed for the autonomous vehicle to safely navigate without human intervention. Rather than using expensive and time consuming mapping fleet processing including high resolution sensor equipped vehicles to gather data for HD maps, embodiments of the present invention use data from the lower resolution sensors of autonomous vehicles as they travel in their environment. The vehicle may not have prior map data for these routes, or even map data for the region. Embodiments of the present invention provide location as a service (LaaS) so that autonomous vehicles of different manufacturers can all access the latest map information created by these embodiments of the present invention.