DE102013201935A1 - Method for determining collision-free paths of e.g. car, in predetermined environment, involves classifying first set of candidate nodes or second set of candidate nodes, and classifying path nodes or open nodes or closed nodes - Google Patents

Abstract

The method involves determining a first set of paths (17) depending on a position of a predetermined output node (15) in an occupancy map (10), where the set of paths terminates a first set of candidate nodes (20). A second set of paths is determined depending on a position of chance nodes randomly selected from regions of the occupancy map. The first set of candidate nodes or a second set of candidate nodes is classified. Path nodes (38) or open nodes (36) or closed nodes are classified. Independent claims are also included for the following: (1) a device for determining collision-free paths for a moving object in a predetermined environment (2) a system for determining collision-free paths for a moving object in a predetermined environment.

Description

The invention relates to a method and a device for determining collision-free paths for a moving object. Furthermore, the invention relates to a system for the moving object.

Modern vehicles are becoming increasingly autonomous, that is, the vehicles are able to provide a driving control with less driver intervention. An essential element of autonomous driving is path planning. In most cases, stored digital map data are used to determine the possible route. However, in this case cases may occur in which a sufficiently accurate localization of the vehicle in the map is not possible, for example in very densely populated areas, and / or the digital map data reflect short-term changes in the routes not or not reliable. Such short-term changes may be due, for example, to roadworks.

The object underlying the invention is to provide a method and apparatus for determining collision-free paths for a moving object, as well as a system for the moving object, which efficiently determine collision-free paths in a given environment of the moving object Enable object.

The object is solved by the features of the independent claims. Advantageous developments of the invention are characterized in the subclaims.

The invention is characterized according to a first and second aspect by a method and a corresponding device for determining collision-free paths for a moving object in a predetermined environment of the moving object. Here, the predetermined environment of the mobile object is represented by a predetermined occupancy map. One or more first paths are determined, each starting at a given output node and ending at respective first candidate nodes determined during the determination of the first paths. Once one or more first paths are determined, a plurality of second paths are determined, each beginning at the given parent node and ending at respective second candidate nodes determined during the determination of the second paths. Here, the first and second paths are determined depending on a position of the origin node in the occupancy map and dependent on a predetermined expansion rule by which incremental expansion nodes are determined, which are classified on fulfillment of a predetermined target condition as the first candidate node or as the second candidate node and otherwise be classified as either a path node or an open node or a closed node, where a closed node represents an expansion node that does not expand further. Further, the second paths are respectively determined depending on a position of a respective random node randomly selected from areas of the occupancy map having no obstacle, and open nodes each of which has detected but not yet expanded expansion nodes.

The first and second paths are each determined independently of a given destination node. This enables collision-free paths to be determined for a moving object, for example for a vehicle or a mobile robot, in a given environment, without one or more destination nodes being known. The paths can be determined quickly and reliably. It makes it possible to determine a first path very quickly and to determine a desired number of second paths depending on temporal or computation-related boundary conditions. The expansion rule can be specified such that the expansion steps for the first path or paths are analogous to the expansion steps of an A * algorithm or a hybrid A * algorithm and the expansion steps for the second paths are analogous to the expansion steps of a Rapidly Exploring Random Tree (FIG. RRT) algorithm. The first and second paths can be determined very quickly and reliably.

Such determination of the first and second paths enables the second paths, which extend into different areas of the occupancy map, to be efficiently determined on the basis of the first path (s). The determined path nodes, open nodes and connections of the first path have a tree structure and thus form a search tree. This determined search tree of the first path can be used for the further expansion, so that each node is determined only once. The random nodes are randomly selected from areas of the occupancy map that have no obstruction. Further, the random nodes are selected so that they do not lie on the first path (s). Alternatively, it is possible that the random nodes are selected from the set of open nodes, especially if the set of open nodes already has a large number of nodes. The closed nodes represent expansion nodes that are not expanding further. The closed nodes may be arranged, for example, in areas that have an obstacle.

The common English term for occupancy card is Occupancy Grid. The Occupancy card has a predetermined number of grid cells. The respective grid cells are assigned occupancy values which represent an occupancy probability, an obstacle being characterized by a high occupancy probability. Advantageously, no object recognition and / or object classification is required to determine the occupancy map. The occupancy card can be easily determined depending on sensor data, which are detected by means of a predetermined environment sensor device. The environmental sensor may include one or more radar sensors and / or ultrasonic sensors and / or lidar sensors. Alternatively or additionally, the occupancy map can also be determined as a function of acquired image data that represents the predetermined environment.

For a further processing of the determined paths, for example in a navigation system, a signal can be provided at a given interface, which represents the determined paths.

In an advantageous embodiment of the first and second aspects, the determined first and second paths are combined into different groups depending on predetermined similarity features. Furthermore, a resulting path for the respective group is determined depending on the first and / or second paths of the respective group. This makes it possible to determine at least one collision-free resulting path for a moving object in the given environment, without knowing one or more destination nodes. The resulting paths can be determined quickly and reliably. For a further processing of the resulting paths, for example in a navigation system, a signal representing the resulting paths can be provided at the given interface or at a further interface. The common English term for such a group, in which data objects are summarized depends on given similarity features, is a cluster. Algorithms that are suitable for detecting similarity structures in databases are also referred to as cluster algorithms.

In a further advantageous embodiment of the first and second aspects, the expansion rule comprises determining a predefined number of expansion nodes for a current path node depending on the position of the current path node and a predetermined movement model for the moving object. Further, the expansion rule includes that the total costs are determined for the respective expansion nodes depending on a sum of connection costs of an optimal path from the root node to the respective expansion node. Further, the expansion rule includes that, if none of the expansion nodes satisfies the predetermined target condition, that expansion node is determined to be the current path node having the lowest total cost and the other expansion nodes are classified as either open node or closed node, respectively respective total costs of these expansion nodes, and if the respective expansion node meets the predetermined target condition, the respective expansion node is classified as a first candidate node or second candidate node. This has the advantage that the first paths determined in this way are cost-optimal paths. Such determination of the first and second paths enables the second paths, which extend into different areas of the occupancy map, to be efficiently determined on the basis of the first path (s). For this purpose, the ascertained search tree of the respective first path can be used directly for the further expansion, so that each node is determined only once. The closed nodes can be arranged, for example, in areas which have an obstacle and thus have high occupancy costs and / or connection costs. Advantageously, this makes it possible to quickly and reliably determine collision-free cost-optimized paths, wherein the areas in which the second paths develop depend on a predetermined direction of movement of the moving object in the output node.

In a further advantageous embodiment according to the first and second aspects, a continuation node for the respective second path is determined from a node set comprising the open nodes, which has a smallest distance to the random node. Starting from the continuation node, the respective second path is expanded depending on the predetermined expansion rule. This has the advantage that the continuation node can be easily determined.

In a further advantageous embodiment according to the first and second aspects, a number of expansion steps for determining the first and / or the second paths is predetermined. This has the advantage that the number of expansion steps can be customized in an application-specific manner, for example, depending on predefined reliability requirements for finding collision-free paths and / or a computing unit's performance that is used to calculate the paths.

In a further advantageous embodiment according to the first and second aspects, the resulting path for the respective group is dependent on the respective total costs of the paths respective group, the total cost of the respective path representing a sum of connection costs of direct connections between the respective path nodes located on the respective path. For example, the first or second path can be selected from the group that has the lowest overall cost. Advantageously, the respectively determined resulting path is a cost-optimal path.

In a further advantageous embodiment according to the first and second aspects, a maximum number of groups that are formed is predetermined. This has the advantage that it also sets the number of resulting paths.

In a further advantageous embodiment according to the first and second aspects, the respective groups are determined in such a way that the first and / or second paths of the respective group in each case have a distance measure which falls below a predetermined threshold value. This has the advantage that the choice of the threshold value can specify how similar the respective paths must be in order to be assigned to a group. In particular, it is possible that the threshold value is adjusted depending on whether the predetermined maximum number of groups to be formed has already been reached.

In a further advantageous embodiment according to the first and second aspects, the occupancy card has a predetermined number of grid cells and each grid cell are assigned to predetermined occupancy costs. Further, associated with the movable object is an object mask that is representative of a size of the object with respect to a size of the grid cells. The respective connection costs for the direct connection between a node and an adjacent node are each dependent on the occupancy costs of the grid cells, which are at least affected by the predetermined object mask, when the movable object moves according to the predetermined movement model along the direct connection. Advantageously, the connection costs can be determined very easily. In particular, this can ensure that the paths and / or the resulting path are collision-free. The occupancy costs are preferably representative of an occupancy probability of the respective grid cells. In a simplified case, the grid cell may have a first value if it is considered occupied, and a second value if it is not occupied.

In a further advantageous embodiment according to the first and second aspects, the respective connection costs for the direct connection from one node to an adjacent node depend on a distance of the direct connection and / or the adjacent node from an obstacle. Advantageously, this makes it possible to ensure that the paths and / or the resulting paths have a certain distance from the obstacles.

In a further advantageous embodiment according to the first and second aspects, the respective connection costs for the direct connection from one node to an adjacent node depend on at least one action to be performed by the mobile object for reaching the neighboring node. This advantageously allows connections that require an action of the moving object to be charged at a higher cost, for example, a steering action to be performed by the moving object to change a current axle and wheel position to reach the adjacent node.

In a further advantageous embodiment according to the first and second aspects, the target condition is dependent on a predetermined maximum path length for the paths. This has the advantage that it can be easily checked whether the determined expansion nodes meet the target condition or not.

In a further advantageous embodiment according to the first and second aspects, the maximum path length represents a predefined maximum depth of a search tree starting from the output node. This has the advantage that it can be very easily checked whether the determined expansion nodes fulfill the target condition or not.

According to a third aspect, the invention is characterized by a system for a moving object. The system comprises a computing unit which is designed to determine an occupancy map, depending on predetermined environmental sensor data representing a given environment of the vehicle, and a device according to the second aspect, which is signal-wise coupled to the arithmetic unit for reading in the occupancy map.

Advantageous embodiments of the first and second aspects also apply to the third aspect.

Embodiments of the invention are explained below with reference to the schematic drawings.

Show it:

1 an exemplary block diagram for a system for a moving object,

2 an exemplary flow diagram for a first program for determining collision-free paths,

3 an exemplary flowchart for a second program for determining resulting paths and

4a to 4c schematic illustrations for explaining the determination of first, second and the resulting paths.

Elements of the same construction or function are provided across the figures with the same reference numerals.

1 shows a block diagram for a system 1 for a moving object. The movable object may be, for example, a vehicle or a robot. The vehicle may be, for example, an automobile. The system 1 is arranged, for example, in the vehicle or the robot.

The system 1 has an arithmetic unit 4 for determining an occupancy card 10 on. Further, the system rejects 1 a path determination unit 2 on. The path determination unit 2 is data-technically coupled with the arithmetic unit 4 for the exchange of occupancy card data. The arithmetic unit 4 and the path determination unit 2 For example, each have a processor unit and an associated memory unit. Alternatively, the arithmetic unit 4 and the path determination unit 2 use a common processor unit and / or shared memory unit.

The system 1 For example, one or more environment sensor units 6 assigned. Alternatively, the system can 1 one or more environment sensor units 6 include. The respective environment sensor unit 6 includes, for example, a radar sensor and / or a lidar sensor and / or infrared sensor. Alternatively or additionally, the environment sensor unit 6 have a camera.

The arithmetic unit 4 is formed, depending on the environment sensor unit 6 determined environment data for a given environment of the moving object, for example, the vehicle, an occupancy card 10 to investigate. The occupancy card 10 has a predetermined number of grid cells. The respective grid cells are assigned occupancy values which represent an occupancy probability, an obstacle being characterized by a high occupancy probability.

The path determination unit 2 is formed, for example, depending on such an occupancy card 10 To determine a variety of possible collision-free paths and forward, for example, to a driver assistance system.

The path determination unit 2 may also be referred to as a collision-free path determination device, and is configured to execute a first program for determining collision-free paths. In addition, the path determination unit 2 formed, for example, a second program for determining at least one resulting path 50 perform. An exemplary flowchart for the first program shows 2 , An exemplary flowchart for the second program shows 3 ,

The first program will, as in 2 shown started in a step S10. In a step S12, one or more first paths become 17 each starting at a given parent node 15 and terminate at respective first candidate nodes 20 while determining the first paths 17 be determined. The first path (s) 17 are determined as a function of a predetermined expansion rule, by means of which incremental expansion nodes are determined which, when a predetermined target condition is met, are the first candidate nodes 20 be classified and otherwise as a path node 38 or open node 36 or closed node. Further, the one or more first paths become 17 determines depending on a position of the source node 15 in the occupancy card 10 that by the arithmetic unit 4 provided.

To calculate the first path (s) 17 is starting from the parent node 15 for a respective current path node determines a predetermined number of expansion nodes to a current path node depending on the position of the current path node and a given motion model for the moving object. For example, the moving object moving model may include a moving length traveled by the moving object in an expansion step. Further, the motion model may include steering angles for the moving object and dimensions of the moving object.

For the respective expansion nodes, the total cost is determined depending on a sum of connection costs of an optimal path from the root node 15 to the respective expansion node. If none of the expansion nodes satisfies the predetermined target condition, that expansion node is determined to be the current path node having the lowest total cost, and the other expansion nodes each become either an open node 36 or as closed node classified depending on the respective total cost of these expansion nodes. When the respective expansion node satisfies the predetermined target condition, the respective expansion node becomes the first candidate node 20 classified.

The optimal path is to be understood as a path optimized in connection costs.

The connection costs of a direct connection between a node and an adjacent node include occupancy costs, for example. Each grid cell of the occupancy card 10 are assigned predetermined occupancy costs. Associated with the moving object is an object mask that is representative of a size of the object with respect to a size of the grid cells. The respective connection costs for the direct connection between the node and the adjacent node are each dependent on the occupancy costs of the grid cells, which are at least affected by the predetermined object mask, when the movable object moves according to the predetermined movement model along the direct connection.

The grid cells have a width and height that is preferably smaller than the respective movement length of the movable object.

The connection costs are additionally or alternatively dependent on a distance of the direct connection and / or the adjacent node from an obstacle. Furthermore, the connection costs are additionally or alternatively dependent on at least one action of the movable object to be performed by the mobile object for reaching the neighboring node.

For example, the target condition depends on a given maximum path length for the paths.

For example, the target condition is satisfied for the respective expansion node if the cost-optimal connection is from the originating node 15 until the expansion node has the maximum path length or exceeds a predetermined tolerance range.

The maximum path length represents, for example, a given maximum depth of a search tree starting from the output node 15 , In this case, for example, the target condition is satisfied for the respective expansion node when the cost-optimal connection from the source node 15 up to the expansion node, the predetermined maximum depth in the search tree from the parent node 15 having.

4a shows a result of a first search section in which first the first paths 17 were determined. There are three first paths 17 to recognize each of the output nodes 15 with the respective first candidate node 20 connect. The first three paths 17 are up to a node immediately before the respective first candidate node 20 was determined to be the current path node. Further, the open nodes 36 the first paths 17 to recognize. The movement model here includes, by way of example, a movement length of 6 m and three steering angles, for example -2 °, 0 ° and 2 °. The occupancy card 10 has, for example, 512 × 512 grid cells. The grid cells have an area of 0.2 m × 0.2 m. The target condition was specified with a search tree depth of 10.

It can be clearly seen that the first paths 17 extend straight between the obstacles. The first paths 17 have a minimal overall cost, since they have the greatest possible distance to the obstacles and require no action, in particular no steering action of the moving object.

Once one or more first paths 17 are determined, the in 2 shown first program in a step S14 continued such that a plurality of second paths 34 are determined, each starting at the given output node 15 and terminate at respective second candidate nodes 30 while determining the second paths 34 be determined. The second paths 34 become analogous to the first paths 17 determined as a function of a position of the output node in the occupancy map and depending on the predetermined expansion rule, by means of which the expansion nodes are determined stepwise, the fulfillment of a predetermined target condition as the second candidate node 30 be classified and otherwise as a path node 38 or open node 36 or closed nodes are classified.

The path nodes 38 the respective second paths 34 are further determined depending on a position of each random random node selected from areas of the occupancy map 10 that have no obstacle and open nodes 36 , each representing identified, but not yet expanded, expansion nodes represent.

To determine the second paths 34 For example, the random node is randomly selected from the areas of the occupancy map 10 that have no obstacle. Further, a continuation node for the respective second path 34 determined from a node set containing the open nodes 36 which has a smallest distance to the random node, and starting from the continuation node, depending on the predetermined expansion rule, the respective second path 34 expanded.

4b shows an intermediate result after 100 expansion steps. The second paths 34 extend evenly in not yet examined areas of the occupancy map during the first search section 10 ,

This in 2 For example, in a step S16, the first program shown becomes after executing a predetermined number of expansion steps for the first and second paths 17 . 34 completed.

4c shows an example in which the first program after performing 500 expansion steps for the first and second paths 17 . 34 has ended.

By means of the second program, which in 3 is shown, for example, depending on the determined first and second paths 17 . 34 at least one resulting path 50 determined. The second program is preferably executed following the first program. Alternatively, the first and second programs may be executed in a program.

The second program is started in a step S20. In a step S22, the determined first and second paths 17 . 34 into different groups 40 summarized depending on predetermined similarity features.

Summarizing the first and / or second paths 17 . 34 into the different groups 40 can be done, for example, hierarchically agglomerative. That is, every object, in this case the respective first and second paths 17 . 34 , form an independent group first, and these groups are then progressively grouped into ever larger groups until a desired number of paths, for example all paths, belong to one of the groups. The maximum number of different groups 40 , which are formed, this can for example be specified.

The groups 40 For example, they may be determined such that the first and / or second paths 17 . 34 of the respective groups 40 each have a smaller pitch than the first and / or second paths 17 . 34 other groups.

Alternatively it is possible that the respective groups 40 be determined such that the first and / or second paths 17 . 34 the respective group 40 each have a distance measure to each other, which falls below a predetermined threshold.

In a step S24 becomes a resulting path 50 for the respective group 40 determines depending on the first and / or second paths 17 . 34 the respective group 40 , The resulting path 50 for the respective group 40 For example, it depends on the respective total costs of the paths of the respective group 40 determined, wherein the total cost of the respective path represent a sum of connection costs of direct connections between the respective path nodes 38 that lie on the respective path.

The second program is ended in a step S26.

4c shows next to the determined first and second paths 17 . 34 the determined groups 40 and the resulting paths 50 , There are three groups 40 of paths that correctly reflect the topology of the search space. There are also three collision-free resulting paths 50 to recognize. The determined resulting paths 50 For example, they can be forwarded via a predefined interface to a driver assistance system for determining a roadway.

LIST OF REFERENCE NUMBERS

1

system

2

Location detection unit

4

computer unit

6

Environment sensor unit

10

occupancy map

15

output node

17

first path

20

first candidate node

30

second candidate node

34

second path

36

open knot

37

path node

40

group

50

resulting path

S10 ... S20

program steps

Claims (15)

Method for determining collision-free paths ( 50 ) for a moving object in a given environment of the moving object, which is defined by a predetermined occupancy map ( 10 ), in which - one or more first paths ( 17 ), each starting at a given output node ( 15 ) and terminate at respective first candidate nodes ( 20 ) during the determination of the first paths ( 17 ), and - as soon as one or more first paths ( 17 ), several second paths ( 34 ), each starting at the given output node ( 15 ) and terminate at respective second candidate nodes ( 30 ) during the determination of the second paths ( 34 ), where - the first and second paths ( 17 . 34 ) are determined depending on a position of the source node ( 15 ) in the occupancy card ( 10 ) and dependent on a predetermined expansion rule, by means of which incremental expansion nodes are determined which, when a predetermined target condition is fulfilled, are selected as the first candidate nodes ( 20 ) or as the second candidate nodes ( 30 ) and otherwise either as a path node ( 38 ) or as an open node ( 36 ) or classified as a closed node, where a closed node represents an expansion node that does not expand further, and - the second paths ( 34 ) are determined in each case depending on a position of a randomly selected random node from areas of the occupancy map ( 10 ), which have no obstacle, and open nodes ( 36 ), which each identified, but not yet expanded, expansion nodes represent. The method of claim 1, wherein - the determined first and second paths ( 17 . 34 ) into different groups ( 40 ) are grouped depending on predetermined similarity features and - a resulting path ( 50 ) for the respective group ( 40 ) is determined depending on the first and / or second paths ( 17 . 34 ) of the respective group ( 40 ). The method of any preceding claim, wherein the expansion rule comprises: determining a predetermined number of expansion nodes at a current path node depending on the location of the current path node and a given moving object motion model, the total cost determined for each Expansion node depending on a sum of connection costs of an optimal path from the parent node ( 15 ) to the respective expansion node, - if none of the expansion nodes meets the predetermined target condition, that expansion node is determined to be the current path node having the lowest total costs and the other expansion nodes are each defined either as an open node ( 36 ) or classified as a closed node is dependent on the respective total costs of these expansion nodes, - if the respective expansion node meets the predetermined target condition, the respective expansion node as a first candidate node ( 20 ) or second candidate node ( 30 ) is classified. Method according to one of the preceding claims, in which a continuation node for the respective second path ( 34 ) is determined from a node set containing the open nodes ( 36 ), which has a smallest distance to the random node, and starting from the continuation node, depending on the predetermined expansion rule, the respective second path ( 34 ) is expanded. Method according to one of the preceding claims, wherein a number of expansion steps for determining the first and / or the second paths ( 17 . 34 ) is given. Method according to one of the preceding claims 2 to 5, wherein the resulting path ( 50 ) for the respective group ( 40 ) depending on the respective total costs of the paths of the respective group ( 40 ), the total cost of the respective path representing a sum of connection costs of direct connections between the respective path nodes ( 38 ), which lie on the respective path. Method according to one of the preceding claims 2 to 6, wherein a maximum number of groups ( 40 ), which are formed, is predetermined. Method according to one of the preceding claims 2 to 7, wherein the respective groups ( 40 ) are determined such that the first and / or second paths ( 17 . 34 ) of the respective group ( 40 ) Each have a distance measure to each other, which falls below a predetermined threshold. Method according to one of the preceding claims, in which - the occupancy card ( 10 ) has a predetermined number of grid cells and each grid cell is assigned predetermined occupancy costs, - the object to be mapped is associated with an object mask which is representative of a size of the object with respect to a size of the grid cells, and - the respective connection costs for the direct connection between A node and an adjacent node are each dependent on the occupancy costs of the grid cells, which are at least affected by the predetermined object mask, when the movable object according to the predetermined movement model moves along the direct connection. Method according to one of the preceding claims, in which the respective connection costs for the direct connection from one node to an adjacent node depend on one Distance of direct connection and / or adjacent node from an obstacle. Method according to one of the preceding claims, in which the respective connection costs for the direct connection from one node to an adjacent node depend on at least one action of the mobile object to be performed by the mobile object for reaching the neighboring node. Method according to one of the preceding claims, wherein the target condition is dependent on a predetermined maximum path length for the paths. The method of claim 12, wherein the maximum path length is a predetermined maximum depth of a search tree from the parent node (16). 15 ). Device for determining collision-free paths for a moving object in a given environment, which is designed to carry out a method according to one of Claims 1 to 13. System ( 1 ) for a moving object, comprising: - a computing unit ( 4 ), which is designed, depending on predetermined environment sensor data representing a predetermined environment of the vehicle, an occupancy map ( 10 ), and - a device according to claim 14, which is signal-wise connected to the arithmetic unit ( 4 ) is coupled to read the occupancy card ( 10 ).

Images