JP6708535B2 - Vehicle trajectory graph generator - Google Patents

Description

The present disclosure relates to a technique of generating a trajectory generation graph that has a plurality of nodes and links connecting the nodes and is a basis of a vehicle trajectory.

In Patent Document 1 shown below, in a graph generation device that generates the above-described trajectory generation graph, a large number of nodes are set at preset positions on a road, and links that connect the nodes are set. A technique for generating a graph in which each node is associated with a degree of risk is disclosed.

JP, 2006-154967, A

In the above technique, a trajectory is generated using the same trajectory generation graph without appropriately considering the traveling state of the vehicle in any situation. Therefore, depending on the state of the vehicle, control restrictions may be included. Due to dynamic constraints, it is likely to produce a non-travelable track. That is, there is a problem that the accuracy of trajectory generation on the graph is low, and the deviation between the discrete trajectory generated on the graph and the continuous trajectory that the vehicle can actually take is large. Due to this, when the generated trajectory on the graph is used for control or prediction, there is a possibility that the amount of calculation in the conversion to the continuous trajectory becomes large or the trajectory is converted to an inappropriate trajectory. When a graph is used for learning a risk function based on a set of vehicle trajectory data, which will be described later, the continuous value trajectory data set is converted to a trajectory set on the graph, and the shape of the risk function is used as learning data. In order to optimize, the risk function obtained by learning the trajectory data on the graph with a large deviation from the actual continuous trajectory data does not sufficiently reflect the original continuous trajectory data. Therefore, if a trajectory for control is generated using it, there is a great risk that it will be unsuitable for vehicle control, such as sudden acceleration/deceleration or too close proximity to other vehicles. In addition, when the trajectory for prediction is generated, it is far from the trajectory actually taken by the vehicle, that is, there is a problem that the prediction accuracy is low.

Therefore, it is desirable that the graph generation device that generates the trajectory generation graph that is the basis of the vehicle trajectory be capable of generating the trajectory generation graph for generating a more accurate trajectory. On the other hand, the present disclosure provides a method for guaranteeing that transitions between nodes corresponding to all links on a graph satisfy necessary constraints in the process of generating a graph for generating a trajectory. , It is possible to represent on a graph a discrete trajectory with high approximation accuracy for a continuous trajectory that a vehicle can actually take.

The graph generation device (1) of the present disclosure includes a state acquisition unit (33, S100), a node search unit (33, S220, S230, S240, S250), a link setting unit (33, S270), and a risk handling unit (37). ) Is provided.

The state acquisition unit is configured to acquire the traveling state of the target vehicle whose track is to be obtained. In addition, the node search unit searches for a reachable node that represents one or a plurality of nodes that the target vehicle can reach in a certain time within a range of dynamic constraints from a reference node that represents a reference node, according to the traveling state. To be configured. The link setting unit is configured to set a link from the reference node to each reachable node, and the risk handling unit is configured to associate the risk amount with each reachable node.

That is, in the graph generation device, a reachable node that represents a node that the target vehicle can reach is searched for, a graph including this node is generated, and a risk amount is associated with each reachable node.

Therefore, according to such a graph generation device, when the trajectory is generated, if the optimal link is selected according to the risk amount and the trajectory on the graph is configured, it is possible to obtain a discrete trajectory on the graph and an actual trajectory. The error from the continuous track on which the vehicle can travel can be suppressed to be small. That is, it is possible to improve the accuracy in generating the trajectory.

When learning the risk function used to calculate the risk amount of each node from the vehicle trajectory data as described later, it is common to replace continuous trajectory data with discrete trajectories on the graph before learning. is there. Here, by using the graph of the present disclosure, learning can be performed based on a discrete trajectory having a high approximation accuracy with respect to a continuous trajectory, and thus a more accurate risk function can be obtained.

It should be noted that the reference numerals in parentheses described in this column and the claims indicate the correspondence with the specific means described in the embodiments described below as one aspect, and do not indicate the technical scope of the present disclosure. It is not limited.

It is a block diagram which shows the structure of a driving assistance device. It is a block diagram showing a configuration of trajectory generation. It is a flowchart which shows a graph generation process. It is a program example which shows a graph generation process. It is a flowchart which shows a link addition process. It is a program example which shows a link addition process. It is explanatory drawing which shows the procedure 1 of a link addition process. It is explanatory drawing which shows the procedure 2 of a link addition process. It is explanatory drawing which shows the procedure 3 of a link addition process. It is a flowchart which shows a node addition process. It is a program example which shows a node addition process. It is explanatory drawing which shows the procedure 1 of a node addition process. It is explanatory drawing which shows the procedure 2 of a node addition process. It is explanatory drawing which shows the procedure 3 of a node addition process. It is a block diagram which shows the structure of risk function learning. It is a flowchart which shows a support process.

Hereinafter, embodiments of the present disclosure will be described with reference to the drawings.
[1. Embodiment]
[1-1. Constitution]
The driving support device 1 is mounted on a vehicle such as a passenger car. The driving support device 1 generates a trajectory generation graph having a plurality of nodes and links connecting these nodes according to the traveling state of the target vehicle that represents the vehicle for which the trajectory is to be generated, and generates the trajectory. Driving support is performed using the graph for the vehicle.

In addition, the target vehicle represents an arbitrary vehicle, and a vehicle in which at least a part of the configuration of the driving support device 1 is mounted among the target vehicles is also referred to as a host vehicle. The driving support device 1 includes a control unit 10 as shown in FIG. The driving support device 1 may include various sensors 21, a notification unit 26, and a support execution unit 27.

The various sensors 21 include a vehicle speed sensor that detects the vehicle speed of the host vehicle, an acceleration sensor that detects the acceleration of the host vehicle, an opening sensor that detects the accelerator opening of the host vehicle, and a steering angle sensor that detects the steering angle of the host vehicle. A well-known sensor such as a brake sensor that detects the amount of brake operation of the host vehicle, a position sensor that detects the current position of the host vehicle, a radar or a camera that detects the traffic object information Oj, and the like. The various sensors 21 send the detection result of each value to the control unit 10.

The traffic object means an object related to traffic, and the traffic object information Oj includes information such as the position and movement speed of the object related to traffic. The traffic objects include, for example, other vehicles, white lines, road edges, obstacles, and the like.

The notification unit 26 has a configuration for notifying the driver of information that requires some attention, such as a display that displays the display image generated by the control unit 10 and a speaker that outputs the sound generated by the control unit 10. To be done. The configuration of the notification unit 26 may be a known configuration.

The support execution unit 27 supports driving of the own vehicle by controlling movement vectors of the own vehicle, such as an accelerator opening degree of the own vehicle, a brake operation amount, and a steering angle of the steering wheel according to a command from the control unit 10.

The control unit 10 mainly includes a well-known microcomputer including a CPU 11 and a semiconductor memory (hereinafter, memory 12) such as a RAM, a ROM, and a flash memory. Various functions of the control unit 10 are realized by the CPU 11 executing a program stored in the non-transitional physical recording medium.

In this example, the memory 12 corresponds to a non-transitional substantive recording medium storing a program. By executing this program, the method corresponding to the program is executed. It should be noted that the non-transitional substantive recording medium refers to the recording medium excluding electromagnetic waves. Further, the number of microcomputers forming the control unit 10 may be one or plural.

The control unit 10 has, as a configuration for generating a trajectory among the functional configurations realized by the CPU 11 executing a program, as shown in FIG. 2, an information collection unit 31, a graph generation unit 33, a risk field. A generation unit 37 and an optimum path search unit 38 are provided. As a configuration for obtaining the risk function, as shown in FIG. 15, a trajectory data discretization unit 34, an inverse reinforcement learning unit 35, an already-explained information collection unit 31, and a graph generation unit 33 are provided. As a configuration for generating a trajectory, as shown in FIG. 2, a risk field generation unit 37, an optimum path search unit 38, an already-explained information collection unit 31, and a graph generation unit 33 are provided.

A vehicle track DB 32 and a risk function DB 36 are provided as a part of the memory 12 described above. The database is referred to as DB.
The method of realizing these elements configuring the control unit 10 is not limited to software, and some or all of the elements may be realized by using one or a plurality of hardware. For example, when the above function is realized by an electronic circuit which is hardware, the electronic circuit may be realized by a digital circuit including a large number of logic circuits, an analog circuit, or a combination thereof.

[1-2. processing]
The configuration of the functions of the control unit 10 is realized by information processing.
The process of generating the trajectory will be described with reference to FIG. As shown in FIG. 2, when generating the trajectory, the function of the information collecting unit 31 acquires the traffic object information Oj according to the current traveling environment of the host vehicle.

Then, in the function of the risk field generation unit 37, the risk amount is calculated for each coordinate value according to the traffic object information Oj obtained by the information collection unit 31 and the risk amount at each coordinate obtained by the risk function described later. The risk field associated with is obtained. In this risk field, the risk amount is set to increase as the position near another object such as another vehicle or an obstacle approaches.

The function as the graph generation unit 33 generates the graph information Gr by executing the graph generation process. The graph generation process will be described later. In the graph generated by the graph generation process, the risk amount is not associated with each node. Therefore, the risk amount is associated with each node in the following process.

The function of the optimum path search unit 38 is to set a start point node and an end point node from a plurality of nodes obtained from the graph information Gr, and to move along the link from the start point node to the end point node as a passing node. The path with the smallest sum of the associated risk amounts is searched for, and the path with the smallest sum of the risk amounts is set as the trajectory Wg of the target vehicle.

At this time, the value of the risk field at each node, that is, the sum of the values obtained by inputting the state quantities of reachable nodes into the risk function is associated as the risk amount of each node. The reachable node means one or a plurality of nodes that the target vehicle can reach from the reference node in a predetermined fixed time when the traveling state is taken into consideration. In other words, the reachable node can also be said to be one or a plurality of nodes that can be reached in a predetermined time according to the vehicle traveling state at the reference node and the dynamic constraints.

Next, the graph generation process will be described with reference to FIG. The graph generation process is a process executed by the function of the graph generation unit 33 of the control unit 10, and is started when a command to generate a graph is input. Note that FIG. 4 is pseudo code showing an example of the graph generation processing. It is presented for reference only.

In the graph generation process, as shown in FIG. 3, first, in S100, vehicle information is acquired. The vehicle state here includes the traveling state of the target vehicle. The running state indicates a value related to running of the vehicle, and corresponds to, for example, the speed, acceleration, yaw rate, or the like of the target vehicle.

Then, in S110, the node set V which comprises a graph is initialized. The initial node set V _initial has, for example, a start point node corresponding to the current traveling state of the vehicle, an end point node corresponding to the end state, and a start point node and an end point node as diagonal vertices or included therein. The nodes may be randomly arranged in a rectangle. The generated initial node set is regarded as the node set V at that time. Then, the reference node x _τ is selected. The reference node represents a reference node among the plurality of nodes. For example, an arbitrary node such as a node indicating the current value of the target vehicle can be the reference node selected first. In this processing, starting from the node indicating the current value of the target vehicle, the processing proceeds while sequentially changing the reference node.

Subsequently, in S120, a link addition process is performed. The link addition process is a process of adding and setting a link from the reference node to the reachable node. The details of the link addition process will be described later.

Then, in S130, a node addition process is implemented. The node addition process is a process of adding the number of links so that the number of links set in the reference node becomes the reference number. The details of the node addition processing will be described later.

Subsequently, in S140, the node set outside the road region preset as the trajectory generation region is set and deleted. In other words, even if the position is a position where the vehicle can travel based on the traveling state of the host vehicle, the node located in the area where traveling is determined to be impossible based on the traveling environment such as the actual road condition is deleted.

Subsequently, in S150, it is determined whether the link addition processing and the node addition processing for the newly added node are completed. If the link addition process and the node addition process for the newly added node are not completed, the process returns to S120. If the link addition processing and the node addition processing for the newly added node have been completed, the newly added node is added to the node set V.

Then, in S180, it is determined whether or not there is an unselected reference node x _τ . If the reference node x _tau unselected at S190, selects one of the reference node x _tau unselected, returning to S120. If there is no unselected reference node x _τ , the graph generation process ends.

Next, details of the link addition process will be described.
In the link addition processing, as shown in FIG. 5, a link is added from the reference node to the reachable node and set. 6 is a pseudo code showing an example of the link addition process. It is presented for reference only.

In the link addition process, as shown in FIG. 5, first, in step S210, the number of steps κ is input and the control inputs u _τ+1:τ+κ are set to arbitrary initial values. The step number κ represents the number of times an arbitrary process that is repeatedly performed is executed. Therefore, for example, when u represents acceleration/deceleration and the steering angle, it can be said that the number of steps κ represents a time of 50 ms*κ when the cycle for determining and executing the acceleration/deceleration and the steering angle is 50 ms.

Subsequently, in S220, a node x other than the reference node x _τ designated in S110 is selected from among the plurality of nodes as a link addition availability determination target. Then, in S230, the state x _τ+κ of the _host vehicle after κ steps is calculated according to the reference node x _τ , the control input u _τ+1:τ+κ , and the state equation f.

In the vehicle motion model state equation f representing the movement of one unit time of the vehicle which has received the control input, i.e., the x _t as input the traveling state x _t-1 and the control input u _t of the vehicle to the control of the time step t This is the output function.

Subsequently, S240 in, the vehicle state x _{tau + kappa} and error δx between the position of the node x other than the reference node, and extended Jacobian matrix J and the control so that the error δx is reduced based on the input u _{τ + 1:} Correct _τ+κ . However, the control input u _τ+1:τ+κ is preset with a range that can be taken according to the current traveling state of the vehicle. For example, the upper limit value and the lower limit value of the acceleration and the limit values such as the upper limit value and the lower limit value of the change amount of the acceleration, which are determined based on the motion characteristics/control characteristics of the vehicle and the current speed, are set.

The extended Jacobian matrix J is set as follows.
For example, the state of your vehicle is continuous

And input

For, the equation of state is

The system is represented by.
The extended Jacobian matrix J can be expressed as follows.

In addition,

And Based on this

Then, the control input u is updated to u+ηδu _τ+1:τ+κ using a predetermined learning coefficient η. However, if there is a dimension that exceeds the limit of u as a result of updating, the limit value is used for updating. By repeating the calculation of x _{τ + κ} and δ u and the update of u based on the updated u, x _{τ + κ} is brought close to x within the limit of u.

That is, in S240, the control inputs u _τ+1:τ+κ are changed so that the error δx becomes smaller.
Succeedingly, in S250, it is determined whether or not the control inputs u _τ+1:τ+κ have converged. The case where the control input u _τ+1:τ+κ converges means that the change in the value of the control input u _τ+1:τ+κ is extremely small. When the control input u _τ+1:τ+κ converges, there are cases where the error δx becomes sufficiently small and cases where the control input u _τ+1:τ+κ matches the upper limit value or the lower limit value.

If the control inputs u _τ+1:τ+κ have not converged, the process returns to S230. If the control inputs u _τ+1:τ+κ have converged, the error δx is compared with a preset threshold ε in S260.

If the error δx is greater than or equal to the threshold value ε, the process proceeds to S280 described below. If the error δx is less than the threshold ε, in S270, a link from the reference node x _τ to the node x is newly set. In other words, if the error δx is sufficiently small, the link is established assuming that the node x can be reached.

Subsequently, in S280, it is determined whether or not all the nodes x have been selected as the nodes to be determined as to whether or not to add a link starting from x _τ . If any node x is not selected, another unselected node x is selected in S290, and the process returns to S230.

If all the nodes x have been selected, the link addition processing ends.
In such link addition processing, for example, as shown in FIG. 7, when the node x ₁ is used as the reference node, the range of the position x _τ+κ of the own vehicle that can be reached from the reference node x ₁ by the number of steps κ is shown. The area is obtained as shown by the broken line 8 in FIG. That is, by calculating the error δx while changing the control input u _τ+1:τ+κ , the position x _τ+κ of the own vehicle can be obtained as a region.

However, it is not necessary to directly obtain this area in this processing. In other words, if the error δx<threshold ε is possible, it is known that the target node is in the area. Therefore, the process of determining whether or not the error δx<threshold ε is performed for each node, and the target node is It is determined whether it is inside. Then, as shown in FIG. 9, links are set to the nodes in this area. Such nodes and links form a trajectory generation graph.

Next, details of the node addition processing will be described. In the node addition processing, as shown in FIG. 10, the number of links is added so that the number of links set in the reference node becomes the reference number. Note that FIG. 11 is a pseudo code showing an example of the node addition processing. It is presented for reference only.

In the node addition processing, as shown in FIG. 10, first, in S310, the control inputs u _τ+1:τ+κ are calculated within the range of the limit value. Then, in S320, the control input u _{τ + 1:} calculated by the node according to _{τ + κ} x _{τ + κ} equation of state f, and a new link from the reference node x _tau to node x _{tau + kappa,} a set of links leaving the _x τ E (x τ ) To.

Then, in S330, the number of links |E( _xτ )| from the reference node x _τ is compared with the preset number of links |A|. If the link number |E( _xτ )| is less than the link setting number |A|, the process returns to S310. If the number of links |E(x _τ )|

In such node addition processing, as shown in FIG. 12, for example, as shown in FIG. 12, a node is less than the link setting number |A| within the range of the vehicle position x _{τ+κ that} can be reached from the reference node x ₁ by the step number κ. Even if there is only this, a new node N is added as shown in FIG. For example, if the current number of links is 3 and the number of link settings |A| is 4, one node N is added. Then, as shown in FIG. 14, a new link E(x _τ ) from the reference node x ₁ to the node N is added.

In the graph generation process described above, each reachable node to which a link is set is used as a reference node to search for a new reachable node, and each time a new reachable node is searched for, each reference node is searched. Link to each reachable node from. However, after the link is set, the node generated in the area where the own vehicle cannot travel even if the trajectory is generated is deleted.

By repeating such processing, the graph is generated within the area where the vehicle can travel and within the area where the trajectory can be generated.
Next, the configuration for obtaining the risk function will be described with reference to FIG.

The function as the information collecting unit 31 acquires the track Wr, which is a route along which an arbitrary vehicle has traveled, and the object information Oj regarding the type and position of the traffic object.
When the function of the information collecting unit 31 acquires the track of the host vehicle as the track Wr, the track is estimated using the detection results of the various sensors 21, and the track is set as the track of the host vehicle. When acquiring the track of the other vehicle as the track Wr, for example, a known communication technique is used to acquire the track of the other vehicle accumulated in the other vehicle or the server. The acquired trajectory is recorded in the vehicle trajectory DB 32.

The function as the graph generation unit 33 generates the graph information Gr by executing the graph generation process. The graph generation process will be described later.
The function as the trajectory data discretization unit 34 discretizes the trajectory Wr obtained from the vehicle trajectory DB 32 using the graph information Gr. That is, based on the graph information Gr, each of the trajectories recorded in the vehicle trajectory DB 32 as a travel history is converted into a discrete trajectory having the most similar shape on the graph, and the coordinate values of the nodes on the discrete trajectory are converted. Get as series Wd. For example, for trajectories {s ₁ , ..., s _T } measured with continuous values, the node x _n ∈ V on the graph that minimizes the cost defined by The state sequence Wd.

At this time, when targeting only a specific area where there is no moving body other than the own vehicle, such as a suburban road or a controlled special work place, the trajectory run as the trajectory Wd can be a route on the graph. Assuming that the trajectory is safer than the trajectory, the node on Wd is classified as a node whose risk amount should be lower than other nodes, and the risk amount of each node is adjusted. The risk amount is a value that quantifies the danger to the vehicle, such as the likelihood of an accident occurring when the vehicle travels.

When a trajectory is acquired in a region where a moving body other than the own vehicle exists, or when trajectories in multiple regions are mixed, the risk amount is not directly assigned to the node, and an object-dependent risk function is used by using inverse reinforcement learning. It is sufficient to obtain a risk function that can calculate the risk amount according to each situation by adjusting. In the function as the inverse reinforcement learning unit 35, inverse reinforcement learning is performed using a set of discrete trajectories Wd obtained as an output by the trajectory data discretization unit 34. Inverse reinforcement learning is an algorithm for obtaining a risk function from a dataset of discrete trajectories. Here, the risk function that represents the function for obtaining the risk amount of a node by inputting the state quantity corresponding to the node on the trajectory To get

The inverse reinforcement learning is a well-known technique, and its details are disclosed in the following documents, for example.
Reference 1: "Algorithms for inverse reinforcement learning", Ng, A. and Russel, S., ICML, 2000,
Reference 2: “Maximum Entropy Inverse Reinforcement Learning”, Ziebart B., et al., AAAI, 2008.

Note that the risk function in the present disclosure may have different names such as a reward function, a cost function, and a potential function depending on the literature.
The risk function obtained by the function of the inverse reinforcement learning unit 35 is recorded in the risk function DB 36. When the risk function obtained by the inverse reinforcement learning is used, the trajectory Wr actually traveled is set to be a safer trajectory with a lower risk amount.

The risk function is a function that receives the running state as an input and returns a risk amount. For example, as shown in feature descriptor f(s) in Reference 3 below, the risk is generally considered to be high using an exponential function. It may be designed so that the value becomes high in the state.

Reference 3: “Modeling Risk Anticipation and Defensive Driving on Residential Roads with Inverse Reinforcement Learning”, Simosaka, M., et al., 2014 IEEE 17th International Conference on Intelligent Transportation Systems (ITSC)).

In Reference 3 above, the risk function is set as follows.

In this formula, by preparing multiple values with different values of s and Σ, it is possible to use in situations where the vehicle is running at excessively high or low speeds, at high speeds near intersections without traffic lights, at low speeds in straight sections, etc. A risk function that reduces the value is prepared. However, s in this notation indicates the running state.

Further, in the above-mentioned Patent Document 1, the risk function due to an obstacle i such as another vehicle is defined as the following, as "a direct contribution term due to the presence of the obstacle i among the elements constituting the risk field". Is designed to

However, O_i is designed as a region occupied by the obstacle i, a_i and b_i are parameters designed by the designer, and x_i(t) and y_i(t) are predicted positions at the time t of the obstacle i. is there. For each obstacle,

Is prepared, and the total sum is used as the risk of the node.
These risk functions are input with the vehicle speed and the position of the vehicle/obstacle as the traveling state, but may be designed based on acceleration or relative speed, for example. The obtained risk function is recorded in the risk function DB 36.

The risk amount of each node in the graph is obtained by inputting the running state corresponding to the node into the risk function. However, in Patent Document 1, “sum of risks” corresponds to the risk amount. The risk amount of the route on the graph is obtained as the sum of the risk amounts of the nodes included in the route.

The graph information Gr and the trajectory Wg thus obtained can be used in any processing. For example, when the trajectory Wg is used, it is possible to carry out a support process as shown in FIG.

When the graph information Gr is provided to another device, the process of associating the risk amount with each node may be performed by another device. That is, the configuration in which the risk amount is associated with each node is not an essential configuration in the driving support device 1.

In the support process, as shown in FIG. 16, first, in S410, the current position of the vehicle is acquired. Then, in S420, the vehicle information about the own vehicle is acquired. As the processing for acquiring the vehicle information, the same processing as the above-described processing of S100 can be adopted.

Then, in S430, a trajectory generation process is performed. The trajectory generation process is a process corresponding to the process for generating the trajectory described with reference to FIG.
Subsequently, in S440, for example, a control amount for causing the host vehicle to travel along the track Wg is calculated. The control amount here may include parameters such as an accelerator opening degree, a brake amount, and a steering amount, which may influence the traveling of the host vehicle. Further, for example, the control amount for the alarm output from the notification unit 26 may be calculated according to the degree of deviation between the track Wg and the actual track of the own vehicle.

Then, in S450, the calculated control amount is output to the notification unit 26 and the support execution unit 27. By this processing, support according to the trajectory Wg is realized.
When such processing ends, the support processing ends.

[1-3. effect]
According to the first embodiment described in detail above, the following effects are achieved.
(1a) In the driving support device 1 described above, the graph generation unit 33 acquires the traveling state of the target vehicle for which the trajectory is to be obtained. Further, the graph generation unit 33 searches for a reachable node that represents one or a plurality of nodes that the target vehicle can reach from a reference node that represents a reference node among the plurality of nodes, according to the traveling state. Further, the graph generation unit 33 sets a link to each reachable node from the reference node, and the optimal path search unit 38 associates each reachable node with a risk amount.

That is, the driving assistance apparatus 1 searches for reachable nodes that represent reachable nodes of the target vehicle, generates a graph including the nodes, and associates the reachable nodes with the risk amounts.

Therefore, according to such a driving support device 1, when a trajectory is generated, there is a high possibility that a travelable trajectory can be generated by selecting an optimum link according to the risk amount. Therefore, the accuracy in generating the trajectory can be improved.

(1b) In the driving support device 1 described above, the information collecting unit 31 acquires the trajectory Wr, which is the route on which an arbitrary vehicle has traveled, and the trajectory data discretizing unit 34 and the inverse reinforcement learning unit 35 are located on the route. Reduce the amount of risk to reachable nodes.

According to the driving support device 1 as described above, the risk amount of the reachable node located on the route on which the arbitrary vehicle travels is reduced, and thus the trajectory is generated assuming that the route on which the vehicle actually travels is safer. can do.

(1c) In the driving support device 1 described above, the optimum path searching unit 38 sets a value obtained by inputting the coordinates of the reachable node into a risk function prepared in advance to obtain a risk amount as a risk amount. Corresponding to each other, the inverse reinforcement learning unit 35 corrects the risk function according to the route.

According to such a driving support device 1, the risk function can be optimized by correcting the risk function according to the route traveled by an arbitrary vehicle, and thus an appropriate risk function can be obtained.
(1d) In the driving support device 1 described above, the optimum path searching unit 38 associates the value obtained by inputting the coordinates of the reachable node into the risk function prepared in advance to obtain the risk amount as the risk amount. ..

According to such a driving support device 1, the risk amount can be calculated only by inputting the coordinates of the reachable node to the risk function, and thus the risk amount can be calculated by a simple process.
(1e) In the driving support device 1 described above, the inverse reinforcement learning unit 35 uses the trajectory generation graph generated by the driving support device 1 to search for a reachable node located on the route along which an arbitrary vehicle has traveled, The risk function is corrected according to the search result.

According to such a driving support device 1, the trajectory is generated using the trajectory generation graph in accordance with the risk function learned by using the trajectory generation graph generated by the driving support device 1, so that a more accurate trajectory is obtained. Can be generated.

(1f) In the driving support device 1 described above, the inverse reinforcement learning unit 35 reduces the risk amount of reachable nodes located on the route by inverse reinforcement learning.
According to such a driving support device 1, the risk amount of reachable nodes can be optimized by inverse reinforcement learning.

(1g) In the driving support device 1 described above, the graph generation unit 33 searches for a new reachable node with each of the reachable nodes for which a link is set as a reference node, and a new reachable node is searched. Then, a link is set from each reference node to each reachable node.

According to such a driving support device 1, it is possible to generate a graph in which links are set to each of a large number of nodes.
(1h) In the driving support device 1 described above, the graph generation unit 33 compares the number of links set in the reference node with a preset reference number, and when the number of links is less than the reference number, A node is added in the feasible area representing the area that the target vehicle can reach so that the number of links becomes the reference number, and links are added to the added node from the reference node and set.

According to the driving support device 1 as described above, when the number of links set to the reference node is less than the reference number, the node is added in the reachable area that is the reachable area, and the link is also added. Not only can the number of links be prevented from becoming excessive, but the number of links in the feasible region can also be prevented from becoming too small. Therefore, it is possible to generate a more appropriate graph for generating the trajectory.

When performing both the process of adding a link to the node added from the reference node and setting the link and the process of setting the link from the reference node to the reachable node, the order of these processes is It doesn't matter.

(1i) In the driving support device 1 described above, the optimum path search unit 38 sets a start point node and an end point node from among a plurality of nodes, and moves from the start point node to the end point node along the link as a passage. It is configured to search for a path having the smallest sum of risk amounts associated with the reachable nodes, and to set the path having the smallest sum of risk amounts as the trajectory of the target vehicle.

According to the driving support device 1 as described above, a path having the smallest sum of risk amounts is searched for and set as the path of the target vehicle. Therefore, the safest path that can be traveled is set. You can

[2. Other Embodiments]
Although the embodiments of the present disclosure have been described above, the present disclosure is not limited to the above-described embodiments, and various modifications can be implemented.

(2a) In the above embodiment, the risk function has been described as a function for obtaining the risk amount for each coordinate value on the road according to the running state by inputting the coordinate value indicating the trajectory, the risk amount, and the like. Other parameters such as

(2b) In the above embodiment, when the number of links is less than the reference number, a node is added in the feasible area representing the area that the vehicle can reach such that the number of links becomes the reference number. Although configured, this configuration may be omitted.

(2c) In the above-described embodiment, the graph information Gr is used to generate the track that the host vehicle should travel in the future, but the graph information Gr may be used when predicting the track of another vehicle. When the graph information Gr is used, one associated with a risk amount may be used, or one associated with no risk amount may be used.

(2d) A plurality of functions of one constituent element in the above embodiment may be realized by a plurality of constituent elements, or one function of one constituent element may be realized by a plurality of constituent elements. .. Further, a plurality of functions of a plurality of constituent elements may be realized by one constituent element, or one function realized by a plurality of constituent elements may be realized by one constituent element. Moreover, you may omit a part of structure of the said embodiment. Further, at least a part of the configuration of the above-described embodiment may be added to or replaced with the configuration of the other above-described embodiment. Note that all aspects included in the technical idea specified by the wording recited in the claims are embodiments of the present disclosure.

(2e) In addition to the driving support device 1 described above, a system including the driving support device 1 as a component, a program for causing a computer to function as the driving support device 1, and a non-transitional type such as a semiconductor memory in which the program is recorded. The present disclosure can be realized in various forms such as a practical recording medium and a driving support method.

[3. Correspondence between Configuration of Embodiment and Configuration of Present Disclosure]
In the embodiment, the driving support device 1 corresponds to the graph generation device in the present disclosure, and in the embodiment, the information collection unit 31 corresponds to the route acquisition unit in the present disclosure. Further, in the embodiment, the trajectory data discretization unit 34 and the inverse reinforcement learning unit 35 correspond to the risk reduction unit and the function correction unit in the present disclosure, and in the embodiment, the optimum path search unit 38 includes the trajectory setting unit and the trajectory setting unit in the present disclosure. Corresponds to the risk response section.

In addition, the process of S100 of the processes executed by the control unit 10 in the embodiment corresponds to the state acquisition unit according to the present disclosure, and the processes of S220, S230, S240, and S250 in the embodiment include the node search unit according to the present disclosure. Equivalent to. Further, in the embodiment, the processing of the graph generation unit 33 and S270 corresponds to the link setting unit in the present disclosure, and in the embodiment, the processing of S320 and S330 corresponds to the additional setting unit in the present disclosure.

DESCRIPTION OF SYMBOLS 1... Driving support device, 10... Control part, 11... CPU, 12... Memory, 21... Various sensors, 26... Notification part, 27... Assistance implementation part, 31... Information collection part, 32... Vehicle track DB, 33... Graph Generation unit, 34... Orbital data discretization unit, 35... Inverse reinforcement learning unit, 36... Risk function DB, 37... Risk field generation unit, 38... Optimal path search unit.

Claims (9)

A graph generation device (1) configured to generate a trajectory generation graph, which has a plurality of nodes and links connecting the nodes and is a basis of a vehicle trajectory, comprising:
A state acquisition unit (S100) configured to acquire the traveling state of the target vehicle for which the trajectory is to be determined;
A node search unit (S220, S230, S240) configured to search for a reachable node that represents one or more nodes that the target vehicle can reach from a reference node that represents a reference node according to the traveling state. , S250),
A link setting unit (S270) configured to set a link from the reference node to each of the reachable nodes,
A risk handling unit (38) configured to associate a risk amount with each of the reachable nodes;
The number of links set in the reference node is compared with a preset reference number, and when the number of links is less than the reference number, the number of links becomes the reference number, An additional setting unit (S320, S330) configured to add a node and set a link to the node added from the reference node in a feasible region representing a reachable region of the target vehicle. When,
A start point node and an end point node are set from among the plurality of nodes, and the sum of the risk amounts associated with the reachable nodes passing therethrough is assumed to move from the start point node to the end point node along the link. A path search unit (38) for searching for a minimum path,
A graph generation device equipped with. The graph generation device according to claim 1, wherein
A route acquisition unit (31) configured to acquire a route traveled by an arbitrary vehicle;
A risk reduction unit (34, 35) configured to reduce the risk amount of reachable nodes located on the path;
A graph generation device further comprising: A graph generation device (1) configured to generate a trajectory generation graph, which has a plurality of nodes and links connecting the nodes and is a basis of a vehicle trajectory, comprising:
A state acquisition unit (S100) configured to acquire the traveling state of the target vehicle for which the trajectory is to be determined;
A node search unit (S220, S230, S240) configured to search for a reachable node that represents one or more nodes that the target vehicle can reach from a reference node that represents a reference node according to the traveling state. , S250),
A link setting unit (S270) configured to set a link from the reference node to each of the reachable nodes,
A risk handling unit (38) configured to associate a risk amount with each of the reachable nodes;
A route acquisition unit (31) configured to acquire a route traveled by an arbitrary vehicle;
A risk reduction unit (34, 35) configured to reduce the risk amount of reachable nodes located on the path;
A start point node and an end point node are set from among the plurality of nodes, and the sum of the risk amounts associated with the reachable nodes passing therethrough is assumed to move from the start point node to the end point node along the link. A path search unit (38) for searching for a minimum path,
A graph generation device equipped with. The graph generation device according to claim 2 or 3 , wherein
The risk handling unit is configured to associate a value obtained by inputting the coordinates of the reachable node into a risk function prepared in advance to obtain a risk amount as the risk amount,
The risk reduction unit, configured graph generator so as to correct the risk function depending on said path. The graph generation device according to claim 4 , wherein
The risk reducing unit searches a reachable node located on the route using the trajectory generating graph generated by the graph generating device, and corrects the risk function according to the search result. Generator. A graph generating apparatus according to Izu Re one of claims 2 to claim 5,
The risk reduction unit is a graph generation device configured to reduce the risk amount of reachable nodes located on the route by inverse reinforcement learning. The graph generation device according to any one of claims 1 to 3 , wherein:
The graph generation device configured to associate the value obtained by inputting the coordinates of the reachable node into a risk function prepared in advance to obtain the risk amount as the risk amount. The graph generation device according to any one of claims 1 to 7 , wherein:
The node search unit is configured to search for a new reachable node with each of the reachable nodes set with a link by the link setting unit as the reference node,
The graph generation device configured such that the link setting unit sets a link from each reference node to each reachable node every time the new reachable node is searched. The graph generation device according to any one of claims 1 to 8, wherein:
The path searching unit, configured graph generator to set the path the sum of the amount of risk is minimum as a track of the target vehicle.