JP7338582B2 - Trajectory generation device, trajectory generation method, and trajectory generation program - Google Patents

Description

The disclosure in this specification relates to a technique for predicting the behavior of a moving object.

Patent Literature 1 discloses a system for controlling travel of a vehicle. This system weights multiple judgments regarding the future behavior of the vehicle and outputs the final judgment result.

JP 2019-164729 A

By the way, it is required to predict the behavior of mobile bodies around the own vehicle and determine the behavior of the own vehicle based on the prediction result. When there are multiple prediction results of the behavior of the moving object, it is necessary to arbitrate these prediction results. Patent Literature 1 does not disclose in detail a method of arbitration for multiple judgments. For this reason, with the technique of Patent Document 1, there is a possibility that a plurality of prediction results cannot be effectively used.

An object of the disclosure is to provide a trajectory generation device, a trajectory generation method, and a trajectory generation program that can effectively utilize a plurality of prediction results.

The multiple aspects disclosed in this specification employ different technical means to achieve their respective objectives. In addition, the symbols in parentheses described in the claims and this section are an example showing the correspondence relationship with the specific means described in the embodiment described later as one aspect, and limit the technical scope isn't it.

One of the disclosed trajectory generation devices is a trajectory generation device that generates a travel trajectory scheduled for the own vehicle (Va),
A behavior prediction unit (110) that acquires prediction results from a plurality of prediction models for the behavior of mobile objects around the own vehicle;
an arbitration unit (120) for arbitrating each prediction result based on at least one of safety, comfort, and fitness for the driving environment of each prediction model when driving based on each prediction result;
a trajectory planning unit (130) that plans a travel trajectory based on the arbitrated prediction results;
with
The behavior prediction unit includes an intention estimation model that outputs prediction results after assuming future changes in the physical quantity of motion based on the estimated motion intention of the moving object, and a model that does not assume future changes in the physical quantity of motion based on the estimated motion intention. Get the prediction results from the intention non-estimation model that outputs the prediction results .

One of the disclosed trajectory generation methods is a trajectory generation method executed by a processor (102) to generate a travel trajectory intended for an own vehicle (Va), comprising:
A behavior prediction process (S10) for acquiring prediction results from a plurality of prediction models, respectively, regarding the behavior of mobile bodies around the own vehicle;
An arbitration process (S20, S30, S40, S50, S60) and
a trajectory planning process (S80) for planning a travel trajectory based on the reconciled prediction results;
including
In the behavior prediction process, there is an intention estimation model that outputs prediction results after assuming future changes in the physical quantities of motion based on the estimated motion intention of the moving object, and a model that does not assume future changes in the physical quantities of motion based on the estimated motion intention. Get the prediction results from the intention non-estimation model that outputs the prediction results .

One of the disclosed trajectory generation programs is a trajectory generation program that is stored in a storage medium (101) and includes instructions to be executed by a processor (102) in order to generate a travel trajectory planned for the own vehicle (Va). There is
the instruction is
a behavior prediction process (S10) for acquiring prediction results from a plurality of prediction models for the behavior of mobile bodies around the own vehicle;
An arbitration process (S20, S30, S40, S50, S60) and
a trajectory planning process (S80) for executing a travel trajectory plan based on the arbitrated prediction results;
including
In the behavior prediction process, there is an intention estimation model that outputs prediction results after assuming future changes in the physical quantities of motion based on the estimated motion intention of the moving object, and a model that does not assume future changes in the physical quantities of motion based on the estimated motion intention. A prediction result is obtained from an intention non-estimation model that outputs a prediction result .

According to these disclosures, each prediction result is arbitrated based on at least one of the degree of safety and comfort when driving based on each prediction result, and the suitability of each prediction model for the driving environment. A trajectory plan is generated based on the prediction results. Therefore, a plurality of prediction results arbitrated based on at least one aspect of safety, comfort and suitability can be used for trajectory planning. As described above, a trajectory generation device, a trajectory generation method, and a trajectory generation program that can effectively utilize a plurality of prediction results can be provided.

1 shows a system including a trajectory generator; FIG. It is a block diagram which shows an example of the function which a trajectory generation apparatus has. It is a figure for demonstrating the safety degree evaluation of each prediction result. It is a graph which shows an example of the weight setting based on safety degree. It is a figure for demonstrating the comfort evaluation of each prediction result. It is a graph which shows an example of weight setting based on comfort level. It is a figure for demonstrating the trajectory plan using a prediction result. It is a flowchart which shows an example of the trajectory generation method which a trajectory generation apparatus performs. FIG. 9 is a flow chart showing the details of a conformity evaluation process in FIG. 8; FIG.

(First embodiment)
A trajectory generation device 100 of the first embodiment will be described with reference to FIGS. 1 to 9. FIG. The trajectory generation device 100 is an electronic control device mounted on the own vehicle Va having at least one of an automatic driving function and an advanced driving support function. The trajectory generation device 100 generates the travel trajectory of the own vehicle Va traveling by the functions described above. The trajectory generation device 100 is connected to the perimeter monitoring ECU 10, the locator 20, the vehicle speed sensor 30, and the vehicle control ECU 40 via a communication bus or the like.

The perimeter monitoring ECU 10 is mainly composed of a microcomputer having a processor, memory, I/O, and a bus connecting them, and executes various processes by executing a control program stored in the memory. The surroundings monitoring ECU 10 acquires the detection result from the surroundings monitoring sensor 11 and recognizes the driving environment of the own vehicle based on the detection result. The surroundings monitoring sensor 11 is an autonomous sensor that monitors the surrounding environment of the vehicle A. It includes a perimeter surveillance camera or the like that captures an image of a predetermined range. Further, the perimeter monitoring sensor 11 may include millimeter wave radar, sonar, and the like. Based on the information detected by the surroundings monitoring sensor 11, the surroundings monitoring ECU 10 recognizes the presence or absence of moving objects such as other vehicles Vb and pedestrians, and their relative positions and relative velocities as surroundings information. The surroundings monitoring ECU 10 also determines whether or not the surroundings monitoring sensor 11 has failed. The perimeter monitoring ECU sequentially provides the perimeter information and the presence or absence of failures to the trajectory generator.

The locator 20 generates vehicle position information and the like by composite positioning that combines a plurality of acquired information. The locator 20 includes a GNSS (Global Navigation Satellite System) receiver 21 , an inertial sensor 22 , a map DB 23 and a locator ECU 24 . The GNSS receiver 21 receives positioning signals from a plurality of positioning satellites. The inertial sensor 22 is a sensor that detects inertial force acting on the vehicle A. As shown in FIG. The inertial sensor 22 includes, for example, a gyro sensor and an acceleration sensor.

The map DB 23 is a non-volatile memory and stores map data such as link data, node data, road shapes, and structures. In addition, the map DB 23 stores information such as the position of traffic lights, the type and position of road signs, and the type and position of road signs. The map data may be a three-dimensional map consisting of point groups of feature points of road shapes and structures. Note that the three-dimensional map may be generated based on captured images by REM (Road Experience Management). The map data may also include traffic regulation information, road construction information, weather information, and the like. The map data stored in the map DB 23 are updated regularly or as needed based on the latest information received by the vehicle-mounted communication device.

The locator ECU 24 mainly includes a microcomputer having a processor, a RAM, a memory, an input/output interface, and a bus connecting them. The locator ECU sequentially locates the position of the vehicle A (hereinafter, vehicle position) by combining the positioning signals received by the GNSS receiver, the map data of the map DB, and the measurement results of the inertial sensor. The position of the vehicle may be represented by coordinates of latitude and longitude, for example. It should be noted that the positioning of the own vehicle position may be performed using the traveling distance obtained from the signals sequentially output from the vehicle speed sensor 30 mounted on the vehicle A. FIG. When using a 3D map consisting of point groups of feature points of road shapes and structures as map data, the locator combines this 3D map with the detection results of the surrounding monitoring sensor without using a GNSS receiver. may be used to identify the position of the vehicle. The locator ECU 24 sequentially provides the trajectory generation device with vehicle position information, map data, information detected by the inertial sensor 22, and the like.

The vehicle control ECU 40 is an electronic control unit that performs acceleration/deceleration control and steering control of the vehicle A. As shown in FIG. The vehicle control ECU includes a steering ECU that performs steering control, a power unit control ECU that performs acceleration/deceleration control, a brake ECU, and the like. The vehicle control ECU acquires detection signals output from each sensor such as a steering angle sensor and a vehicle speed sensor mounted on vehicle A, and controls each traveling control such as an electronically controlled throttle, a brake actuator, and an EPS (Electric Power Steering) motor. Outputs control signals to the device. The vehicle control ECU acquires a control instruction for the vehicle A from the driving support device, and controls each driving control device so as to realize autonomous driving or driving support according to the control instruction.

The trajectory generation device 100 has a configuration mainly including a computer having a memory 101, a processor 102, an input/output interface, and a bus connecting these. The processor 102 is hardware for arithmetic processing. The processor 102 includes, as a core, at least one type of CPU (Central Processing Unit), GPU (Graphics Processing Unit), RISC (Reduced Instruction Set Computer)-CPU, and the like.

The memory 101 stores or stores computer-readable programs and data in a non-temporary manner, and includes at least one type of non-transitional physical storage medium (non-transitional storage medium) such as a semiconductor memory, a magnetic medium, an optical medium, or the like. transitory tangible storage medium). The memory 101 stores various programs executed by the processor 102, such as a trajectory generation program to be described later.

Processor 102 executes a plurality of instructions contained in a trajectory generation program stored in memory 101 . Thereby, the trajectory generation device 100 constructs a plurality of functional units for generating the future trajectory of the own vehicle Va. Thus, in the trajectory generation device 100, the program stored in the memory 101 causes the processor 102 to execute a plurality of instructions, thereby constructing a plurality of functional units. Specifically, functional units such as a behavior prediction unit 110, a prediction result arbitration unit 120, and a trajectory planning unit 130 are constructed in the trajectory generation device 100, as shown in FIG.

The behavior prediction unit 110 predicts the behavior of mobile objects around the own vehicle Va. The moving object to be predicted may be another vehicle Vb, a pedestrian, an animal, or the like. The behavior prediction unit 110 inputs input information to a plurality of prediction models and causes each prediction model to output a prediction result. The input information includes, for example, the position information and speed information of the mobile body, the position information, speed information and acceleration information of the own vehicle Va. Further, the input information includes traffic signal information such as the position and lighting color of the traffic signal, sign information such as the position and type of the sign, and road marking information such as the position and type of the road marking (stop line, etc.). Input information to be input to each prediction model may be all common, or may be at least partly different. Each prediction model outputs, for example, the prediction result of the behavior of the moving object as a probability distribution for each position. The behavior prediction unit 110 sequentially provides the output prediction results to the prediction result arbitration unit 120 .

A plurality of prediction models are defined by different policies. The multiple prediction models include, for example, an intention non-estimating model that does not estimate the motion intention of the moving object to be predicted, and multiple intention estimation models that estimate the motion intention of the moving object. The non-intention estimation model outputs a prediction result by linear prediction based on current moving object information. That is, the intention non-estimating model predicts the behavior of the moving object by simply assuming that the current motion physical quantity of the moving object is maintained.

On the other hand, the plurality of intention estimation models predict the behavior of the mobile object by estimating the motion intention, assuming that the current physical quantity of motion of the mobile object will change in the future. The intention estimation model is provided by, for example, a learned model trained by machine learning so as to output behavior prediction of a mobile object in response to input of mobile object information. Multiple intention estimation models include a model optimized for prediction on expressways, a model optimized for prediction on general roads, a model optimized for prediction in specific areas, and so on. That is, each of the multiple intention estimation models is a trained model specialized for a specific driving environment. The intention estimation model may include models optimized for areas with many bicycles and motorcycles, areas with many animals, areas with many rough driving, and the like. A plurality of intention estimation models may be machine-learned based on a common model, or may be based on different models. Note that the intention estimation model may be provided by a rule-based model instead of a learned model learned by machine learning.

The prediction result arbitration unit 120 evaluates the prediction results of each prediction model based on multiple evaluation criteria. Prediction result arbitration unit 120 includes arbitration necessity determination unit 121, safety evaluation unit 122, comfort evaluation unit 123, suitability evaluation unit 124, and determination unit 125 as sub-function units.

The arbitration necessity determining unit 121 determines whether or not it is necessary to arbitrate a plurality of prediction results. The arbitration necessity determination unit 121 determines that arbitration is necessary when the difference between the plurality of prediction results is larger than the allowable range.

Specifically, the arbitration necessity determination unit 121 calculates the variance of a plurality of prediction results, determines that arbitration is necessary when the variance is greater than a threshold, and determines that arbitration is not required when the variance is smaller than the threshold. I judge. Alternatively, the arbitration necessity determination unit 121 may plan the travel trajectory of the own vehicle Va based on each prediction result, and determine that arbitration is not necessary when the matching degree of each travel trajectory exceeds a predetermined value. .

The degree-of-safety evaluation unit 122 evaluates the degree of safety for a plurality of prediction results. Specifically, the safety evaluation unit 122 first generates an evaluation route Re, which is a provisional route, based on each prediction result, and calculates the safety evaluation value EVs of the evaluation route Re.

The safety level evaluation value EVs is calculated based on an index parameter that is an index of the safety level of the evaluation route Re. For example, the safety level evaluation unit 122 uses the time to collision (TTC) between the mobile body and the host vehicle Va as an index parameter. The safety level evaluation unit 122 evaluates the safety level evaluation value EVs smaller as the TTC is smaller. That is, the safety level evaluation unit 122 lowers the safety level evaluation value EVs as the safety level of the evaluation route Re is lower. As an example, the safety level evaluation unit 122 may determine the safety level evaluation value EVs according to the graph of the correspondence relationship between the TTC and the safety level evaluation value EVs shown in FIG.

Further, the safety level evaluation unit 122 sets a weight corresponding to the safety level evaluation value EVs to the prediction result. Specifically, the safety level evaluation unit 122 sets a larger weight for a prediction result with a lower safety level evaluation value EVs. In other words, the safety level evaluation unit 122 sets weights so that prediction results that have a large impact (for example, collision risk) on the safety of the own vehicle Va contribute more to the trajectory planning. In the example shown in FIG. 3, the weight w1 of the prediction result of the first prediction model that blocks the right turn on the evaluation route Re is set larger than a predetermined specified value, and the weight w1 of the prediction result of the second prediction model that does not block the right turn on the evaluation route Re is set. The prediction result weight w2 is not changed from a predetermined specified value. In FIG. 3, the prediction result of each prediction model is represented by grayscale mosaics shown in the traveling direction of the other vehicle Vb. The darker the mosaic, the higher the probability of existence of the other vehicle Vb (the same applies to FIGS. 5 and 7). For example, the safety level evaluation unit 122 determines the weight of each prediction result based on the relationship between the safety level evaluation value EVs and the weight _wn defined in the graph shown in FIG. Specifically, when the safety evaluation value EVs exceeds a predetermined value, the safety evaluation unit 122 sets a constant weight (base weight w_b) regardless of the magnitude of the safety evaluation value. Then, the safety evaluation unit 122 sets a larger weight to the base weight as the safety evaluation value EVs is smaller than a predetermined value.

Note that the safety level evaluation unit 122 may use a value other than the TTC as the index parameter. For example, the safety level evaluation unit 122 may use the vehicle speed or type of the other vehicle Vb as an index parameter. Specifically, the safety level evaluation unit 122 may set the safety level evaluation value EVs lower as the vehicle speed of the other vehicle Vb increases or as the class of the vehicle increases. The degree-of-safety evaluation unit 122 may appropriately change the index parameter according to the current running environment.

The comfort level evaluation unit 123 evaluates the comfort levels of a plurality of prediction results. Specifically, the comfort level evaluation unit 123 generates an evaluation route Re based on each prediction result, and calculates a comfort level evaluation value EVc for the evaluation route Re.

The comfort level evaluation value EVc is calculated based on an index parameter that is an indicator of the comfort level of the evaluation route Re. For example, the comfort level evaluation unit 123 uses the travel time of the evaluation route Re as an index parameter. Specifically, the comfort level evaluation unit 123 determines that the shorter the evaluation route Re can be traveled, the larger the comfort level evaluation value EVc.

Specifically, regarding the behavior of another vehicle Vb traveling in a lane adjacent to the current lane of the host vehicle Va, when the second prediction model outputs a prediction result that the other vehicle Vb will enter the current lane, the evaluation route Re is: , decelerates to avoid approaching another vehicle Vb (see FIG. 5). In this case, the comfort level evaluation unit 123 calculates the comfort level evaluation value EVc of the prediction result of the second prediction model to be smaller than the prediction result of the first prediction model on which the evaluation route Re that overtakes the other vehicle Vb is generated. do. As an example, the comfort level evaluation unit 123 may determine the comfort level evaluation value EVc according to the graph of the correspondence relationship between the travel time of the evaluation route Re and the comfort level evaluation value EVc shown in FIG.

The comfort level evaluation unit 123 may use a value other than the travel time of the evaluation route Re as the index parameter. For example, the comfort level evaluation unit 123 may calculate the comfort level evaluation value EVc to be smaller as the acceleration change is larger, using the lateral acceleration change of the own vehicle Va as an index parameter. The comfort level evaluation unit 123 may appropriately change the index parameter according to the current running environment.

Furthermore, the comfort level evaluation unit 123 sets a weight corresponding to the comfort level evaluation value EVc to the prediction result. Specifically, the comfort level evaluation unit 123 assigns a smaller weight to a prediction result with a lower comfort level evaluation value EVc. In other words, the comfort level evaluation unit 123 sets the weights so that the contribution of prediction results with low comfort levels to the trajectory planning becomes smaller. For example, the comfort level evaluation unit 123 determines the weight of the prediction result based on the relationship between the comfort level evaluation value EVc defined in the graph shown in FIG. 6 and the weight. Specifically, when the comfort evaluation value EVc exceeds a predetermined value, the comfort evaluation unit 123 sets a constant weight (base weight w_b) regardless of the magnitude of the comfort evaluation value EVc. Then, the comfort level evaluation unit 123 sets a smaller weight for the base weight w_b as the comfort level evaluation value EVc is smaller than a predetermined value.

However, the comfort level evaluation unit 123 cancels setting to reduce the weight for prediction results in which the safety level evaluation value EVs is outside the allowable range in the safety level evaluation unit 122, for example, prediction results that fall below the threshold. . In other words, the comfort level evaluation unit 123 stops reducing the degree of contribution to the trajectory planning according to the comfort level evaluation value EVc for prediction results in which the safety level evaluation value EVs is below the threshold.

The suitability evaluation unit 124 sets the weight of each prediction result based on the driving environment of the own vehicle Va. Specifically, the fitness evaluation unit 124 sets the weight according to the fitness between the driving environment of the own vehicle Va and each prediction model.

Adaptability evaluation unit 124 sets a greater weight for the prediction result of the prediction model as the adaptability of the prediction model to the driving environment is higher. In other words, the suitability evaluation unit 124 sets a large weight for the prediction result of the prediction model when the prediction model is driving in a driving environment in which the prediction model is good, and when driving in a driving environment in which the prediction model is weak. , the weight of the prediction result is set small.

For example, the adaptability evaluation unit 124 determines that the intention estimation model adapts more when driving in an area where the difficulty of driving decisions is relatively high than when driving in an area where the difficulty is relatively low. judged to be high. The degree of difficulty of driving judgment may be determined, for example, based on the estimated frequency of appearance of mobile objects that may impede the progress of the own vehicle Va. Adaptability evaluation unit 124 determines the degree of difficulty of driving judgment according to whether the current travel area is a predetermined specific area. The specific area includes an area with a relatively large number of pedestrians crossing, an area with a junction or junction, an area with a roundabout, and the like. When it is determined that the vehicle is traveling in these specific areas, the suitability evaluation unit 124 weights the prediction result based on the intention estimation model higher than when it is determined that the vehicle is not traveling in the specific area.

Further, the conformity evaluation unit 124 determines whether or not at least one of the driving environment specialized for each prediction model, that is, the assumed environment of each prediction model, matches the current driving environment. When it is determined that there is a match, the fitness evaluation unit 124 makes the weight of the prediction result by the matching prediction model larger than when it is determined that there is no match. The specific driving environment of the prediction model is, for example, an area with many bicycles and motorbikes, an area with many animals, an area with many rough driving, and the like.

In addition, the conformity evaluation unit 124 determines that the degree of conformity of the intention estimation model is low as it is estimated that the detection error of the surroundings monitoring sensor 11 increases. For example, the goodness-of-fit evaluation unit estimates that the farther the prediction target exists, the larger the detection error. Also, the conformity evaluation unit 124 estimates that the detection error increases as the road curvature increases. Furthermore, the conformity evaluation unit 124 estimates that the larger the rainfall, the larger the detection error. Also, the conformity evaluation unit 124 estimates that the smaller the light intensity, the larger the detection error. In addition, the adaptability evaluation unit 124 estimates that the detection error is larger when the vehicle is traveling inside a tunnel than when traveling outside the tunnel.

Furthermore, when the current driving scene is a non-target scene that is not targeted for intention estimation, the fitness evaluation unit 124 determines that the fitness of the intention estimation model is lower than when the current driving scene is not a target scene. Excluded scenes include, for example, accident sites, construction sites, sensor limits, and the like.

Further, the fitness evaluation unit 124 determines that the fitness of the intention estimation model is lower when the specific function of the own vehicle Va cannot be used than when it can be used. The specific function is a detection function of the perimeter monitoring sensor 11, an advanced driving support function, and the like.

The determination unit 125 integrates each weighted prediction result to generate an arbitrated prediction result. Assuming that the weight set for the i-th prediction model is wi and the prediction result Pa,i of the i-th prediction model is set, the arbitrated prediction result Pa is obtained based on the following formula (1).

However, when the arbitration necessity determination unit 121 determines that arbitration of the prediction results is not necessary, the determination unit 125 stops arbitration of the prediction results. In this case, the determination unit 125 determines the prediction result of a specific prediction model among the plurality of prediction results as the final prediction result of the moving object. A specific prediction model at this time is set in advance. Alternatively, a specific prediction model may be changed as appropriate according to the running environment of the own vehicle Va.

The trajectory planning unit 130 generates a future trajectory to be followed by the own vehicle Va based on the moving object prediction result determined by the determination unit 125 . The future trajectory is a planned travel trajectory that follows the future behavior, and defines the travel position of the own vehicle Va according to the progress of the own vehicle Va. In addition, the future trajectory may define the speed of the own vehicle Va at each travel position.

Specifically, as shown in FIG. 7, the trajectory planning unit 130 defines a travelable area Ap for the own vehicle Va based on the prediction result. The trajectory planning unit 130 may set, for example, an area in which the existence probability of a moving object is below a threshold as the travelable area Ap. The trajectory planning unit 130 plans the future trajectory so as to fit within this travelable area Ap. The trajectory planning unit 130 sequentially provides the generated trajectory planning to the vehicle control ECU 40 .

Next, the flow of the trajectory generation method executed by the trajectory generation device 100 in cooperation with the functional blocks will be described below with reference to FIGS. In the flow described later, "S" means multiple steps of the flow executed by multiple instructions included in the program.

First, in S10 of FIG. 8, the behavior prediction unit 110 acquires a plurality of prediction results of the moving object based on each prediction model. Next, in S20, the arbitration necessity determination unit 121 determines whether arbitration of the prediction result is necessary. If it is determined that arbitration is necessary, the safety level evaluation unit 122 evaluates the safety level in S30. In subsequent S40, the comfort level evaluation unit 123 evaluates the comfort level. Furthermore, in S50, the fitness evaluation unit 124 evaluates the fitness between the current driving environment and each prediction model. Then, in S60, the determination unit 125 determines the prediction result based on the evaluation results in S30 to S50.

On the other hand, if it is determined in S20 that arbitration of the prediction results is unnecessary, then in S70 the determination unit 125 determines the prediction results based on one prediction model. After the process of S60 or S70, in S80, the trajectory planning section 130 generates the travel trajectory of the own vehicle Va based on the prediction result, and then the series of processes ends.

Next, the flow of the suitability evaluation method by the suitability evaluation unit 124 will be described below with reference to FIG.

First, in S51, the difficulty level of driving judgment is determined for the current driving environment. Next, in S52, the degree of matching of the current driving environment with the specialized region of the prediction model is determined. In the subsequent S53, the magnitude of the detection error of the input information to the prediction model is determined. Furthermore, in S54, it is determined whether or not a specific function of the own vehicle Va has failed. Next, in S55, it is determined whether or not the current driving scene is an exceptional scene. In S56, the weight of each prediction result is determined based on the judgment results of S51 to S55, and the flow returns to FIG.

Note that S10 described above is an example of the "behavior prediction process," S20 to S60 are an example of the "arbitration process," and S80 is an example of the "trajectory planning process."

According to the first embodiment described above, each prediction result is arbitrated based on at least one of the degree of safety and comfort when driving based on each prediction result, and the degree of suitability of each prediction model for the driving environment. , a trajectory plan is generated based on the reconciled prediction results. Therefore, a plurality of prediction results arbitrated based on at least one aspect of safety, comfort and suitability can be used for trajectory planning. As described above, it is possible to effectively utilize a plurality of prediction results.

Further, according to the first embodiment, the lower the degree of safety, the more the degree of contribution of the prediction result to the trajectory planning is arbitrated. Therefore, when the degree of safety when traveling based on the prediction result is relatively low, the prediction result is given more importance in trajectory planning. Therefore, trajectory planning with a greater emphasis on safety can be achieved.

Furthermore, according to the first embodiment, the lower the comfort level, the lower the degree of contribution of the prediction result to the activation plan. In addition, the decrease in the degree of contribution to the trajectory planning based on the degree of comfort is stopped for prediction results in which the degree of safety is out of the allowable range. Therefore, as long as the degree of safety is within the allowable range, the prediction result with the relatively high degree of comfort is emphasized. As a result, it is possible to plan a trajectory that enhances comfort while ensuring the safety of travel.

In addition, according to the first embodiment, the adaptability is evaluated based on the degree of difficulty of driving judgment required in the driving environment, so the prediction result is arbitrated according to the degree of difficulty. In particular, in the first embodiment, the prediction result by the intention estimation model may be emphasized as the degree of difficulty of driving judgment increases. Therefore, the prediction result can be adjusted appropriately in a driving environment more suitable for the intention estimation model.

Further, according to the first embodiment, the degree of conformity is evaluated based on the degree of conformity between the driving environment and the assumed environment of each prediction model, so the prediction results are adjusted according to the degree of conformity. Therefore, prediction results can be adjusted appropriately according to the driving environment suitable for each prediction model.

In addition, according to the first embodiment, since the degree of conformity is evaluated based on the magnitude of the detection error of the information input to the prediction model, the prediction result is arbitrated according to the detection error. . In particular, in the first embodiment, the greater the detection error, the lower the degree of contribution of the prediction result of the intention estimation model to the trajectory planning. Therefore, prediction results can be properly arbitrated when more errors in intention estimation can occur.

Furthermore, according to the first embodiment, the suitability is evaluated based on whether or not the specific function of the own vehicle can be used, so the prediction result is arbitrated according to the availability. In particular, in the first embodiment, the contribution of the prediction result by the intention estimation model to the trajectory planning is reduced when the specific function is unavailable. Therefore, even in a driving environment in which the intention estimation model is not good at, the prediction result can be adjusted appropriately.

Furthermore, according to the first embodiment, based on whether or not the current driving scene is a non-target scene that is not subject to the estimation of the motion intention of the own vehicle, the degree of suitability is evaluated and the evaluation result is evaluated. Reconciliation of prediction results is performed. In particular, in the first embodiment, the degree of contribution of the prediction result of the intention estimation model to the trajectory planning is reduced when the current driving scene is not the target scene. Therefore, prediction results can be appropriately adjusted in a driving environment not covered by the intention estimation model.

Furthermore, according to the first embodiment, it is determined whether or not each prediction result needs to be arbitrated. determined as a result. Therefore, only certain prediction results are used for trajectory planning when reconciliation of prediction results is not required. Therefore, the computational load required for arbitration of prediction results can be reduced.

(Other embodiments)
The disclosure herein is not limited to the illustrated embodiments. The disclosure encompasses the illustrated embodiments and variations thereon by those skilled in the art. For example, the disclosure is not limited to the combinations of parts and/or elements shown in the embodiments. The disclosure can be implemented in various combinations. The disclosure can have additional parts that can be added to the embodiments. The disclosure encompasses omitting parts and/or elements of the embodiments. The disclosure encompasses permutations or combinations of parts and/or elements between one embodiment and another. The disclosed technical scope is not limited to the description of the embodiments. The disclosed technical scope is indicated by the description of the claims, and should be understood to include all changes within the meaning and range of equivalents to the description of the claims. .

In the above-described embodiment, the trajectory generation device 100 arbitrates each prediction result based on all of safety, comfort, and suitability. Just do it.

The trajectory generator 100 may be a dedicated computer that includes at least one of a digital circuit and an analog circuit as a processor. Here, digital circuits in particular include, for example, ASIC (Application Specific Integrated Circuit), FPGA (Field Programmable Gate Array), SOC (System on a Chip), PGA (Programmable Gate Array), and CPLD (Complex Programmable Logic Device). at least one of Such digital circuits may also include memory storing programs.

Trajectory generator 100 may be provided by a single computer or a set of computer resources linked by a data communication device. For example, some of the functions provided by the trajectory generation device 100 in the above-described embodiments may be realized by another ECU.

100 trajectory generation device 101 memory (storage medium) 102 processor 110 behavior prediction unit 120 prediction result arbitration unit (arbitration unit) 130 trajectory planning unit Va own vehicle.

Claims (18)

A trajectory generation device that generates a travel trajectory scheduled for an own vehicle (Va),
a behavior prediction unit (110) that acquires prediction results from a plurality of prediction models for the behavior of mobile objects around the own vehicle;
an arbitration unit (120) for arbitrating each prediction result based on at least one of the degree of safety and comfort when driving based on each prediction result, and the fitness of each prediction model to the driving environment; ,
a trajectory planning unit (130) that plans the travel trajectory based on the arbitrated prediction results;
with
The behavior prediction unit includes an intention estimation model that outputs the prediction result after assuming a future change in the physical quantity of motion based on the estimation of the motion intention of the moving object, and a future change in the physical quantity of motion based on the estimation of the motion intention. and a trajectory generation device that acquires the prediction result from an intention non-estimation model that outputs the prediction result without assuming the . The trajectory generation device according to claim 1, wherein the arbitration unit performs arbitration to increase the degree of contribution of the prediction result to the plan as the degree of safety is lower. The arbitration unit performs arbitration to reduce the degree of contribution of the prediction result to the plan as the comfort level is lower, and for the prediction result in which the safety level is outside an allowable range, the comfort level is used as the basis. The trajectory generation device according to claim 2, wherein the reduction of the degree of contribution to the plan is stopped. The trajectory generation device according to any one of claims 1 to 3, wherein the arbitration unit evaluates the degree of suitability based on the degree of difficulty of driving judgment required in the running environment. The trajectory generation device according to any one of claims 1 to 4, wherein the arbitration unit evaluates the degree of conformity based on the degree of conformity between the driving environment and the assumed environment of each of the prediction models. The trajectory generation device according to any one of claims 1 to 5 , wherein the arbitration unit evaluates the degree of fitness based on a magnitude of detection error of information input to the prediction model. The trajectory generation device according to any one of claims 1 to 6 , wherein the arbitration unit evaluates the suitability based on availability of a specific function of the host vehicle. The arbitration unit evaluates the suitability based on whether or not the current driving scene is a non-target scene for estimating the motion intention of the host vehicle. The trajectory generation device according to any one of items 1 and 2. The arbitration unit determines whether arbitration of each of the prediction results is necessary, and if it is determined that the arbitration is unnecessary, selects a specific prediction result from among the plurality of prediction results as the prediction result that contributes to the plan. A trajectory generator according to any one of claims 1 to 8 , wherein the trajectory generator is determined. A trajectory generation method executed by a processor (102) for generating a planned travel trajectory for an own vehicle (Va), comprising:
a behavior prediction process (S10) for acquiring prediction results from a plurality of prediction models for the behavior of mobile bodies around the own vehicle;
An arbitration process (S20, S30) for arbitrating each of the prediction results based on at least one of the degree of safety and comfort when driving based on each of the prediction results, and the fitness of each of the prediction models to the driving environment. , S40, S50, S60) and
a trajectory planning process (S80) for planning the travel trajectory based on the arbitrated prediction results;
including
In the behavior prediction process, an intention estimation model for outputting the prediction result on the assumption of a future change in the physical quantity of motion based on the estimation of the motion intention of the moving object, and a future change in the physical quantity of motion based on the estimation of the motion intention. and a trajectory generation method for obtaining the prediction result from an intention non-estimating model that outputs the prediction result without assuming the . 11. The trajectory generation method according to claim 10 , wherein in the arbitration process, the lower the degree of safety, the more the degree of contribution of the prediction result to the plan is increased. In the arbitration process, the lower the comfort level, the lower the degree of contribution of the prediction result to the plan. The trajectory generation method according to claim 11 , wherein the reduction of the degree of contribution to the plan is stopped. 13. The trajectory generation method according to any one of claims 10 to 12, wherein in the arbitration process, the suitability is evaluated based on the degree of difficulty of driving decisions required in the driving environment. 14. The trajectory generation method according to any one of claims 10 to 13, wherein in the arbitration process, the degree of conformity is evaluated based on the degree of conformity between the driving environment and the envisaged environment of each of the prediction models. 15. The trajectory generation method according to any one of claims 10 to 14 , wherein the arbitration process evaluates the goodness of fit based on a magnitude of detection error of information input to the prediction model. 16. In the arbitration process, the suitability is evaluated based on whether or not the current driving scene is a non-target scene for estimating the motion intention of the own vehicle. The trajectory generation method according to any one of the items. In the mediation process, it is determined whether or not mediation of each of the prediction results is necessary, and if it is determined that mediation is not necessary, a specific prediction result among the plurality of prediction results is selected as the prediction result that contributes to the plan. A trajectory generation method according to any one of claims 10 to 16 , wherein the trajectory generation method is determined. A trajectory generation program stored in a storage medium (101) and containing instructions to be executed by a processor (102) in order to generate a travel trajectory scheduled for an own vehicle (Va),
Said instruction
a behavior prediction process (S10) for obtaining prediction results from a plurality of prediction models for the behavior of mobile bodies around the own vehicle;
An arbitration process (S20, S30) for arbitrating each of the prediction results based on at least one of the degree of safety and comfort when driving based on each of the prediction results, and the fitness of each of the prediction models to the driving environment. , S40, S50, S60) and
a trajectory planning process (S80) for executing the planning of the traveling trajectory based on the arbitrated prediction results;
including
In the behavior prediction process, an intention estimation model for outputting the prediction result on the assumption of a future change in the physical quantity of motion based on the estimation of the motion intention of the moving object, and a future change in the physical quantity of motion based on the estimation of the motion intention. and a trajectory generation program for obtaining the prediction result from an intention non-estimation model that outputs the prediction result without assuming .

Images