Methods, programs, storage media, and support systems for assisting the operator of one's own agent.

Abstract

To provide a method and a system for supporting an operator by transmitting a risk in a future state to the operator of an ego-agent.SOLUTION: An action plan algorithm is applied by using a first value and a second value of a parameter in a cost function of the action plan algorithm, for determining first and second planned actions. A present state of the ego-agent is identified, for determining an actual action. A personalized parameter value is estimated on the basis of a relation between the first and second planned actions and the actual action of the ego-agent. A parameter correction value is determined on the basis of the personalized parameter value and a target parameter value. The personalized parameter value is corrected by using the parameter correction value, for generating an adapted parameter value. The action plan algorithm is applied on the basis of the adapted parameter value, and a risk is estimated. The risk is then transmitted to an operator.SELECTED DRAWING: Figure 5

Description

[Technical Field]

[0001] This disclosure relates to the field of assisting the operator of an agent. More specifically, a method for assisting the operator of an agent, a corresponding program including program code, a corresponding non-transient computer-readable storage medium, and an assistance system for assisting the operator of an agent are proposed. [Background technology]

[0002] When people operate their agents in environments where other agents are present, there is generally a risk of collisions between agents. Traffic density has increased significantly in recent years, making it extremely difficult for an agent operator to observe all other agents in the environment in order to react in a timely manner to the actions of all other agents so that collisions can be avoided. Furthermore, there are other risks encountered during vehicle operation, such as curve risks or regulation risks. The performance of the latest processors and the availability of sensors such as radar sensors, cameras, and LIDAR sensors that provide the processor with information about the agent's environment make it possible to assist the operator of the agent while the agent is in operation. However, the style in which an agent is operated varies greatly from operator to operator. Therefore, in many situations, the assistance system may suggest actions in a way that distracts the operator's attention or (for example, in the case of semi-autonomous driving) may even interfere with the operator's control actions.

[0003] Several approaches have been attempted to adapt systems to the behavior of individual operators in order to reduce operator annoyance and thus increase the tolerance for system-assisted behavior. For example, U.S. Patent No. 9,623,878 (B2) proposes a personalized driver assistance system that learns driver habits. However, the proposed system adjusts control parameters, such as the target distance of ACC (Adaptive Cruise Control), to match the driver's specific driving style. Unfortunately, such approaches result in assistance systems that reflect individual driving styles but do not influence the driver's behavior.

[0004] Numerous methods have been developed for generating control signals for autonomous or partially autonomous driving. One such method, described in U.S. Patent No. 9,463,797(B2), predicts the future trajectory of the agent and generates several alternative trajectories for the agent from this prediction. Furthermore, a hypothetical future trajectory of another agent is determined, and a risk function over time is calculated based on at least one pair of the agent's trajectory and the other agent's trajectory, or alternatives along the agent's calculated hypothetical trajectory. These risk functions are then incorporated into a risk map, which is subsequently analyzed to generate control signals.

[0005] U.S. Patent No. 10,627,812(B2) relates to another known problem in assistive systems. Accurate predictions require accurate information about the environment used in the prediction algorithm. However, since most information is obtained using one or more sensors, certain areas within the relevant environment may be obscured from accurately assessing traffic conditions. U.S. Patent No. 10,627,812(B2) mitigates the risks of such blind spots by assuming virtual traffic entities in areas where the reliability of sensor data falls below a certain threshold or where sensor data is unavailable at all, allowing virtual traffic entities to interact with the agent's predicted actions. Each risk measure is estimated, and the estimated risk measures are taken into consideration in the agent's control actions.

[0006] U.S. Patent Application Publication 2020 / 0231149(A1) supports the driver of an agent by considering the priority relationship between the agent and at least one other traffic participant and selecting a predictive model for each traffic participant. Thus, the selection of predictive models takes into account the determined priority relationship between the agents involved and thus improves the support by more rigorous prediction of the future behavior of the other agents.

[0007] EP4068153A1 describes an advanced driver assistance system in which environmental features are identified based on information received from sensors, and risk zones for those features are estimated. The features and their risk zones are then displayed on the screen along with the vehicle's environment in a map. [Prior art documents] [Patent Documents]

[0008] [Patent Document 1] U.S. Patent No. 9,623,878(B2) [Patent Document 2] U.S. Patent No. 9,463,797 (B2) [Patent Document 3] U.S. Patent No. 10,627,812(B2) [Patent Document 4] U.S. Patent Application Publication No. 2020 / 0231149(A1) [Patent Document 5] European Patent Application Publication No. 4068153(A1) [Overview of the project] [Problems that the invention aims to solve]

[0009] As the examples of known support systems provided above clearly demonstrate, there are several different methods for estimating risk and providing information about that risk to the agent's operator, or for adapting the agent's control to the operator's habits. However, the problem remains that the style of one agent's operator in a traffic situation may differ significantly from that of another agent. This can lead to misinterpretation of a particular situation and is difficult for both human operators and automated systems. Therefore, it is still necessary to find ways to harmonize the operating styles of multiple agent operators involved in a traffic situation. Such harmonizing would significantly improve the quality and reliability of predictions and directly enhance traffic safety. [Means for solving the problem]

[0010] In this invention, the problem is solved by adapting the estimation of future risks based on the habits of individual operators, but in contrast to what is known from the prior art, it is solved by not only attempting to meet the operator's expectations but also by further considering the differences between the operator's style and the target style. This is achieved by adapting the estimation of future risks so that the transmitted output ultimately educates the operator (driver) toward the target style. If this target style is, for example, an average driving style, and if behavior aligned with the target style is achieved for any operator of the agent, dangerous differences in driving styles between different agents can be mitigated.

[0011] According to the present invention, this is achieved by a method, corresponding program, computer-readable storage medium, and support system for assisting an operator by communicating estimated risks to the operator, wherein the risk estimation begins with an analysis of the operator's driving habits and is adapted to affect the operation of the agent. , information regarding the state of the agent (4) is acquired from the sensor of the agent (4), an action planning algorithm is applied using a risk map (3, 10, 11) generated from at least the predicted trajectories (1, 2) of the agent (4) calculated from the acquired information, at least the first and second planned actions (12, 13) of the agent (4) are determined using at least the first value of the parameter (α, β, γ) and the second value of the parameter (α, β, γ) in the cost function of the action planning algorithm, the current state of the agent (4) is identified using the acquired information, the actual action of the agent (4) is determined, and based on the relationship between at least the first and second planned actions (12, 13) and the actual action of the agent (4), personalized parameter values ​​(α, β, γ) in the parameter (α, β, γ) of the cost function are determined. estimated A step of estimating the personalized parameter value (α estimated Estimating the acceleration values ​​of at least the first and second planned actions (12,13) ​​of the agent (4) includes the step of interpolating the actual action with at least the first and second planned actions (12,13) ​​by comparing the acceleration values ​​of at least the first and second planned actions (12,13) ​​of the agent (4) with the acceleration values ​​of the actual action of the agent (4), Based on at least the difference between the personalized parameter value and the target parameter value in the parameters (α, β, γ) of the cost function, and a correction coefficient, a parameter correction value is determined, and the personalized parameter value is corrected using the parameter correction value, thereby the adapted parameter value (α) of the parameters (α, β, γ) of the cost function. adapted ) generates the adapted parameter value (α adapted ) is applied to the action planning algorithm, and the applied parameter value (α) of the action planning algorithm is applied. adapted The system generates a risk map using the above method to estimate the future risks and communicates the estimated future risks to the operator. For example, by knowing the target parameter value and the estimated personalized parameter value for each operator, it becomes possible to shift the personalized parameter value toward the target parameter value, thereby generating adapted parameter values ​​by correcting the personalized parameter value using parameter correction values.

[0012] It is also possible to determine three or more planned actions using three or more parameter values. Therefore, using these three or more planned actions, improved estimations of personalized parameter values ​​can be made, for example, through better interpolation.

[0013] Next, this adapted parameter value is applied to the action plan algorithm and the risk is estimated. This risk is ultimately communicated to the operator. Therefore, the risk communicated to the operator is more closely aligned with the risk communicated to an operator with an average style. Since the estimation and communication of risk no longer strictly reflect the operator's style, this results in a training effect for the assisted operator and ultimately harmonizes the styles of all assisted operators.

[0014] In the description of the embodiments, reference is made to the accompanying drawings.

Brief Description of the Drawings

[0015] [Figure 1] It is a diagram showing various elements of the cost function used in the action plan. [Figure 2] It is a diagram showing the meaning of the risk map. [Figure 3] It is a diagram showing the relationship between the action plan based on the risk map and the parameter values in the cost function. [Figure 4] It is a block diagram of the process for estimating personalized parameter values. [Figure 5] It is a diagram showing the process of creating an output for communicating risk based on the adapted parameter value. [Figure 6] It is a diagram showing the effect of the correction value.

Mode for Carrying Out the Invention

[0016] According to one embodiment, the personalized parameter value is estimated using interpolation of the actual action with the first and second planned actions. Using interpolation to estimate the personalized parameter value has the advantage that the estimation can be done online, i.e., during the operation of the self-agent. As a result, the guidance remains effective even if the operator's style changes.

[0017] Furthermore, personalized parameter values ​​can be estimated using a comparison of the acceleration of actual behavior with the acceleration of first and second planned behaviors. Acceleration is easily perceptible to the agent itself, and little to no pre-processing of sensor values ​​is required. This reduces computational costs and improves real-time application, for example, in vehicles.

[0018] According to one embodiment of the present invention, a parameter correction value is calculated based on the difference between the target parameter value and the personalized parameter value and a correction coefficient. This has the advantage, on the one hand, of directly considering how much the actual behavior differs from the target behavior, and as a result, a larger correction is made if a larger difference is recognized. On the other hand, the correction value allows for further adjustment regarding the strength of the correction. Specifically, it is possible to determine the value of the correction coefficient based on at least one of the amount of difference between the personalized parameter value and the target parameter value and the operator condition. Thus, starting with the difference between the actual behavior and the target behavior and a constant correction coefficient, many more embodiments can be taken into consideration.

[0019] In another advantageous embodiment, the estimation of personalized parameter values ​​is performed repeatedly during the operation of the agent. By repeatedly estimating personalized parameter values, it becomes possible to adjust the corrections to match the actual behavior that may change during the operation of the agent.

[0020] On the other hand, it may be preferable to estimate personalized parameter values ​​based on multiple actual parameter values ​​estimated based on the relationship between the first and second planned actions and the actual actions of the agent. By considering multiple actual parameter values, it is avoided that the personalized parameter values ​​follow the changes every time the estimated actual parameter values ​​change. For example, a moving average can be calculated from a certain number of actual parameter values, or hysteresis can be applied to filter out small changes in the estimated actual parameter values. In that case, the personalized parameter values ​​are updated only when a significant change in the operator's style is recognized.

[0021] According to another preferred embodiment, the last personalized parameter value estimated during the agent's previous operation is used as the personalized parameter value for the risk estimation of the current operation. If the starting point is the personalized parameter value estimated during the agent's previous operation, it is particularly preferable to update the last personalized parameter value estimated during the vehicle's previous operation with the personalized parameter value estimated during the vehicle's current operation, and then use the updated personalized parameter value as the personalized parameter value for the risk estimation of the current operation. To enable the use of previously estimated personalized parameter values ​​in this manner, the system includes a non-volatile memory in which the most recent personalized parameter value is stored.

[0022] Actual parameter values ​​can be iteratively estimated based on the relationship between the first and second planned actions and the agent's actual actions, and a confidence scale can be calculated for the most recent actual parameter value. If the confidence scale for this last actual parameter value exceeds a confidence threshold, this last actual parameter value is carried over as the personalized parameter value. The confidence threshold quantifies, for example, the frequency of changes in the personalized parameter value.

[0023] Parameter values ​​lie within a certain interval, and the lower bound of the interval is used as the first parameter value and the upper bound of the interval is used as the second parameter value to determine the first and second planned actions. Therefore, when the extreme values ​​defined by the interval limits are used to determine the first and second planned actions, it is readily possible to estimate personalized parameter values ​​with high accuracy using interpolation. Interpolation may use a step function, a linear function, two connected linear functions, or a sigmoid function to estimate the actual parameter values.

[0024] In embodiments where actual parameter values ​​are not processed to derive personalized parameters, it should be noted that personalized parameter values ​​and actual parameter values ​​may be used interchangeably. Estimating personalized parameter values ​​is the same as estimating actual parameter values. However, in descriptions of embodiments where actual parameters are processed to determine personalized parameters, a distinction is made between "personalized" and "actual."

[0025] Preferably, the cost function includes a risk element, as well as at least one of a utility element and a comfort element, each element being weighted by a dedicated parameter. Having such a cost function that includes multiple different elements makes it possible to guide the agent's operator toward a desired action not only in terms of risk but also in terms of other aspects such as utility or comfort. In such a case, the method is performed with respect to the risk element, and further performed with respect to at least one of the utility element and the comfort element.

[0026] Furthermore, the change in the correction value over time can be evaluated, and feedback on the evaluation results is provided to the operator. The correction value is then stored in non-volatile memory, and the process of the correction value over time is evaluated after a predetermined time interval or upon request (e.g., from the operator).

[0027] It should be noted that the present invention can be applied to all kinds of vehicles, including airplanes, boats, and micromobility robots, as well as pedestrians. In most cases, the operator of the agent is the driver of the vehicle, the pilot of the airplane, etc. However, the operator may also control the agent remotely.

[0028] Before describing all the estimation and decision steps in more detail, we will explain the general background of the present invention and its effects on educating operators.

[0029] The individual style of an operator when executing its own agent can be modeled using coefficients that weight the individual elements in the cost function used in the action plan. The cost function is: cost=α*risk-β*utility+γ*comfort It can be described as follows.

[0030] Parameters α, β, and γ adjust the overall cost function and can have values ​​from an interval between a lower and upper bound. Note that the interval does not need to be the same for all parameters. Figure 1 shows the effect of the possible values ​​of the parameters. On the left side, the first element of the cost function is shown, explaining the safety preference.

[0031] The left side of Figure 1 shows the preference for safety. When the first parameter α has a high value, the resulting cost function leads to planned behavior that involves braking in advance to reduce risk. Thus, the first parameter α considers the acceptable risk in the planned behavior. Similarly, the second parameter β defines the preference for utility, which means that, for example, when the value of the second parameter β is set high, the planned behavior will allow for high-speed overtaking. Finally, as shown on the right side of Figure 1, the third parameter γ relates to the preference for comfort, and when the third parameter γ is set high, it leads to planned behavior that is rather wait-and-see rather than frequently changing the actual behavior.

[0032] On the other hand, the cost function, and the planned actions resulting from the execution of the action planning algorithm using the cost function with given parameters, make it possible to model the operator's behavioral style. Therefore, while the values ​​of the three parameters originally define how actions are planned in a particular situation, it is also possible to determine corresponding parameter values ​​that describe the operator's habits and style from the actual observed actions of the agent itself. For example, the first parameter α describes how close the operator (the agent itself) is to other agents, or how long the operator is close to other agents. The second parameter β describes how fast the operator wants to arrive at the destination, or how often the operator changes lanes. Finally, the third parameter γ describes how often the operator accelerates or decelerates, or how strong the acceleration or deceleration is.

[0033] Figure 2 illustrates the principle of how a risk map is used in action planning. Based on the predicted trajectories 1 and 2 of the local agent and another agent, and considering Gaussian and Poisson distributions for spatial and temporal uncertainties that may arise, for example from sensor uncertainties, a risk map 3 can be created in which velocity v is presented over time. The right side of Figure 2 shows the intersecting trajectories 1 and 2 of the local agent 4 and another agent 5, as well as the distribution of the positions of the local agent 4 and the other agent 5 over time, i.e., along trajectories 1 and 2. These time-dependent distributions of the positions of agents 4 and 5 make it possible to determine the predicted collision area 6 between the agents. This predicted collision is included in the risk map 3 shown on the left side of Figure 2, and in addition to the risk, it also includes methods for the local agent 4 to avoid the risk. In this illustration, the collision can be avoided by performing deceleration 7 on the local agent.

[0034] It should be noted that, apart from action plans, risk maps are also used to generate alerts and communicate identified risks to the agent's operators. Such risk communication or alert generation, and action plans using risk maps, are known in the art.

[0035] The trajectory is calculated in a known manner using information about the vehicle state obtained from sensors mounted or carried by agents. This information may be enhanced by inter-agent information. The obtained information is processed by a processor, which performs not only the calculation steps for planning described above, but also, in relation to the present invention, estimation, correction, and all other calculations described below.

[0036] In the following explanation, the term "risk" is primarily used. However, these explanations are equally valid when estimating personalized parameter values ​​β and γ for the elements of "usefulness" and / or "comfort."

[0037] Personalized parameter values ​​can be estimated from the cost function given below, based on the risk map used in the behavioral planner as described above.

[0038] As explained earlier, the planned actions differ depending on the values ​​of the three parameters. The first parameter, α, relates specifically to the risk of collision when another agent is involved. As shown at the top of Figure 3, this can be understood as a measure of the size of the risk zone. The smaller the value of the first parameter α, the smaller the size of the risk zone, and vice versa.

[0039] The center of Figure 3 shows the effect of the value of the second parameter β. This relates to the situation when agent 4 is operating alone, as the value of the second parameter β affects the target velocity. Similarly, the value of the third parameter γ relates to acceleration. These relationships between the values ​​of the three parameters α, β, and γ are used to estimate personalized parameter values. In the following explanation, for the sake of brevity, only the first parameter α will be referred to.

[0040] Figure 4 illustrates the procedure for estimating the personalized value of the first parameter. First, two risk maps are created based on the actual vehicle condition. The first risk map 10 is created using a first value of the first parameter α, and the second risk map 11 is created using a second value of the first parameter α. Preferably, the parameter values ​​used correspond to the lower limit (e.g., α=0, novice driver) and upper limit (e.g., α=1, experienced driver) of the range of values ​​for the first parameter α. Action plans are executed for both created risk maps 10 and 11, resulting in two different planned actions 12 and 13, labeled “Expert Action” and “Defensive Action” in Figure 4.

[0041] Next, interpolation 14 is performed using the current vehicle state, which is also used to determine the planned actions 12 and 13, based on the two planned actions 12 and 13 and the actual actions of the agent, in order to derive the current actions of agent 4 corresponding to the actual control performed by the operator and the resulting actual parameter values ​​from the vehicle state. Note that if no further processing of estimates is performed, the actual parameter values ​​will be the same as the personalized parameter values. This interpolation may be performed, for example, by comparing the acceleration values ​​of the planned actions with the acceleration values ​​of the actual actions.

[0042] The lower right of Figure 4 shows the range of possible values ​​for the first parameter α, and the resulting personalized operator (driver) parameter values ​​for the planned action obtained from a planning algorithm that uses the lower and upper limits as parameter values.

[0043] Once personalized parameter values ​​are estimated, corrections are performed to generate adapted parameter values, which are then used to estimate risk by generating a risk map using the adapted parameter values. This process is shown in Figure 5. After the personalized parameter values ​​are determined in 15, parameter correction values ​​are determined in 16 and applied to the personalized parameter values. The estimation in 15 may also include processing the actual parameter values ​​to determine the personalized parameter values, as described above. By correcting the personalized parameter values, adapted parameter values ​​are generated. These adapted parameter values, indicated as α(k) in the figure, are input into the risk model. The risk model generates warning signals to be communicated to the operator based on the adapted parameter values ​​and vehicle conditions. This communication may be a continuous output to the operator providing warnings on a human-machine interface, or it may be done by semi-autonomously adjusting control inputs from the operator. The calculation of the risk map and the determination of the risk communicated to the operator are no different from prior art methods. However, according to the present invention, the parameter values ​​for applying the risk model or generating the risk map are not only adapted to the operator's style but are also further corrected to achieve an educational effect. For communication, the system may include a display that can show warnings or risk indicators. Alternatively, communication may use a speaker or an audio signal supplied to headphones or a similar communication device.

[0044] Figure 6 shows, in addition to the first value of the first parameter and the second value of the first parameter, the personalized parameter value and its shift by the correction value. The figure also shows the "normal value" of the first parameter, which is a target value that can be selected to correspond to an average skilled operator. This "normal value" constitutes the target value in the described embodiment. However, this value does not have to be the average of all operators. This value can be set to lead all operators to a safer operation of their respective agents.

[0045] The adapted parameter value is α adapted (k)=α estimated +(α normal -α estimated )*k calculated as.

[0046] k is a correction factor that enables adjustment of how strongly the correction is for a given difference between the personalized parameter value α estimated and the target value α normal . The basis for calculating the adapted parameter value α adapted (k) is the deviation between the estimated parameter value α estimated and the normal parameter value α normal , and k can be freely set, for example, in the interval from 0 to 1 according to the desired effect. For example, a correction value k = 0.1 may lead the driver slowly to an average safe driving style.

[0047] To adjust the strength of the effect, the correction factor may be adjusted. For example, when it is recognized that the driver's awareness is low, the driver is sleepy, and / or the driver is stressed, it may be desirable to provide a smaller correction value. The estimated parameter value α estimated and the normal parameter value α normalSince the difference between the two remains the same, the correction coefficient k is adjusted to 0 in that case. On the other hand, if the operator is highly alert and not feeling drowsy or stressed, a stronger correction may be desired. This effect can be achieved by adjusting the coefficient to a higher value (k->1).

[0048] Another way to apply the correction coefficient k is to use the estimated parameter value α estimated and normal parameter value α normal The amount of deviation from the original value is taken into consideration. For example, the correction factor k may be increased when the difference is small and decreased when the difference is large. Reducing the correction factor k when the difference is large avoids the driver becoming overly stressed. On the other hand, increasing the correction factor k when the difference is small ensures that the adaptation of the parameter value is still effective. Otherwise, the closer the operator's style is to a normal style, the more negligible the difference becomes. A table relating the correction value to the calculated difference may be stored in memory and retrieved by a processor that performs all the calculations and decisions described above in order to estimate the risk.

[0049] It should be noted that the explanations given so far do not distinguish between personalized parameter values ​​and actual parameter values ​​for the agent's current behavior. This means that actual parameter values ​​estimated by interpolation, for example, as described above, are used directly as personalized parameter values ​​for further processing and risk estimation communicated to the operator. However, it may be preferable to estimate multiple actual parameter values ​​and determine personalized parameter values ​​based on them. This avoids the direct impact of fluctuations in actual parameter values ​​on risk estimation.

Claims

1. A method for supporting the operator of the agent (4) by communicating the risks in future circumstances to the operator, The steps include obtaining information regarding the state of the self-agent (4) from the sensor of the self-agent (4), The steps include: applying an action planning algorithm using a risk map (3, 10, 11) generated from at least the predicted trajectories (1, 2) of the agent (4) calculated from the acquired information, and determining at least first and second planned actions (12, 13) of the agent (4) using at least a first value of the parameter (α, β, γ) and a second value of the parameter (α, β, γ) in the cost function of the action planning algorithm; Using the acquired information, the current state of the agent (4) is identified, and the actual actions of the agent (4) are determined. Based on the relationship between at least the first and second planned actions (12, 13) and the actual actions of the agent (4), the personalized parameter values ​​(α, β, γ) in the parameters (α, β, γ) of the cost function are determined. ｅｓｔｉｍａｔｅｄ A step of estimating the personalized parameter value (α ｅｓｔｉｍａｔｅｄ Estimating the acceleration of the self-agent (4) includes the step of interpolating the actual action with at least the first and second planned actions (12, 13) by comparing the acceleration values ​​of at least the first and second planned actions (12, 13) of the self-agent (4) with the acceleration values ​​of the actual action of the self-agent (4), The steps include determining a parameter correction value based on at least the difference between the personalized parameter value and the target parameter value in the parameters (α, β, γ) of the cost function, and a correction coefficient, By correcting the personalized parameter values ​​using the parameter correction values, the adapted parameter values ​​(α, β, γ) of the cost function are obtained. ａｄａｐｔｅｄ The steps to generate ) and The applied parameter value (α ａｄａｐｔｅｄ ) is applied to the action planning algorithm, and the applied parameter value (α) of the action planning algorithm is applied. ａｄａｐｔｅｄ The steps include: estimating the future risks by generating a risk map using ) The steps include communicating the estimated future risks to the operator and Methods that include...

2. The value of the correction coefficient is the personalized parameter value (α ｅｓｔｉｍａｔｅｄ The method according to claim 1, wherein the value of the correction coefficient is determined based on the difference between the target parameter value and and / or the operator of the agent (4) is in a state of low awareness, drowsiness, or stress.

3. The method according to claim 1, wherein the estimation of the personalized parameter values ​​is repeatedly performed during the operation of the agent (4).

4. The method according to claim 1, wherein the personalized parameter values ​​are estimated based on a plurality of actual parameter values ​​estimated based on the relationship between at least the first and second planned actions (12, 13) and the actual actions of the agent (4).

5. The method according to claim 1, wherein the last personalized parameter value estimated during a previous operation of the agent (4) is used as the personalized parameter value for estimating the future risk of the current operation.

6. The method according to claim 1, wherein the last personalized parameter value estimated during a previous operation of the agent (4) is updated with the personalized parameter value estimated during the current operation of the agent (4), and the updated personalized parameter value is used as the personalized parameter value for estimating the future risk of the current operation.

7. The method according to claim 1, wherein the actual parameter values ​​are repeatedly estimated based on the relationship between at least the first and second planned actions (12, 13) and the actual actions of the agent (4), a confidence scale is calculated for the most recent actual parameter values, and if the confidence scale for the most recent actual parameter values ​​exceeds a confidence threshold, the most recent actual parameter values ​​are carried over as the personalized parameter values.

8. The method according to claim 1, wherein the values ​​of the parameters (α, β, γ) are within a certain interval, the lower limit of the interval is used as the first value of the parameters (α, β, γ), and the upper limit of the interval is used as the second value of the parameters (α, β, γ).

9. The method according to claim 1, wherein the cost function includes, in addition to a risk element, at least one of a utility element and a comfort element, and each element is weighted by a dedicated parameter (α, β, γ).

10. The method according to claim 9, wherein the method is performed with respect to the risk element and further with respect to at least one of the utility element and the comfort element.

11. The method according to claim 1, wherein the change in the parameter correction value over time is evaluated, and feedback regarding the results of the evaluation is provided to the operator.

12. A program, when executed on a computer or digital signal processor, includes program code means for performing a method to assist an operator of its agent (4) by communicating risks in future circumstances to the operator, wherein the method is The steps include obtaining information regarding the state of the self-agent (4) from the sensor of the self-agent (4), The steps include: applying an action planning algorithm using a risk map (3, 10, 11) generated from at least the predicted trajectories (1, 2) of the agent (4) calculated from the acquired information, and determining at least first and second planned actions (12, 13) of the agent (4) using at least a first value of the parameter (α, β, γ) and a second value of the parameter (α, β, γ) in the cost function of the action planning algorithm; Using the acquired information, the current state of the agent (4) is identified, and the actual actions of the agent (4) are determined. A step of estimating a personalized parameter value (α estimated) in the parameters (α, β, γ) of the cost function based on the relationship between at least the first and second planned actions (12, 13) and the actual actions of the agent (4), wherein estimating the personalized parameter value (α estimated) includes interpolating the actual actions with at least the first and second planned actions (12, 13) by comparing the acceleration values ​​of at least the first and second planned actions (12, 13) of the agent (4) with the acceleration values ​​of the actual actions of the agent (4), The steps include determining a parameter correction value based on at least the difference between the personalized parameter value and the target parameter value in the parameters (α, β, γ) of the cost function, and a correction coefficient, The steps include: generating adapted parameter values ​​(α adapted) for the parameters (α, β, γ) of the cost function by correcting the personalized parameter values ​​using the parameter correction values; The steps include: applying the adapted parameter value (α adapted) to the action planning algorithm and generating a risk map using the adapted parameter value (α adapted) of the action planning algorithm to estimate the future risk; The steps include communicating the estimated future risks to the operator and including, program.

13. A computer-readable storage medium storing a program that causes a computer or digital signal processor to execute a method for assisting an operator of its agent (4) by communicating risks in future circumstances to the operator, wherein the method is The steps include obtaining information regarding the state of the self-agent (4) from the sensor of the self-agent (4), The steps include: applying an action planning algorithm using a risk map (3, 10, 11) generated from at least the predicted trajectories (1, 2) of the agent (4) calculated from the acquired information, and determining at least first and second planned actions (12, 13) of the agent (4) using at least a first value of the parameter (α, β, γ) and a second value of the parameter (α, β, γ) in the cost function of the action planning algorithm; Using the acquired information, the current state of the agent (4) is identified, and the actual actions of the agent (4) are determined. A step of estimating a personalized parameter value (α estimated) in the parameters (α, β, γ) of the cost function based on the relationship between at least the first and second planned actions (12, 13) and the actual actions of the agent (4), wherein estimating the personalized parameter value (α estimated) includes interpolating the actual actions with at least the first and second planned actions (12, 13) by comparing the acceleration values ​​of at least the first and second planned actions (12, 13) of the agent (4) with the acceleration values ​​of the actual actions of the agent (4), The steps include determining a parameter correction value based on at least the difference between the personalized parameter value and the target parameter value in the parameters (α, β, γ) of the cost function, and a correction coefficient, The steps include: generating adapted parameter values ​​(α adapted) for the parameters (α, β, γ) of the cost function by correcting the personalized parameter values ​​using the parameter correction values; The steps include: applying the adapted parameter value (α adapted) to the action planning algorithm and generating a risk map using the adapted parameter value (α adapted) of the action planning algorithm to estimate the future risk; The steps include communicating the estimated future risks to the operator and including, Computer-readable storage medium.

14. A support system for assisting the operator of the agent (4) by communicating risks in future situations to the operator, The aforementioned support system comprises a processor for executing programs, memory, and a display or speaker. The processor that executes the aforementioned program is Information regarding the state of the self-agent (4) is obtained from the sensor of the self-agent (4), Apply an action planning algorithm using a risk map (3, 10, 11) generated from at least the predicted trajectories (1, 2) of the agent (4) calculated from the acquired information, and determine at least the first and second planned actions (12, 13) of the agent (4) using at least the first value of the parameter (α, β, γ) and the second value of the parameter (α, β, γ) in the cost function of the action planning algorithm. Using the acquired information, the current state of the agent (4) is identified, and the actual actions of the agent (4) are determined. Based on the relationship between at least the first and second planned actions (12, 13) and the actual actions of the agent (4), the personalized parameter values ​​(α, β, γ) in the parameters (α, β, γ) of the cost function are determined. ｅｓｔｉｍａｔｅｄ ) is estimated, and the personalized parameter value (α ｅｓｔｉｍａｔｅｄ Estimating the acceleration values ​​of at least the first and second planned actions (12, 13) of the agent (4) includes comparing the acceleration values ​​of at least the first and second planned actions (12, 13) of the agent (4) with the acceleration values ​​of the actual actions of the agent (4), thereby interpolating the actual actions with at least the first and second planned actions (12, 13). A parameter correction value is determined based at least on the difference between the personalized parameter value and the target parameter value in the parameters (α, β, γ) of the cost function, and a correction coefficient. By correcting the personalized parameter value using the parameter correction value, an adapted parameter value (α ａｄａｐｔｅｄ ) of the parameters (α, β, γ) of the cost function is generated. The applied parameter value (α ａｄａｐｔｅｄ ) is applied to the action planning algorithm, and the applied parameter value (α) of the action planning algorithm is applied. ａｄａｐｔｅｄ By generating a risk map using ), the aforementioned future risks are estimated. The future risks estimated by the display or speaker are communicated to the operator. Support system.

Images