An Improved Cognition-Driven Decision-Making Approach for Human Decision Prediction

Abstract

This invention discloses an improved cognitive-driven decision-making method for predicting human decisions, comprising: acquiring the current situation; based on the current situation, performing feature recognition based on a preset memory model to obtain a prototype corresponding to the current situation; based on the prototype corresponding to the current situation, performing memory retrieval to obtain memory retrieval information; based on the memory retrieval information, performing an expectation check to obtain memory retrieval information that passes the check; and based on the memory retrieval information that passes the check, performing a simulated action to obtain the final decision action. On the one hand, the preset memory model can effectively store continuous memory information and can match the corresponding prototype based on continuous information in the current situation; on the other hand, this invention uses a joint probability density function, which has the ability to simulate the uncertainty of human decision-making.

Description

Technical Field

[0001] This invention relates to an improved cognitive-driven decision-making method for predicting human decisions, belonging to the field of human decision-making theory technology. Background Technology

[0002] Predicting human decision-making is a crucial need in various fields, including industrial production, aerospace, and economic management. Human decision-making is highly complex, influenced not only by external factors such as the completeness of information, workload, and time constraints, but also by internal factors such as experience, cognitive differences, and rationality. However, the patterns of these external and internal factors' effects on human decision-making are extremely complex, and these factors themselves are often difficult to quantify or define. Therefore, accurately predicting human decisions is very difficult and complex. Current prediction methods often only apply to specific scenarios, and their effectiveness rapidly diminishes beyond the pre-defined conditions of that scenario. The Cognition-Driven Decision-Making Method, derived from observing and analyzing a large number of actual human decision-making processes, exhibits extremely high performance in predicting decisions made by individuals under high workloads or facing time-sensitive situations. Experimental results show that in these scenarios, over 95% of decision-making processes conform to the Cognition-Driven Decision-Making Method. Even in decisions where the above conditions are not met, the Cognition-Driven Decision-Making Method still demonstrates considerable predictive performance. Because cognitive-driven decision-making methods have good predictive performance and strong scalability, they have become a hot topic in the field of human decision prediction and have shown great application potential in many fields such as military command, fire rescue, building evacuation, and automobile / aircraft driving.

[0003] The cognitive-driven decision-making approach was initially proposed as a framework, describing the steps involved in human decision-making. While numerous researchers have subsequently developed different mathematical descriptions of these methods for various decision problems, resulting in many important and practical cognitive-driven decision-making methods, existing approaches face the following pressing issues as real-world decision problems become increasingly complex and their applications expand:

[0004] (1) As decision-making information becomes increasingly abundant, people encounter more and more information with continuous characteristics, such as flight altitude and latitude and longitude. According to the idea of ​​cognitive-driven decision-making, this decision-making information will be stored in memory in some form, and when encountering a new situation, similar situations in memory will be matched based on the decision-making information. For continuous information, how to store it in memory and how to match the situations in memory based on continuous information are still unsolved problems.

[0005] (2) As decision-making systems become increasingly complex, especially with the rapidly growing need to model multiple people's current situations, the inherent uncertainty of individual decisions will significantly affect the system's outcome through complex transmission processes. How to consider the uncertainty of individual decisions in the methodology is an important problem that current cognitive-dominated decision-making methods urgently need to solve.

[0006] (3) With the deepening of biological and psychological research, a large number of documents have confirmed that frequency / probability information is ubiquitous in human memory, and this frequency / probability information extensively influences human decision-making processes. The core of the cognitive-driven decision-making method is to make decisions by matching similar scenarios in memory. Whether memory incorporates this frequency / probability information not only affects the entire decision-making process, but more importantly, it determines whether the decision-making method conforms to the actual human decision-making process at the mechanistic level. Therefore, how to obtain this probability / frequency information, in what form this information will be stored in memory, and how this information will play a role in the entire decision-making process are currently difficult problems that need to be solved.

[0007] The information disclosed in this background section is intended only to enhance the understanding of the overall background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of the prior art and provide an improved cognitive-driven decision-making method for predicting human decisions. On the one hand, the preset memory model can effectively store continuous memory information and match the corresponding prototype according to the continuous information in the current situation. On the other hand, this invention adopts a joint probability density function, which has the ability to simulate the uncertainty of human decision-making.

[0009] To achieve the above objectives, the present invention is implemented using the following technical solution:

[0010] This invention discloses an improved cognitive-driven decision-making method for predicting human decisions, comprising:

[0011] Obtain the current context;

[0012] Based on the current scenario, feature recognition is performed using a preset memory model to obtain the memory prototype corresponding to the current scenario;

[0013] Based on the memory prototype corresponding to the current scenario, memory retrieval is performed to obtain memory retrieval information;

[0014] Based on the memory retrieval information, an expectation check is performed to obtain the memory retrieval information that passes the check;

[0015] Based on the memory retrieval information that has passed the check, simulated actions are performed to obtain the final decision action;

[0016] The method for constructing the preset memory model is as follows:

[0017] Construct a memory structure based on the probability density function;

[0018] Obtain historical event information;

[0019] Based on the memory structure and historical event information, a memory model is obtained.

[0020] Furthermore, the memory structure is represented as a series of joint probability density functions: P1(X), P2(X), ... P k (X), where each probability density function represents a prototype of a class of events; where P1(X) represents the first probability density function; P2(X) represents the second probability density function; P k (X) represents the k-th probability density function; X represents the information stored in memory;

[0021] The i-th probability density function P i The expression for (X) is as follows:

[0022]

[0023] Among them, P i (X) represents the i-th joint probability density function, i.e., the i-th prototype; n represents the prototypes that a person has experienced, P. i The number of historical events in (X); W represents the diagonal weight matrix; ρ represents the length of vector X, i.e., the amount of information contained in X; X represents the information stored in memory; X j This represents the j-th time a person has conformed to prototype P. i The information contained in the event (X); (·) T It is represented as the transpose matrix; h is represented as the smoothing parameter.

[0024] Furthermore, the information stored in the memory includes the characteristics of the event, expectations, relevant clues, possible goals, and actions.

[0025] Furthermore, based on the aforementioned memory structure and historical event information, a memory model is obtained, including:

[0026] For all prototypes in the memory structure, calculate the probability density corresponding to the historical event information to obtain the maximum probability density value;

[0027] When the maximum probability density value is greater than or equal to a preset probability density value threshold, the historical event belongs to the prototype corresponding to the maximum probability density value, the prototype corresponding to the maximum probability density value is updated, and the updated prototype is obtained.

[0028] When the maximum probability density value is less than a preset probability density value threshold, the historical event does not belong to any prototype in the memory structure, and a new prototype is constructed.

[0029] Based on the updated prototype and the new prototype, a memory model is obtained.

[0030] Furthermore, the updated prototype's expression is as follows:

[0031]

[0032] Among them, P i (X)′ represents the prototype corresponding to the updated maximum probability density value, X s It represents information about historical events.

[0033] Furthermore, the expression for constructing the new prototype is as follows:

[0034]

[0035] Among them, P k+1 (X) represents the (k+1)th joint probability density function, i.e., the (k+1)th prototype; X s It represents information about historical events.

[0036] Furthermore, based on the current scenario, feature recognition is performed using a preset memory model to obtain the prototype corresponding to the current scenario, including:

[0037] For all prototypes in the preset memory model, calculate the marginal distribution probability density of the current scenario to obtain the maximum marginal distribution probability density value; wherein, the prototype corresponding to the current scenario is the prototype corresponding to the maximum marginal distribution probability density value.

[0038] The expression for the probability density of the marginal distribution is as follows:

[0039]

[0040] Among them, P i (F) represents the marginal distribution probability density function corresponding to the i-th prototype, which is P i Marginal distribution of (X); W F It is represented as the first r-order submatrix of the diagonal weight matrix W; r is the length of the eigenvector F, representing the number of features; F is denoted as the eigenvector; F j This is represented by the values ​​of the feature vectors in the experience.

[0041] Furthermore, based on the memory prototype corresponding to the current scenario, memory retrieval is performed to obtain memory retrieval information; the expression for memory retrieval is as follows:

[0042]

[0043] Among them, P i (Z) represents the probability density function for memory retrieval; α j Represented as coefficients derived from features; W Z This is represented as a new matrix in the diagonal weight matrix W, consisting of elements corresponding to Z; Z represents the memory retrieval information, including expectations, relevant cues, possible goals, and actions; Z j Represented as the value of Z in historical events; F s F is represented as the feature vector of the current scenario. j This represents the values ​​of the feature vectors in historical events.

[0044] Furthermore, the expected check includes the following steps:

[0045] Based on the memory retrieval information, the prototype retrieval expectation corresponding to the current scenario is obtained;

[0046] Check whether the expected results of the prototype extraction match the expected results of the current scenario;

[0047] If the prototype extraction expectation is discrete and matches the expectation of the current scenario, then the memory retrieval information that has passed the check is obtained, and the process proceeds to the next step.

[0048] If the prototype extraction expectation is discrete and does not match the expectation of the current scenario, then the information of the current scenario is reacquired.

[0049] If the prototype extraction expectation is continuous and the probability density value of the expectation of the current scenario is greater than or equal to a preset threshold, then the memory extraction information that has passed the check is obtained, and the process proceeds to the next step.

[0050] If the prototype extraction expectation is continuous and the probability density value of the expectation of the current scenario is less than a preset threshold, then the information of the current scenario is reacquired.

[0051] Furthermore, based on the memory retrieval information obtained from the passed check, simulated actions are performed to obtain the final decision action, including:

[0052] Based on the memory retrieval information that has passed the check, the prototype retrieval action corresponding to the current scenario is obtained;

[0053] The execution process of the prototype extraction action is simulated to obtain the simulation execution result;

[0054] If the simulation result is feasible, then the prototype extraction action becomes the final decision-making action.

[0055] If the simulation result is infeasible, other possible actions are extracted from the prototype corresponding to the current scenario, and the above steps are repeated.

[0056] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0057] The present invention provides an improved cognitive-driven decision-making method for predicting human decisions. On the one hand, it employs a pre-set memory model that can effectively store continuous memory information and match corresponding prototypes based on continuous information in the current situation. On the other hand, the present invention uses a joint probability density function, which has the ability to simulate the uncertainty of human decision-making.

[0058] This invention effectively solves the problem of acquiring and storing frequency / probability information, and can describe the role of this information in the decision-making process. The problem of storing frequency / probability information is solved by storing a joint probability density function in a memory structure. Furthermore, the probability density function in the memory model is obtained by fitting historical time information experienced by the individual, thus solving the problem of acquiring frequency / probability information. Attached Figure Description

[0059] Figure 1 This is a flowchart of an improved cognitive-driven decision-making method for predicting human decisions;

[0060] Figure 2 This is a module diagram of an improved cognitive-driven decision-making method for predicting human decisions;

[0061] Figure 3 This is a diagram illustrating human obstacle avoidance.

[0062] Figure 4 This is a schematic diagram of the current scenario;

[0063] Figure 5 It is the actual obstacle avoidance process of a person. Detailed Implementation

[0064] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0065] Example 1

[0066] This embodiment 1 discloses an improved cognitive-driven decision-making method for predicting human decisions, including:

[0067] Obtain the current context;

[0068] Based on the current scenario, feature recognition is performed using a pre-defined memory model to obtain the prototype corresponding to the current scenario;

[0069] Based on the prototype corresponding to the current situation, memory retrieval is performed to obtain memory retrieval information;

[0070] Based on the memory retrieval information, an expectation check is performed to obtain the memory retrieval information that passes the check;

[0071] Based on the information retrieved from the memory that has passed the inspection, simulated actions are performed to obtain the final decision action;

[0072] The pre-defined memory model is constructed as follows:

[0073] Construct a memory structure based on the probability density function;

[0074] Obtain historical event information;

[0075] Based on memory structure and historical event information, a memory model is obtained. The technical concept of this invention is as follows: on the one hand, the pre-set memory model can effectively store continuous memory information and can match the corresponding prototype according to the continuous information in the current situation; on the other hand, this invention uses a joint probability density function, which has the ability to simulate the uncertainty of human decision-making.

[0076] The specific steps are as follows: Figure 1 and 2 As shown:

[0077] Step 1: Determine the memory structure.

[0078] Determine the memory structure, that is, determine what kind of information is stored and the form in which this information is stored.

[0079] In this embodiment, the memory structure is represented as a series of joint probability density functions: P1(X), P2(X), ... P k Each probability density function (X) represents a prototype of a class of events, representing the common characteristics satisfied by a large number of similar events; where P1(X) represents the first probability density function; P2(X) represents the second probability density function; P k (X) represents the k-th probability density function. X represents the information stored in memory, including four categories of information: expectation, relevant cues, possible goals, and actions; where expectation is represented as E = {e1, e2, ..., e}. p The relevant clues are represented as C = {c1, c2, ..., c}. q The possible target can be represented as G = {g1, g2, ... g}. s The action is represented as A = {a1, a2, ... a}. o}

[0080] The i-th probability density function P i The expression for (X) is as follows:

[0081]

[0082] Among them, P i (X) represents the i-th joint probability density function, i.e., the i-th prototype; n represents the prototypes that a person has experienced, P. i (X) represents the number of historical events; W represents the diagonal weight matrix; ρ represents the length of vector X, i.e., the amount of information contained in X; X represents the information stored in memory, including the characteristics of the event, expectations, relevant clues, possible goals and actions; X j This represents the j-th time a person has conformed to prototype P. i The information contained in the event (X); (·) T It is represented as the transpose matrix; h is represented as the smoothing parameter, and the value of h needs to satisfy the following condition.

[0083]

[0084] Step 2: Construct a memory model.

[0085] Building a memory model: storing information from events experienced by a person into a memory structure in a specific way.

[0086] Memory is built upon past experiences. Suppose a person experiences a historical event Ex, which contains historical event information X. s To construct a memory model, we first need to determine the prototype to which the historical event Ex belongs. This is done as follows:

[0087] First, store the historical event information X s Substituting into equation (1), we calculate the probability density of all prototypes, i.e., P1(X). s ),P2(X s ),…P n (X s Then, find the maximum probability density value, assumed to be P. i (X s Finally, it is determined whether the result matches the preset probability density threshold. The size relationship.

[0088] When the maximum probability density value is greater than or equal to a preset probability density value threshold, i.e. Then the historical event Ex belongs to the maximum probability density value P i (X s The corresponding prototype P) i (X), historical event Ex is applied to update prototype P in the memory structure. i (X), update the maximum probability density value P. i (X s The corresponding prototype P)i (X) yields the updated prototype.

[0089] The update method is shown in equation (3):

[0090]

[0091] Among them, P i (X)′ represents the prototype corresponding to the updated maximum probability density value, X s It represents information about historical events.

[0092] When the maximum probability density value is less than a preset probability density value threshold, i.e. This indicates that the historical event Ex should belong to a new prototype, which has not yet been established in memory. If the historical event information does not belong to any prototype in the memory structure, a new prototype needs to be constructed, as shown in equation (4):

[0093]

[0094] Among them, P k+1 (X) represents the (k+1)th joint probability density function, i.e., the (k+1)th prototype; X s It represents information about historical events.

[0095] Based on the updated prototype and the new prototype, a complete memory model is obtained.

[0096] Step 3: Feature-based scenario recognition.

[0097] Based on the characteristics of the current situation, identify similar events experienced from the memory model.

[0098] When faced with a new scenario, the task of scenario recognition is to determine which prototype in the memory model it belongs to. The determination method is as follows: based on the preset memory model, calculate the marginal probability density of the current scenario's feature information, and obtain the marginal probability density value with the largest marginal probability density value; where the prototype corresponding to the current scenario is the prototype corresponding to the maximum marginal probability density value.

[0099] First, based on the feature vector F of the current scenario, calculate the joint probability density as shown in equation (5).

[0100]

[0101] Among them, P i (F) is P i The marginal distribution of (X), that is, equation (5) is the marginal distribution of equation (1); W FIt is represented as the first r-order submatrix of the diagonal weight matrix W; r is the length of the eigenvector F, representing the number of features; F is denoted as the eigenvector; F j F and W represent the values ​​of the feature vector in historical events. F The correspondence between X and W in equation (1) can be expressed as equation (6).

[0102]

[0103] For all prototypes in the memory model, the probability density in formula (5) is calculated, and the prototype corresponding to the largest marginal probability density value is the prototype corresponding to the current scenario.

[0104] Step 4: Memory retrieval.

[0105] Memory retrieval: Based on the memory prototypes corresponding to the current situation, information about such events is retrieved from memory to obtain memory retrieval information, including possible goals, relevant clues, expectations, and actions, which are used to solve the problems faced in the current situation.

[0106] If the current scenario belongs to the i-th prototype P i (X), then it needs to be from P i Extract four types of information from (X): expectation, relevant clues, possible targets, and actions. The extraction method is to directly sample the probability distribution P shown in equation (7). i (Z).

[0107]

[0108] Among them, P i (Z) represents the probability distribution function for retrieving memories of the prototype corresponding to the current situation; α j Represented as coefficients derived from features; W Z Z is represented as a new matrix in the diagonal weight matrix W, consisting of elements corresponding to Z; Z represents the extracted memory information, which includes four categories of information: expectation, relevant cues, possible targets, and actions, i.e., Z = {E, C, G, A}; Z j Represented as the value of Z in historical events; F s F is represented as the feature vector of the current scenario. j These are represented as the values ​​of the feature vectors in historical events. Z and W Z The definition is similar to that in equation (6).

[0109] Step 5: Expected Check.

[0110] Based on the memory retrieval information, the expected prototype retrieval for the current scenario is obtained. It is then checked whether this expected prototype retrieval contradicts the expectation in the current scenario. If they contradict each other, the current scenario is considered incorrectly identified, and the current scenario needs to be retrieved again. If they do not contradict each other, the memory retrieval information passes the check, and the next step is performed.

[0111] To check whether the identified memory prototype matches the current situation, it is necessary to check whether the prototype retrieval expectation E corresponding to the current situation matches the expectation E of the current situation. The method of checking is:

[0112] If the prototype retrieval expectation E is discrete and matches the expectation of the current situation (E=E), then the memory retrieval information that has passed the check is obtained, and the process proceeds to the next step.

[0113] If the prototype extraction expectation E is discrete and does not match the expectation of the current situation (E≠E), then return to re-perceive the current situation, obtain feature information, and re-identify the prototype corresponding to the current situation.

[0114] If the prototype extraction expectation E is continuous, then calculate the probability density as shown in equation (8).

[0115]

[0116] In the formula, W FP It is a new weight matrix composed of elements from the weight matrix W corresponding to the features F and the expectations E. p It is a new weight matrix composed of the elements corresponding to the expected value E in the weight matrix W, where p is the length of E. j It is the value of E in the historical event.

[0117] In response to the prototype extraction expectation E being continuous, and the probability density value of the expectation of the current scenario being greater than or equal to a preset threshold. The memory retrieval information that has passed the check is then obtained, and the process proceeds to the next step;

[0118] In response to the prototype extraction expectation E being continuous, and the probability density value of the expectation in the current scenario being less than a preset threshold. Then, the current situation is re-perceived, feature information is obtained, and the prototype corresponding to the current situation is re-identified.

[0119] Step Six: Simulate the execution of the action.

[0120] Based on the memory retrieval information that has passed the inspection, the prototype retrieval action corresponding to the current situation is obtained;

[0121] The execution process of the prototype extraction action is simulated to obtain the simulation execution result;

[0122] If the simulation result is feasible, then the prototype-extracted action becomes the final decision-making action.

[0123] If the simulation result is infeasible, other possible actions are extracted from the prototype corresponding to the current scenario, and the above steps are repeated.

[0124] The specific steps are as follows:

[0125] Based on the characteristics F of the current situation obtained through perception s And action A extracted from the memory prototype, simulate the execution of action A extracted from the prototype, and obtain the simulation execution result. If the result shows that the simulation execution result is feasible, then action A extracted from the prototype is taken as the final decision execution action; otherwise, other possible actions are extracted from the prototype corresponding to the current situation, i.e., the next possible action.

[0126] The method for extracting the next possible action is to determine the feature F. s After obtaining the expected value E, relevant clues C, and possible target G, the sampling probability distribution is shown in equation (9).

[0127]

[0128] In the formula, A j It is action information in historical events, W A Y is a new weight matrix composed of the elements corresponding to action A in the weight matrix W. s ={F s {E,C,G} represent known information, Y j Y represents the characteristics, expectations, relevant clues, and potential target information of historical events. s Values ​​taken in historical events.

[0129] In summary, firstly, the method proposed in this patent can effectively store continuous information and match similar scenarios based on the continuous information in the current scenario when faced with a new scenario. As shown in Equation (1), in the memory structure proposed in this patent, continuous information (included in X) is stored in memory in the form of a probability density function. According to the multi-parameter probability density fitting theory, the probability density function shown in Equation (1) represents the common characteristics of a series of data, that is, it represents a series of prototypes of similar events, which is consistent with the principle of human memory. Therefore, continuous information can be effectively stored in memory through Equation (1).

[0130] In the method proposed in this patent, upon encountering a similar scenario, the model will match similar scenarios (prototypes) in memory based on the features F (containing continuous information) of the current scenario. The matching is achieved by calculating the probability density function shown in equation (5). The probability density in equation (5) is calculated for all prototypes in memory, and the prototype corresponding to the highest probability density is the prototype to which the current scenario belongs. In the method proposed in this patent, the matching is based on the magnitude of the probability density, rather than a direct comparison of continuous information. As an effective measure of the closeness of continuous information, probability density can effectively match the current scenario with similar scenarios in the memory model.

[0131] Secondly, the model proposed in this patent has the ability to simulate the uncertainty of human decision-making. The uncertainty of human decision-making manifests itself in the fact that, when faced with similar current situations multiple times, although most decision results may be similar, there may still be very different decision results. In the model proposed in this patent, after successfully identifying similar situations in memory, four types of information are obtained by sampling according to Equation (7): extraction expectation, relevant clues, possible goals, and actions, where actions are the result of decision-making. Although most of the actions obtained by sampling are concentrated in areas with high probability density, there is still a very small probability that very different results will occur.

[0132] Finally, the method proposed in this patent can effectively solve the problem of acquiring and storing frequency / probability information, and can describe the role of this information in the decision-making process. First, this patent represents memory as a series of joint probability density functions: P1(X), P2(X), ... P k (X), where each probability density function represents a prototype of a class of events, representing the common characteristics satisfied by a large number of similar events. As shown in Equation (1), since each prototype stored in memory is represented by a joint probability density function, for a specific event, the information X it contains actually satisfies a certain probability distribution, rather than a fixed value. Therefore, the model solves the problem of storing frequency / probability information by storing the joint probability density function in memory. Furthermore, as shown in Equation (1), the probability density function is obtained by fitting the information X from n events experienced by a person. Therefore, the model obtains frequency / probability information by statistically analyzing different information from multiple experiences, thus solving the problem of acquiring frequency / probability information.

[0133] Based on the probability density function stored in memory, scene recognition is then achieved according to the marginal probability density function shown in Equation (5). Since the marginal probability density function in Equation (5) is the marginal distribution of the joint probability density function in Equation (1), the frequency / probability information will be transferred from the joint probability density function to the marginal probability density function during the scene recognition process, thus affecting the result of scene recognition.

[0134] After scene recognition is completed, the present invention will use the prototype P identified in memory. i (X), by sampling the probability density function (Equation (7)), four types of information are extracted: expectation, relevant clues, possible target, and action. Equation (7) is the probability density function of the remaining information Z obtained by dimensionality reduction after determining the feature F using Equation (1). Therefore, Equation (7) contains the frequency / probability information of information Z in Equation (1), which will then affect the sampling result, i.e., the decision result. Thus, the most important situational category and decision generation process in cognitive-dominated decision-making both involve the role of frequency / probability information. Therefore, this invention can effectively simulate the role of frequency / probability information in the decision-making process.

[0135] Example 2

[0136] This embodiment 2 provides an improved cognitive-driven decision-making method for human obstacle avoidance, where the stored information includes both continuous and discrete parameters. This embodiment will illustrate that the invention can effectively process continuous information. Figure 3 As shown, a person is moving forward at a constant speed of V = 2 m / s. There is an obstacle with a diameter of D at a distance R ahead. The person changes the direction of the speed, so that the new speed V′ rotates by an angle θ from the original speed to avoid the obstacle.

[0137] Step (1) Determine the memory structure.

[0138] The memory structure consists of k joint probability density functions P1(X), P2(X), ... P k (X) represents treating each obstacle avoidance experience of the robot as an event, with each probability density function representing a prototype of a class of events. Any probability density function P... i (X) can be represented as:

[0139]

[0140] Where X represents the information stored in memory, X j This represents the value of X in each event. In this example, X is specifically shown in Table 1. From Table 1, we know that ρ = 7. In this example, W and h can take the following values:

[0141]

[0142] If the information stored in X is continuous (e.g., the distance between the robot and the obstacle, the size of the obstacle, the time when the person and the obstacle meet, the speed, direction, and rotation angle), it can be directly constructed using equation (1). If the information stored in X is discrete (e.g., from the person's perspective, whether the obstacle is above or below the person; whether the person may pass directly above, below, or collide with the obstacle; whether the robot successfully avoids the obstacle), since equation (1) is defined based on the joint probability density function of continuous variables, the discrete information needs to be mapped to the continuous domain before the memory structure can be constructed using equation (1). Taking the discrete parameter "from the robot's perspective, whether the obstacle is above or below the robot" as an example, in the discrete domain, P∈{0,1}, and after mapping to the continuous domain, P∈{-∞,∞}, but in the event, the value of P can only be 0 or 1.

[0143] Table 1 Information stored in memory

[0144]

[0145] Step (II) Memory Construction.

[0146] The memory model is built based on past events. Here, we assume that 50 events have been experienced (as shown in Table 2), and the memory model will be constructed based on these 50 events. The more events experienced, the richer the human's obstacle avoidance experience, and the higher the obstacle avoidance performance (success rate and efficiency). Since the purpose of constructing the cognitive-driven decision-making model is to enable the robot to have autonomous obstacle avoidance capabilities, we only select events of successful obstacle avoidance (G=1) for memory construction.

[0147] Information from the 50 events experienced in Table 2

[0148]

[0149]

[0150]

[0151] The memory model is constructed by storing each event into the memory structure according to the event number in Table 2. The storage process of the first three memories is taken as an example below. First, for the first event, since there is no prototype in the memory at present, the new prototype shown in equation (3) will be constructed directly.

[0152]

[0153] Where W is as shown in equation (2), X1=[21 1 15.6 0 16.8 1 52.4] T Next, regarding the second event, X2 = [29.6 1 13.2 2 13.8 1 52.4]T At this point, we only have one prototype P1(X) in our memory. Substituting X2 into P1(X), we calculate the value of P1(X2), which gives us P1(X2) = 4.61 × 10 -17 Less than the threshold Therefore, a new prototype P2(X) should be established as shown in equation (4).

[0154]

[0155] For the third event, X3 = [29.8 1 16 2 14.5 1 19.1] T At this point, there are two prototypes in memory, P1(X) and P2(X), and the calculated value is P1(X3) = 2.68 × 10⁻⁶. -19 P2(X3) = 1.76 × 10 -7 Clearly, P2(X3) > P1(X3). We continue by comparing P2(X3) with the threshold. Discover Then X3 should be stored in the prototype P2(X), and the method of storage is to update P2(X) using equation (5).

[0156]

[0157] Using this method, each event is processed, and the memory model is completed after processing 50 events.

[0158] Step (3) Feature-based scenario recognition. Suppose we are facing a new scenario, such as... Figure 4 As shown, the characteristic of this scenario is F = [R PD] = [29.5014.3]. To identify which prototype this scenario belongs to, we should consider all prototypes P. i (X) Calculate its marginal distribution P i (F), as shown in equation (6).

[0159]

[0160] In this example, After calculation, P 15 (X) is the largest. Therefore, the current scenario belongs to the 15th prototype in the memory model.

[0161] Step (iv) Memory Retrieval. Based on the results of scene recognition, four types of information need to be extracted from the 15th prototype: expectation, relevant clues, possible targets, and actions. The extraction method is to sample the joint probability density function as shown in equation (7).

[0162]

[0163] In this embodiment 2, the 15th prototype contains information on only 3 events, therefore, n = 3. These three events correspond to the 26th, 36th, and 50th events in Table 2, respectively. Using rejection sampling, the expected value E = 1, the relevant clues T = 10.5, the possible target G = 1, and the action θ = -20.6 can be extracted.

[0164] Step (5) Expected Check. For example... Figure 4 As shown, the expected value of the current scenario can be obtained. Since the expected value E is discrete, we can directly check E against the given value. Are they equal? ​​In this example, Therefore, the expectation is met, and the scenario recognition is correct.

[0165] Step (VI) Simulate the execution of the action. The person simulates the action θ = -20.6 in the current situation and finds that the person can successfully avoid the obstacle. Therefore, the action is executed directly, and the result is as follows: Figure 5 As shown. By Figure 5 The results show that the present invention can successfully achieve obstacle avoidance for humans. This embodiment illustrates that even when humans have very little experience in obstacle avoidance (only 3 events in the 15th prototype identified), obstacle avoidance can still be effectively achieved.

[0166] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An improved cognitive dominant decision making method for human decision prediction, characterized in that, The method comprises the following steps: acquiring a current situation; performing feature recognition based on a preset memory model according to the current situation, to obtain a prototype corresponding to the current situation; performing memory extraction according to the prototype corresponding to the current situation, to obtain memory extraction information; performing expectation checking according to the memory extraction information, to obtain memory extraction information that passes the checking; performing simulation action according to the memory extraction information that passes the checking, to obtain a final decision action, so as to realize autonomous obstacle avoidance of the robot; wherein the preset memory model is constructed by the following method: constructing a memory structure based on a probability density function; acquiring historical event information; obtaining a memory model according to the memory structure and the historical event information; The memory structure is represented as a series of joint probability density functions: Each probability density function represents a prototype of a class of events; wherein, is represented as the first probability density function; is represented as the second probability density function; is represented as the kth probability density function; represents information stored in the memory; The expression of the first probability density function is as follows: The expression of the second probability density function ； in, Represented as the first The joint probability density function, i.e. the th joint probability density function One prototype; The archetype representing a person's experience is The number of historical events; Represented as a diagonal weight matrix; Represented as a vector The length, i.e. The amount of information contained; This represents information stored in memory; It represents the first experience a person has had. Sub-conformity prototype The information contained in the event; Represented as a transpose matrix; Represented as a smoothing parameter; the information saved in the memory includes features, expectations, related clues, possible targets and actions of events; the features of the events include the distance between a person and an obstacle, whether the obstacle is above or below the person from the perspective of the person, and the size of the obstacle; the expectations include that the person will probably pass directly above the obstacle, below the obstacle, or collide with the obstacle the related clues include the time when the person meets the obstacle; the possible targets include that the person successfully avoids the obstacle; the actions include speed, direction and rotation angle.

2. The improved cognitive dominant decision making method for human decision prediction as claimed in claim 1 wherein, obtaining a memory model according to the memory structure and the historical event information, comprising: calculating the probability density corresponding to the historical event information for all prototypes in the memory structure, to obtain a maximum probability density value; in response to the maximum probability density value being greater than or equal to a preset probability density value threshold, the historical event belongs to the prototype corresponding to the maximum probability density value, the prototype corresponding to the maximum probability density value is updated, to obtain an updated prototype; in response to the maximum probability density value being less than the preset probability density value threshold, the historical event information does not belong to any prototype in the memory structure, a new prototype is constructed; obtaining a memory model according to the updated prototype and the new prototype.

3. The improved cognitive dominant decision making method for human decision prediction as claimed in claim 2, wherein, the expression of the updated prototype is as follows: ； wherein, represents a prototype corresponding to the updated maximum probability density value, represents historical event information.

4. The improved cognitive dominant decision making method for human decision prediction as claimed in claim 3 wherein, the expression of the new prototype is as follows: ； wherein, denotes the joint probability density function, i.e. the prototype; denotes the historical event information.

5. The improved cognitive dominant decision making method for human decision prediction as claimed in claim 4 wherein, performing feature recognition based on the preset memory model according to the current situation, to obtain a prototype corresponding to the current situation, comprising: calculating the edge distribution probability density of the current situation for all prototypes in the preset memory model, to obtain a maximum edge distribution probability density value; wherein the prototype corresponding to the current situation is the prototype corresponding to the maximum edge distribution probability density value; the expression of the edge distribution probability density is as follows: ； wherein, denotes the edge distribution probability density function corresponding to the prototype of the order, is the edge distribution of denotes the first order sub-matrix of the diagonal weight matrix is the length of the eigenvector representing the number of features; denotes the eigenvector denotes the value of the eigenvector in the historical event.​ 6. The improved cognitive dominant decision making method for human decision prediction as claimed in claim 5 wherein, the expression of the memory extraction performed according to the prototype corresponding to the current situation is as follows: ； wherein, is represented as a probability density function of memory retrieval; is represented as a coefficient derived from the features; is represented as a diagonal weight matrix is represented as a new matrix composed of the elements of corresponding to is represented as memory retrieval information, including expectations, relevant cues, possible goals, and actions; is represented as a value of in a historical event; is represented as a feature vector of the current situation; is represented as a value of the feature vector in a historical event.

7. The improved cognitive dominant decision making method for human decision prediction as claimed in claim 1 wherein, the expectation checking comprises the following steps: obtaining a prototype extraction expectation corresponding to the current situation according to the memory extraction information; checking whether the prototype extraction expectation is consistent with the expectation of the current situation; in response to the prototype extraction expectation being discrete and the prototype extraction expectation being consistent with the expectation of the current situation, obtaining memory extraction information that passes the checking, and entering the next step; in response to the prototype extraction expectation being discrete and the prototype extraction expectation being inconsistent with the expectation of the current situation, reacquiring information of the current situation; In response to the prototype extraction expectation being continuous and the probability density value of the current situation expectation being greater than or equal to a preset threshold, memory extraction information passing the check is obtained, and the next step is entered; In response to the prototype extraction expectation being continuous and the probability density value of the current situation expectation being less than a preset threshold, information of the current situation is reacquired.

8. The improved cognitive dominant decision making method for human decision prediction as claimed in claim 1 wherein, According to the memory extraction information passing the check, a simulation action is performed to obtain a final decision action, including: According to the memory extraction information passing the check, a prototype extraction action corresponding to the current situation is obtained; A simulation execution process of the prototype extraction action is simulated to obtain a simulation execution result; In response to the simulation execution result being a feasible result, the prototype extraction action is the final decision execution action; In response to the simulation execution result being an infeasible result, other possible actions are reextracted from the prototype corresponding to the current situation, and the above steps are repeated.

Images