Obstacle intention prediction method, device, apparatus and medium

Abstract

The present disclosure relates to the technical field of unmanned driving, and particularly relates to a method and device for obstacle intention prediction, equipment and a medium. The method comprises: obtaining a plurality of causal arguments and a plurality of observation arguments of a target obstacle at a current time, wherein the causal argument is used to represent conditional information of whether the obstacle intention is possible to exist, and the observation argument is used to represent result information of the obstacle intention generation; determining a prediction probability of a preset intention of the target obstacle based on the causal argument and the observation argument, wherein the preset intention comprises a plurality of intentions; and determining the intention of the target obstacle based on the prediction probability of the preset intention. The technical scheme of the present disclosure supports the prediction of the obstacle intention in multiple scenarios while ensuring that the intention prediction is highly explainable and the intention prediction result is stable.

Description

Technical Field

[0001] This disclosure relates to the field of autonomous driving technology, and in particular to a method, apparatus, device and medium for predicting obstacle intent. Background Technology

[0002] In recent years, intelligent driving technology has become a hot research topic in the automotive industry. To enable vehicles to drive safely and smoothly, a crucial yet often overlooked issue is how to predict the behavior of obstacles around the vehicle over a period of time—that is, how to predict the intentions of obstacles. Predicting the intentions of obstacles is a prerequisite for planning and controlling the vehicle's next actions.

[0003] Current obstacle intent prediction algorithm frameworks can be broadly categorized into two types. One type directly generates predicted trajectories for a future period based on the obstacle's historical state information and other environmental information. The other type first predicts the obstacle's intent and then uses different trajectory generation models to generate corresponding predicted trajectories based on that intent. The former primarily utilizes various deep learning algorithms, such as deep neural networks, long short-term memory artificial neural networks, and graph neural networks. These algorithms require processing, labeling, and training large amounts of data, have weak interpretability, cannot guarantee the stability of prediction results, and lack semantic information about the predicted trajectory's intent, hindering subsequent vehicle planning and control. Currently, they remain in the academic research and pre-implementation research stage. In contrast, the latter type of algorithm framework can output the probability of the obstacle's intent, has strong interpretability, and provides stable prediction results. However, the latter type of algorithm framework often only targets predictions for a single scenario (such as whether a vehicle needs to change lanes). Summary of the Invention

[0004] In order to solve the above-mentioned technical problems, or at least partially solve the above-mentioned technical problems, this disclosure provides an obstacle intention prediction method, apparatus, device and medium.

[0005] In a first aspect, embodiments of this disclosure provide a method for predicting obstacle intent, including:

[0006] The system acquires multiple causal arguments and multiple observational arguments regarding the target obstacle at the current moment. The causal arguments are used to characterize the conditional information of whether the obstacle's intention is possible, and the observational arguments are used to characterize the result information of the obstacle's intention.

[0007] Based on the causal arguments and the observational arguments, the predicted probability of the target obstacle's pre-defined intention is determined, and the pre-defined intention includes multiple intentions;

[0008] The intention of the target obstacle is determined based on the predicted probability of the preset intention.

[0009] In some embodiments, obtaining multiple causal arguments and multiple observational arguments regarding the target obstacle at the current moment includes:

[0010] Acquire current sensing information, including perception information, map information, and positioning information;

[0011] Based on the sensor information, the causal arguments and the observational arguments are determined.

[0012] In some embodiments, determining the predicted probability of the target obstacle's intended purpose based on the causal arguments and the observational arguments includes:

[0013] Based on the causal arguments, determine the conditional probability of the intention under the causal arguments;

[0014] Based on the observed arguments, determine the conditional probability of the observed arguments under the preset intention;

[0015] Based on the conditional probability of the intent corresponding to the target obstacle and the conditional probability of the observation argument, the predicted probability of the preset intent of the target obstacle is determined.

[0016] In some embodiments, determining the conditional probability of intent under the causal argument based on the causal argument includes:

[0017] Based on the aforementioned causal arguments, potential first intentions and impossible second intentions are determined from the predetermined intentions;

[0018] For the first intention, the conditional probability of the intention is determined based on the pre-defined correspondence between the causal arguments and the conditional probability of the intention;

[0019] For the second intention, a preset probability is used as the conditional probability of the intention, and the preset probability is used to make the predicted probability approach 0.

[0020] In some embodiments, the method further includes:

[0021] If any of the preset intentions does not have the causal argument, then the conditional probability of the intention corresponding to any of the intentions is 1.

[0022] In some embodiments, the method further includes:

[0023] Obtain the historical predicted probability of the target obstacle's preset intention at the previous moment;

[0024] Determining the predicted probability of the target obstacle's intended purpose based on the causal arguments and the observational arguments includes:

[0025] Based on the causal arguments, the observed arguments, the historical predicted probabilities, and the conversion probabilities between the pre-set intentions, the predicted probability of the preset intention of the target obstacle is determined.

[0026] In some embodiments, determining the predicted probability of the target obstacle's intended purpose based on the causal arguments and the observational arguments includes:

[0027] Invoke the dynamic Bayesian network to determine the transition probability between each intent node in the dynamic Bayesian network, as well as the historical intent posterior probability of the intent node in the dynamic Bayesian network at the previous time step;

[0028] Based on the causal arguments and the observation arguments, determine the intention conditional probability and the observation argument conditional probability of the intention node in the dynamic Bayesian network at the current moment.

[0029] Based on the historical posterior probability of intent, the transition probability, and the conditional probability of intent, the prior probability of intent nodes in the dynamic Bayesian network at the current moment is determined.

[0030] Based on the observed conditional probability and the prior probability of the intention, the current posterior probability of the intention node in the dynamic Bayesian network at the current time is determined, and the current posterior probability of the intention is the predicted probability.

[0031] In a second aspect, embodiments of this disclosure provide an obstacle intent prediction device, comprising:

[0032] The argument acquisition module is used to acquire multiple causal arguments and multiple observation arguments of the target obstacle at the current moment. The causal arguments are used to characterize the conditional information of whether the obstacle's intention is possible, and the observation arguments are used to characterize the result information of the obstacle's intention.

[0033] An intent probability determination module is used to determine the predicted probability of the preset intent of the target obstacle based on the causal arguments and the observational arguments, wherein the preset intent includes multiple intents;

[0034] The intent determination module is used to determine the intent of the target obstacle based on the predicted probability of the preset intent.

[0035] Thirdly, embodiments of this disclosure provide an electronic device, including:

[0036] Memory and one or more processors;

[0037] The memory is communicatively connected to the one or more processors, and the memory stores instructions that can be executed by the one or more processors. When the instructions are executed by the one or more processors, the electronic device is used to implement the obstacle intent prediction method provided in any embodiment of this disclosure.

[0038] Fourthly, embodiments of this disclosure provide a computer-readable storage medium storing computer-executable instructions thereon, which, when executed by a computing device, can be used to implement the obstacle intent prediction method provided in any embodiment of this disclosure.

[0039] The technical solution provided in this disclosure has the following advantages compared with the prior art:

[0040] The technical solution provided in this disclosure, based on conditional information characterizing the existence of an obstacle's intent (i.e., causal arguments), can determine the presence and absence of obstacle intents in various scenarios. Simultaneously, by combining the result information characterizing the obstacle's intent (i.e., observational arguments), it can determine the prediction probability of each intent among the preset intents of the target obstacle, thereby determining the target obstacle's intent based on the prediction probabilities of each intent. Thus, this technical solution, based on causal and observational arguments, determines the prediction probability of preset intents, supporting obstacle intent prediction in multiple scenarios while ensuring strong interpretability and stable prediction results. Attached Figure Description

[0041] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0042] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 A functional block diagram of an autonomous driving system provided in this disclosure embodiment;

[0044] Figure 2 A flowchart illustrating an obstacle intent prediction method provided in this embodiment of the disclosure;

[0045] Figure 3 A structural diagram of a dynamic Bayesian network provided in an embodiment of this disclosure;

[0046] Figure 4 This is a functional block diagram of an obstacle intent prediction device provided in an embodiment of the present disclosure;

[0047] Figure 5 This is a schematic diagram of the structure of an electronic device suitable for implementing the embodiments of this disclosure. Detailed Implementation

[0048] To better understand the above-mentioned objectives, features, and advantages of this disclosure, the solutions disclosed herein will be further described below. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0049] Numerous specific details are set forth in the following description in order to provide a full understanding of this disclosure, but this disclosure may also be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only some, and not all, of the embodiments of this disclosure.

[0050] Figure 1 This is a functional block diagram of an autonomous driving system provided in an embodiment of the present disclosure. Figure 1 As shown, the autonomous driving system may include a perception module 1, a map module 2, a positioning module 3, an obstacle intention prediction module 4, a planning and decision-making module 5, and a control module 6.

[0051] In some embodiments, the perception module 1 is used to acquire perception information about the environment surrounding the autonomous vehicle. In some embodiments, the perception module 1 acquires sensing information from a sensor array, interaction information from road equipment, or a cloud server to generate perception information. The perception information includes, but is not limited to, at least one of the following: obstacle information, road signs / markings, pedestrian / vehicle information, and drivable areas.

[0052] In some embodiments, map module 2 is used to acquire map information, including lanes and intersections. In some embodiments, map module 2 is a high-precision map module.

[0053] In some embodiments, the positioning module 3 is used to acquire the vehicle's positioning information. The positioning information includes the vehicle's pose, which includes the vehicle's coordinates and the angles between the vehicle's heading and each coordinate axis. In some embodiments, the positioning module 3 can perform positioning based on multiple positioning methods or devices. Positioning sources include, but are not limited to, GPS positioning sources, visual positioning sources, and lidar positioning sources. In some embodiments, when multiple positioning sources exist, the positioning module 3 can select a positioning source with higher confidence or perform fusion positioning.

[0054] In some embodiments, the obstacle intent prediction module 4 is used to predict the intent of an obstacle based on the perception information generated by the perception module 1, the map information generated by the map module 2, and the positioning information generated by the positioning module 3. Specifically, in some embodiments, the obstacle intent prediction module 4 is used to acquire causal arguments and observational arguments of the target obstacle at the current moment. The causal arguments characterize the conditional information indicating whether the obstacle's intent exists, and the observational arguments characterize the result information indicating the generation of the obstacle's intent. Based on the causal arguments and observational arguments, the module 4 determines the prediction probability of the target obstacle's preset intent, which includes multiple intents. Based on the prediction probability of the preset intent, the module 4 determines the intent of the target obstacle.

[0055] In some embodiments, the planning and decision-making module 5 is used for path planning and decision-making. In some embodiments, the planning and decision-making module 5 generates planning and decision-making information based on the perception information generated by the perception module 1, the positioning information generated by the positioning module 3, and the prediction results of the obstacle intent by the obstacle intent prediction module 4. In some embodiments, the planning and decision-making module 5 can combine at least one of the following data: V2X data, high-precision maps, etc., to generate planning and decision-making information. The decision-making information may include, but is not limited to, at least one of the following: behavior (e.g., including but not limited to following, overtaking, stopping, detouring, etc.), vehicle heading, vehicle speed, vehicle's desired acceleration, desired steering wheel angle, etc.

[0056] In some embodiments, the control module 6 is used to generate control commands for the vehicle's underlying execution system based on planning and decision information, and to issue the control commands so that the vehicle's underlying execution system controls the vehicle to travel along a desired path. These control commands may include, but are not limited to, steering wheel commands, lateral control commands, and longitudinal control commands.

[0057] Figure 2 This is a flowchart illustrating an obstacle intent prediction method provided in an embodiment of this disclosure. This obstacle intent prediction method is applicable to predicting the intent of dynamic obstacles in various scenarios and can be executed by the obstacle intent prediction device provided in this embodiment. This obstacle intent prediction device can be implemented using software and / or hardware. Figure 2 As shown, the obstacle intent prediction method includes the following steps:

[0058] S101. Obtain multiple causal arguments and multiple observational arguments for the target obstacle at the current moment.

[0059] In this context, causal arguments characterize the conditional information indicating whether an obstacle's intention is likely to exist, while observational arguments characterize the resulting information about the obstacle's intention. Combining causal arguments with observational arguments reveals the potential behavior of a target obstacle over a future period in various scenarios, i.e., the obstacle's intention. In some embodiments, causal arguments determine the probability of an intention existing, and by introducing causal arguments, different intention prediction results can be controlled based on prior knowledge of different scenarios. For example, the causal argument might be whether a left-turn lane exists ahead, representing scenarios with and without a left-turn lane. Clearly, the probability of a left-turn intention differs significantly between the two scenarios represented by the causal arguments. In scenarios with a left-turn lane, a left-turn intention is likely, with a relatively high probability; in scenarios without a left-turn lane, a left-turn intention is impossible, with an extremely low probability (this disclosure eliminates the intention by setting an extremely low probability) or 0, as it needs to predict the probabilities of multiple intentions. In some embodiments, observational arguments characterize the resulting information about the obstacle's intention. Considering that the behavior of an obstacle changes when it makes a certain intention, this change is reflected in the observational evidence. For example, when an obstacle intends to turn left, its velocity relative to the lane direction and its distance from the left lane edge will both change due to the left-turn intention. Therefore, in this case, the obstacle's velocity relative to the lane direction and its distance from the left lane edge can be used as the resulting information of the left-turn intention. Based on the changes in the obstacle's velocity relative to the lane direction and its distance from the left lane edge, it can be deduced that the obstacle's intention is likely to be a left turn.

[0060] In some embodiments, obtaining multiple causal arguments and multiple observational arguments for a target obstacle at the current moment includes: obtaining sensor information at the current moment, the sensor information including perception information, map information and positioning information; and determining causal arguments and observational arguments based on the sensor information.

[0061] In some embodiments, the sensing information can be derived from... Figure 1 The sensing module 1, map module 2, and positioning module 3 shown are used to generate and acquire sensing information generated by the sensing module 1, map module 2, and positioning module 3. In some embodiments, the sensing information is related to causal arguments and observational arguments, including but not limited to obstacle information (including the position, outline, orientation, and speed of obstacles, etc.), lane information (including lane lines and lane directions, etc.), and road sign information (including left-turn landmarks, right-turn landmarks, and straight-ahead landmarks, etc.). Exemplarily, embodiments of this disclosure provide some causal arguments and some observational arguments. The causal arguments are shown in Table 1, and the observational arguments are shown in Table 2.

[0062] Table 1. Abbreviations and Meanings of Causal Arguments

[0063] Cause and Argument (abbreviation) meaning SLE Is there an adjacent lane? PIL Are all obstacle borders located within the lane? IR Is the obstacle facing the opposite direction to the lane? LTE Is there a left-turn lane ahead? RTE Is there a right-turn lane ahead?

[0064] Table 2 Abbreviations and Meanings of Observational Arguments

[0065] Observational arguments (abbreviated) meaning min_dist_to_ls Shortest distance between the obstacle and the lane edge line theta The angle between the obstacle and the direction of the lane v_lat The velocity of the obstacle relative to the lane direction normal dist_to_left_ls Distance of obstacle from the left lane edge line dist_to_right_ls Distance of obstacle from the right lane edge line

[0066] For Table 1, the existence of adjacent lanes can be determined based on the location of the obstacle and the lane lines. For example, if the location of the obstacle and the lane lines indicate that there are no lane lines to the left and right of the lane line where the obstacle is located, then there is no adjacent lane; if the location of the obstacle and the lane lines indicate that there is a lane line to the left of the lane line where the obstacle is located, then there is a left lane; if the location of the obstacle and the lane lines indicate that there is a lane line to the right of the lane line where the obstacle is located, then there is a right lane; if the location of the obstacle and the lane lines indicate that there are lane lines to the left and right of the lane line where the obstacle is located, then both left and right lanes are present. Additionally, the existence of obstacle boundaries within the lane can be determined based on the obstacle's outline and the lane lines. The orientation of the obstacle can be determined based on the direction of the lane. The existence of a left-turn lane and a right-turn lane ahead can be determined based on road sign information. The specific determinations for these various reasons and arguments can be obtained without question from the corresponding perceptual information, and will not be elaborated here.

[0067] For Table 2, the shortest distance from an obstacle to the lane edge can be determined based on the obstacle's outline and lane lines. The angle between the obstacle's orientation and the lane direction can be determined based on the obstacle's orientation and lane direction. The obstacle's velocity relative to the lane direction's normal velocity can be determined based on the obstacle's velocity. The distances from the obstacle to the left lane edge and the right lane edge can be determined based on the obstacle's outline and lane lines. Similarly, the specific determination of the above observational arguments can be obtained without question based on the corresponding perceptual information, and will not be elaborated here.

[0068] It should be noted that the causal arguments can be set flexibly, and causal arguments can be added flexibly according to the actual scenarios that may be encountered.

[0069] S102. Based on the causal arguments and the observational arguments, determine the predicted probability of the target obstacle's intended purpose.

[0070] The preset intents include a variety of intents. For example, Table 3 shows some of the intents.

[0071] Table 3. Intended abbreviations and their meanings

[0072] Intent Abbreviation meaning KL Keep your lane CL Change lanes TL Entering the lane LL Leave the lane TUL Turn left TUR Turn right CRO Crossing the lane

[0073] In some embodiments, an intention may be associated with one or more causal arguments and one or more observational arguments, and the impact of different causal arguments and observational arguments on such an intention is likely to be different. Therefore, the predicted probability of each intention can be determined based on the influence of the relevant causal arguments and relevant observational arguments on it. In some embodiments, the influence of causal arguments on an intention can be the conditional probability of the intention, and the influence of observational arguments on an intention can be the conditional probability of the observational arguments. Based on this, in some embodiments, determining the predicted probability of a preset intention of a target obstacle based on causal arguments and observational arguments includes: determining the conditional probability of the intention under the causal arguments based on the causal arguments; determining the conditional probability of the observational arguments under the preset intention based on the observational arguments; and determining the predicted probability of the preset intention of the target obstacle based on the conditional probability of the intention and the conditional probability of the observational arguments corresponding to the target obstacle. Specifically, for each intention, its relevant causal arguments and observational arguments are obtained, and then the conditional probability of the intention under the causal arguments and the conditional probability of the observational arguments under that intention are determined. The conditional probabilities of the intention under various causal arguments and the conditional probabilities of the observational arguments under various intentions can be preset and can all be learned (trained) by collecting data. Accordingly, based on the conditional probability of the intention corresponding to the target obstacle and the conditional probability of the observation argument, the predicted probability of the preset intention of the target obstacle can be determined using the following formula:

[0074]

[0075] Among them, Π k P(e k |m=m i ) for the intention m i The conditional probability of the observational argument, ∑ c P(m=m i |c) represents the conditional probability of intention (prior probability of intention) under the causal argument c, ∑ m P(m, e) represents the marginal probability (the marginal probability of intention under each observation argument can be learned in advance), and e is the probability relative to intention m. i The relevant observational arguments, where k is the number of observational arguments.

[0076] For example, taking a left-turn intention (TUL) as an example, the causal arguments related to the left-turn intention include whether the obstacle's orientation is opposite to the lane direction and whether there is a left-turn lane ahead. The observational arguments related to the left-turn intention include the angle between the obstacle's orientation and the lane direction, the obstacle's velocity relative to the lane direction's normal direction, and the obstacle's distance from the left lane edge. Based on the above prediction probability formula, the predicted probability of the target obstacle's left-turn intention (TUL) can be obtained:

[0077]

[0078] Where k = {1, 2, 3}, e1 is the angle between the obstacle's orientation and the lane direction, e2 is the normal velocity of the obstacle relative to the lane direction, and e3 is the distance of the obstacle from the left lane edge.

[0079] Based on the above embodiments, possible intentions and impossible intentions can be determined based on causal arguments. This disclosure embodiment determines the conditional probability of possible intentions and the conditional probability of impossible intentions in the following manner. In some embodiments, determining the conditional probability of an intention under a causal argument includes: determining a potential first intention and an impossible second intention from preset intentions based on the causal arguments; for the first intention, determining the conditional probability of the intention based on a pre-defined correspondence between causal arguments and intention conditional probabilities; for the second intention, using the preset probability as the conditional probability of the intention, whereby the preset probability is used to make the predicted probability approach 0. Specifically, the potential first intention is the possible intention. As mentioned above, the conditional probability of the intention under each causal argument can be obtained through learning from mobile phone data. Therefore, a correspondence between causal arguments and intention conditional probabilities can be established. Thus, after determining the first intention, the conditional probability of the intention corresponding to the first intention can be determined by looking up this correspondence. For the second intention, which is an impossible intention, it needs to be removed from the final prediction result. Therefore, the predicted probability of the second intention needs to be set to a minimum value. According to the formula for predicting the probability of intent, by assigning a minimum value to the conditional probability of the second intent, the calculated prediction probability can be made to approach 0.

[0080] Furthermore, considering that some intentions may lack causal arguments, such as the intention to cross a lane, in order not to affect the calculation of the predicted probability of such intentions, in some embodiments, the method further includes: if any intention in the preset intentions lacks a causal argument, then the intention conditional probability corresponding to that intention is 1. In this case, it is only necessary to calculate the predicted probability of any of the above intentions based on the observed argument conditional probability and the marginal probability. In some embodiments, if the argument associated with an intention only includes the observed argument, it indicates that the intention lacks a causal argument.

[0081] Since the calculation of the intention prediction probability at the current time t is affected not only by the causal argument and the observation argument at the current time t, but also by the prediction probability at the previous time t-1, in some embodiments, the method further includes: obtaining the historical prediction probability of the target obstacle's preset intention at the previous time; correspondingly, determining the prediction probability of the target obstacle's preset intention based on the causal argument and the observation argument, including: determining the prediction probability of the target obstacle's preset intention based on the causal argument, the observation argument, the historical prediction probability, and the pre-set transition probability between intentions. Specifically, the historical prediction probability of the target obstacle's preset intention at the previous time t-1 can be obtained based on the causal argument and the observation argument at the previous time t-1, combined with the above prediction probability formula. The conditional probability of the intention at the current time t under causal argument c is:

[0082]

[0083] Wherein, P(m) t-1 =m j |e t-1 P(m) represents the historical predicted probability of the target obstacle's intended behavior at the previous time t-1. t =m i |m t-1 =m j ) represents other intentions m from the previous time t-1. j Change to intention m i The transition probability.

[0084] Furthermore, by combining the formula for the predicted probability of the above intention, we can obtain the predicted probability of the preset intention considering the influence of the time factor:

[0085]

[0086] Thus, the above technical solution takes into account the influence of time factors on the prediction probability of the preset intention at the current time, thereby improving the accuracy of prediction probability calculation.

[0087] S103. Based on the predicted probability of the preset intent, determine the intent of the target obstacle.

[0088] Based on S101 and S102, the predicted probability of each intention in the preset intention can be obtained. The predicted probabilities of each intention are compared, and the intention with the highest predicted probability is determined as the target obstacle.

[0089] Based on the embodiments described above, a probabilistic graphical model can be used to directly output the predicted probability of a preset intention based on the input state variables (such as causal arguments and observational arguments). In some embodiments, the probabilistic graphical model is a dynamic Bayesian network. Figure 3 A structural diagram of a dynamic Bayesian network is shown. Figure 3Each circle in the diagram represents a node (the meanings of the abbreviations within the circles are shown in Tables 1, 2, and 3). Arrows indicate causal relationships between nodes, and dashed arrows represent the influence of the previous moment on the current moment. In some embodiments, determining the predicted probability of the target obstacle's preset intention based on causal and observational arguments includes: invoking a dynamic Bayesian network to determine the transition probabilities between each intention node in the dynamic Bayesian network, as well as the historical posterior probability of the intention nodes in the dynamic Bayesian network at the previous moment; determining the conditional probability of the intention node and the conditional probability of the observational arguments in the dynamic Bayesian network at the current moment based on causal and observational arguments; determining the prior probability of the intention node in the dynamic Bayesian network at the current moment based on the historical posterior probability, transition probability, and conditional probability of the intention; and determining the current posterior probability of the intention node in the dynamic Bayesian network at the current moment based on the conditional probability of the observational arguments and the prior probability of the intention, where the current posterior probability of the intention is the predicted probability.

[0090] Specifically, the posterior probability of historical intent is the predicted probability output by the dynamic Bayesian network at the previous time step. Based on the sensor information, the causal argument and the observed argument are determined, and these are input into the corresponding causal argument node and observed argument node, respectively. Combining the posterior probability of historical intent and the transition probability, the predicted probability of the intent node in the dynamic Bayesian network at the current time step can be directly output.

[0091] It should be noted that when the current time is the initial time, it is only necessary to calculate the predicted probability of the intention node in the dynamic Bayesian network at the current time based on the causal argument and observation argument at the current time.

[0092] The obstacle intent prediction method provided in this disclosure, based on conditional information characterizing the existence of obstacle intent (i.e., causal arguments), can determine the presence and absence of obstacle intents in various scenarios. Simultaneously, by combining the result information characterizing the generation of obstacle intents (i.e., observational arguments), it can determine the prediction probability of each preset intent of the target obstacle, thereby determining the target obstacle's intent based on the prediction probabilities of each intent. Thus, this technical solution, based on causal and observational arguments, determines the prediction probability of preset intents, supporting obstacle intent prediction in multiple scenarios while ensuring strong interpretability and stable prediction results.

[0093] This disclosure also provides an obstacle intent prediction device. Figure 4 A schematic diagram of an obstacle intent prediction device provided for embodiments of this disclosure is shown below. Figure 4As shown, the obstacle intent prediction device includes: an argument acquisition module 201, used to acquire multiple causal arguments and multiple observation arguments of the target obstacle at the current moment, wherein the causal arguments are used to characterize the conditional information of whether the obstacle intent may exist, and the observation arguments are used to characterize the result information of the obstacle intent; an intent probability determination module 202, used to determine the predicted probability of the target obstacle's preset intent based on the causal arguments and observation arguments, wherein the preset intent includes multiple intents; and an intent determination module 203, used to determine the target obstacle's intent based on the predicted probability of the preset intent.

[0094] In some embodiments, the argument acquisition module 201 is used for:

[0095] Acquire current sensor information, which includes perception information, map information, and location information;

[0096] Based on sensor information, determine causal arguments and observational arguments.

[0097] In some embodiments, the intent probability determination module 202 is used for:

[0098] Based on the causal arguments, determine the conditional probability of the intention under the causal arguments;

[0099] Based on observed arguments, determine the conditional probability of observed arguments under the pre-set intention;

[0100] Based on the conditional probability of the intention corresponding to the target obstacle and the conditional probability of the observation argument, the predicted probability of the preset intention of the target obstacle is determined.

[0101] In some embodiments, the intent probability determination module 202 is used for:

[0102] Based on causal arguments, identify potential first intentions and impossible second intentions from presupposed intentions;

[0103] For the first intention, the conditional probability of the intention is determined based on the pre-defined correspondence between the causal arguments and the conditional probability of the intention;

[0104] For the second intention, the preset probability is used as the intention conditional probability, and the preset probability is used to make the predicted probability approach 0.

[0105] In some embodiments, the apparatus further includes an intent conditional probability setting module, for:

[0106] If any intention in the preset intentions lacks a causal argument, then the conditional probability of the intention corresponding to any intention is 1.

[0107] In some embodiments, the apparatus further includes a historical prediction probability acquisition module, used for:

[0108] Obtain the historical predicted probability of the target obstacle's intended behavior at the previous moment;

[0109] Accordingly, the intent probability determination module 202 is used for:

[0110] Based on causal arguments, observational arguments, historical predicted probabilities, and the conversion probabilities between pre-set intentions, the predicted probability of the target obstacle's pre-set intention is determined.

[0111] In some embodiments, the intent probability determination module 202 is used for:

[0112] Invoke the dynamic Bayesian network to determine the transition probabilities between each intent node in the dynamic Bayesian network, as well as the posterior probability of the historical intent of the intent node in the dynamic Bayesian network at the previous time step.

[0113] Based on causal arguments and observational arguments, determine the intention conditional probability and observational argument conditional probability of the intention node in the dynamic Bayesian network at the current moment;

[0114] Based on the historical posterior probability of intent, the transition probability, and the conditional probability of intent, the prior probability of intent nodes in the dynamic Bayesian network at the current moment is determined.

[0115] Based on the conditional probability of the observation argument and the prior probability of the intention, the posterior probability of the current intention of the intention node in the dynamic Bayesian network at the current moment is determined, and the posterior probability of the current intention is the predicted probability.

[0116] The obstacle intent prediction device disclosed in the above embodiments can execute the obstacle intent prediction method disclosed in the above embodiments and has the same or corresponding beneficial effects. To avoid repetition, it will not be described again here.

[0117] This disclosure also provides an electronic device, including: a memory and one or more processors; wherein the memory is communicatively connected to the one or more processors, and the memory stores instructions that can be executed by the one or more processors. When the instructions are executed by the one or more processors, the electronic device is used to implement the obstacle intent prediction method described in any embodiment of this disclosure.

[0118] Figure 5 This is a schematic diagram of the structure of an electronic device suitable for implementing the embodiments of the present disclosure. For example... Figure 5As shown, the electronic device 300 includes a central processing unit (CPU) 301, which can execute various processes described in the foregoing embodiments according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage section 808 into a random access memory (RAM) 303. The RAM 303 also stores various programs and data required for the operation of the electronic device 300. The CPU 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0119] The following components are connected to I / O interface 305: an input section 306 including a keyboard, mouse, etc.; an output section 307 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 308 including a hard disk, etc.; and a communication section 309 including a network interface card such as a LAN card, modem, etc. The communication section 309 performs communication processing via a network such as the Internet. A drive 310 is also connected to I / O interface 305 as needed. A removable medium 311, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 310 as needed so that computer programs read from it can be installed into storage section 308 as needed.

[0120] In particular, according to embodiments of this disclosure, the methods described above can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program tangibly contained on a readable medium thereof, the computer program containing program code for performing the aforementioned obstacle avoidance method. In such embodiments, the computer program can be downloaded and installed from a network via communication section 309, and / or installed from removable medium 311.

[0121] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0122] The units or modules described in the embodiments of this disclosure can be implemented in software or hardware. The described units or modules can also be located in a processor, and the names of these units or modules do not necessarily constitute a limitation on the unit or module itself.

[0123] Furthermore, this disclosure also provides a computer-readable storage medium, which may be the computer-readable storage medium included in the apparatus described in the above embodiments; or it may be a standalone computer-readable storage medium not assembled into the device. The computer-readable storage medium stores computer-executable instructions, which, when executed by a computing device, can be used to implement the obstacle intent prediction method described in any embodiment of this disclosure.

[0124] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0125] The above description is merely a specific embodiment of this disclosure, enabling those skilled in the art to understand or implement it. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this disclosure. Therefore, this disclosure is not to be limited to the embodiments described herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An obstacle intention prediction method characterized by, include: The system acquires multiple causal arguments and multiple observational arguments regarding the target obstacle at the current moment. The causal arguments are used to characterize the conditional information indicating whether the obstacle's intention is possible, and the observational arguments are used to characterize the result information generated by the obstacle's intention. Based on the causal arguments and the observational arguments, the predicted probability of the target obstacle's pre-defined intention is determined, and the pre-defined intention includes multiple intentions; Based on the predicted probability of the preset intent, the intent of the target obstacle is determined; The step of determining the predicted probability of the target obstacle's intended purpose based on the causal arguments and the observational arguments includes: Based on the causal arguments, determine the conditional probability of the intention under the causal arguments; Based on the observed arguments, determine the conditional probability of the observed arguments under the preset intention; Based on the conditional probability of the intent corresponding to the target obstacle and the conditional probability of the observation argument, the predicted probability of the preset intent of the target obstacle is determined.

2. The method of claim 1, wherein, The multiple causal arguments and multiple observational arguments for obtaining the target obstacle at the current moment include: Acquire current sensing information, including perception information, map information, and positioning information; Based on the sensor information, the causal arguments and the observational arguments are determined.

3. The method of claim 1, wherein, Determining the conditional probability of intent under the causal argument based on the causal argument includes: Based on the aforementioned causal arguments, potential first intentions and impossible second intentions are determined from the predetermined intentions; For the first intention, the conditional probability of the intention is determined based on the pre-defined correspondence between the causal arguments and the conditional probability of the intention; For the second intention, a preset probability is used as the conditional probability of the intention, and the preset probability is used to make the predicted probability approach 0.

4. The method of claim 1, wherein, The method further includes: If any of the preset intentions does not have the causal argument, then the conditional probability of the intention corresponding to any of the intentions is 1.

5. The method of claim 1, wherein, The method further includes: Obtain the historical predicted probability of the target obstacle's preset intention at the previous moment; Determining the predicted probability of the target obstacle's intended purpose based on the causal arguments and the observational arguments includes: Based on the causal arguments, the observed arguments, the historical predicted probabilities, and the conversion probabilities between the pre-set intentions, the predicted probability of the preset intention of the target obstacle is determined.

6. The method of claim 1, wherein, Determining the predicted probability of the target obstacle's intended purpose based on the causal arguments and the observational arguments includes: Invoke the dynamic Bayesian network to determine the transition probability between each intent node in the dynamic Bayesian network, as well as the historical intent posterior probability of the intent node in the dynamic Bayesian network at the previous time step; Based on the causal arguments and the observation arguments, determine the intention conditional probability and observation argument conditional probability of the intention node in the dynamic Bayesian network at the current moment; Based on the historical posterior probability of intent, the transition probability, and the conditional probability of intent, the prior probability of intent nodes in the dynamic Bayesian network at the current moment is determined. Based on the observed conditional probability and the prior probability of the intention, the current posterior probability of the intention node in the dynamic Bayesian network at the current time is determined, and the current posterior probability of the intention is the predicted probability.

7. An obstacle intention prediction device characterized by comprising: include: The argument acquisition module is used to acquire multiple causal arguments and multiple observation arguments of the target obstacle at the current moment. The causal arguments are used to characterize the conditional information of whether the obstacle's intention exists, and the observation arguments are used to characterize the result information of the obstacle's intention. An intent probability determination module is used to determine the predicted probability of the preset intent of the target obstacle based on the causal arguments and the observational arguments, wherein the preset intent includes multiple intents; An intent determination module is used to determine the intent of the target obstacle based on the predicted probability of the preset intent; The step of determining the predicted probability of the target obstacle's intended purpose based on the causal arguments and the observational arguments includes: Based on the causal arguments, determine the conditional probability of the intention under the causal arguments; Based on the observed arguments, determine the conditional probability of the observed arguments under the preset intention; Based on the conditional probability of the intent corresponding to the target obstacle and the conditional probability of the observation argument, the predicted probability of the preset intent of the target obstacle is determined.

8. An electronic device, comprising: include: Memory and one or more processors; The memory is communicatively connected to the one or more processors, and the memory stores instructions that can be executed by the one or more processors. When the instructions are executed by the one or more processors, the electronic device is used to implement the obstacle intent prediction method as described in any one of claims 1-6.

9. A computer-readable storage medium having stored thereon computer- executable instructions, wherein, When the computer-executable instructions are executed by a computing device, they can be used to implement the obstacle intent prediction method as described in any one of claims 1-6.

Images