Obstacle prediction trajectory decision method and device, electronic equipment and storage medium

Abstract

The method disclosed by the embodiments of the present disclosure comprises the following steps: obtaining each candidate trajectory set composed of one predicted trajectory corresponding to each obstacle; determining the yield corresponding to each sampling longitudinal behavior scheme in each candidate trajectory set to obtain a target longitudinal behavior scheme; determining a target trajectory set through the yield corresponding to the target longitudinal behavior scheme of each candidate trajectory set; determining decision information through the target trajectory set and the corresponding target longitudinal behavior scheme; and realizing the decision of the longitudinal behavior of the obstacle and the current vehicle. The method solves the problem of low decision accuracy caused by only considering the trajectory of the obstacle interacting with the ego vehicle in the prior art, and solves the problem of difficulty in evaluating the similar probability of the predicted trajectory.

Description

Technical Field

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

[0002] In autonomous driving systems, the decision-making module is a crucial component. For self-driving vehicles, the most challenging aspect at the decision-making level is handling interactions with obstacles. If the autonomous vehicle incorrectly predicts the current vehicle's intentions, it will make inappropriate decisions.

[0003] Because the intentions of obstacles are constantly changing over time and cannot be directly observed, for interactive obstacles, the intentions of obstacles often depend on the behavior of the vehicle. Therefore, the behavior of obstacles is uncertain in most cases before the vehicle's trajectory is obtained.

[0004] The decision-making module can obtain the predicted trajectory of each obstacle from the prediction module. However, in many cases, the decision-making module will obtain multiple predicted trajectories with similar probabilities for a single obstacle. Current methods typically make behavioral decisions for the vehicle based on the predicted trajectories that interact with the vehicle, neglecting the predicted trajectories that do not interact with the vehicle. This leads to inaccurate predictions of the obstacle's intent, thus affecting the accuracy of decisions regarding both the vehicle and the obstacle. Summary of the Invention

[0005] To address or at least partially address the aforementioned technical problems, embodiments of this disclosure provide a decision-making method, apparatus, electronic device, and storage medium for predicting obstacle trajectories. This enables the decision-making of the lateral and longitudinal behaviors of obstacles, as well as the longitudinal behavior of the current vehicle. It solves the problem of low decision-making accuracy caused by prior art that only considers the trajectories of obstacles interacting with the vehicle, and also addresses the problem of difficulty in assessing the similarity of multiple predicted trajectories.

[0006] In a first aspect, embodiments of this disclosure provide a decision-making method for predicting obstacle trajectories, the method comprising:

[0007] Obtain multiple sets of alternative trajectories for obstacles, wherein each set of alternative trajectories consists of a predicted trajectory corresponding to each obstacle, and the sets of alternative trajectories are not completely identical;

[0008] For each set of candidate trajectories, the benefit corresponding to each sampled longitudinal behavior scheme in the set of candidate trajectories is determined, and the target longitudinal behavior scheme of the set of candidate trajectories is determined based on each benefit. The sampled longitudinal behavior scheme is composed of the sampled longitudinal acceleration sequence of the current vehicle and the sampled longitudinal acceleration sequence of the obstacle.

[0009] Based on the benefits corresponding to the target longitudinal behavior schemes of each set of candidate trajectories, a target trajectory set is determined from each set of candidate trajectories, and decision information is determined based on the target trajectory set and the target longitudinal behavior schemes of the target trajectory set.

[0010] Secondly, embodiments of this disclosure also provide a decision-making device for predicting obstacle trajectories, the device comprising:

[0011] The set acquisition module is used to acquire multiple candidate trajectory sets for obstacles, wherein each candidate trajectory set consists of a predicted trajectory corresponding to each obstacle, and the candidate trajectory sets are not completely identical.

[0012] The longitudinal behavior determination module is used to determine the benefit corresponding to each sampled longitudinal behavior scheme in each candidate trajectory set for each candidate trajectory set, and to determine the target longitudinal behavior scheme of the candidate trajectory set based on each benefit, wherein the sampled longitudinal behavior scheme is composed of the sampled longitudinal acceleration sequence of the current vehicle and the sampled longitudinal acceleration sequence of each obstacle;

[0013] The trajectory decision module is used to determine the target trajectory set from the candidate trajectory sets based on the benefits corresponding to the target longitudinal behavior schemes of each candidate trajectory set, and to determine decision information based on the target trajectory set and the target longitudinal behavior schemes of the target trajectory set.

[0014] Thirdly, embodiments of this disclosure also provide an electronic device, the electronic device comprising: one or more processors; a storage device for storing one or more programs; and when the one or more programs are executed by the one or more processors, causing the one or more processors to implement the obstacle prediction trajectory decision method as described above.

[0015] Fourthly, embodiments of this disclosure also provide a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the obstacle prediction trajectory decision method as described above.

[0016] This disclosure provides a method for determining the predicted trajectory of an obstacle. It acquires a set of candidate trajectories, each consisting of a predicted trajectory corresponding to one of all obstacles. For each candidate trajectory set, it determines the benefit corresponding to each sampled longitudinal behavior scheme to obtain a target longitudinal behavior scheme. Then, based on the benefits corresponding to the target longitudinal behavior schemes of each candidate trajectory set, it determines the target trajectory set, thus realizing the decision on the lateral behavior of the obstacle. Furthermore, by determining the decision information through the target trajectory set and the corresponding target longitudinal behavior scheme, it realizes the decision on the longitudinal behavior of the obstacle and the current vehicle. This method can perform intent analysis on all obstacles, solving the problem of low decision accuracy caused by prior art that only considers the trajectories of obstacles interacting with the vehicle. Moreover, this method can determine the optimal lateral behavior and corresponding optimal longitudinal behavior of the obstacle, as well as the optimal longitudinal behavior of the current vehicle, even when the probabilities of multiple predicted trajectories of the obstacle are similar, thus better solving the problem of difficulty in evaluating the similarity of predicted trajectory probabilities from a global perspective. Attached Figure Description

[0017] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0018] Figure 1 This is a flowchart of a decision-making method for predicting obstacle trajectories in an embodiment of this disclosure;

[0019] Figure 2 This is a schematic diagram illustrating a setting of a time window in an embodiment of this disclosure;

[0020] Figure 3 This is a schematic diagram illustrating how a first correction result is further corrected based on road information obtained within a set time window, according to an embodiment of this disclosure.

[0021] Figure 4 This is a schematic diagram illustrating how a third pose obtained through relocation is used to further optimize the second correction result in an embodiment of this disclosure.

[0022] Figure 5 This is a schematic diagram of a positioning architecture in an embodiment of this disclosure;

[0023] Figure 6 This is a schematic diagram of the structure of a decision-making device for predicting obstacle trajectories according to an embodiment of this disclosure;

[0024] Figure 7 This is a schematic diagram of the structure of an electronic device according to an embodiment of this disclosure. Detailed Implementation

[0025] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0026] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0027] Before providing a detailed description of the obstacle prediction trajectory decision-making method provided in the embodiments of this disclosure, the technical problem solved by this method will be explained first.

[0028] In existing technologies, trajectory decisions are typically made only for obstacles that the vehicle interacts with. Furthermore, during trajectory decision-making, the path with the highest probability from multiple predicted trajectories is usually selected as the obstacle's intent. However, when multiple predicted trajectories of an obstacle represent different behaviors with similar probabilities, this method cannot accurately predict the obstacle's intent.

[0029] Because the intentions of obstacles are constantly changing over time and are not directly observable, for interactive obstacles, the intentions of the obstacles often depend on the behavior of the vehicle. Therefore, the behavior of obstacles is usually uncertain before the vehicle's trajectory is obtained.

[0030] Therefore, in order to accurately and timely characterize the uncertainty of the future behavior of obstacles, solve the problem of uncertainty in obstacle behavior, and address the problem of low decision accuracy caused by only considering the trajectory of obstacles interacting with the vehicle, this disclosure provides a decision-making method for predicting obstacle trajectories, which analyzes the impact of the trajectories of different obstacles on the vehicle's behavior, thereby obtaining a predicted trajectory that meets the requirements for safe vehicle driving.

[0031] Figure 1 This is a flowchart illustrating a decision-making method for predicting obstacle trajectories according to an embodiment of this disclosure. This method can be executed by an obstacle prediction trajectory decision-making device, which can be implemented in software and / or hardware. This device can be configured in an electronic device, such as in the decision-making module of the current vehicle. Figure 1 As shown, the method may specifically include the following steps:

[0032] S110. Obtain multiple alternative trajectory sets for obstacles, wherein each alternative trajectory set consists of a predicted trajectory corresponding to each obstacle, and the alternative trajectory sets are not completely identical.

[0033] Here, the current vehicle can be understood as the vehicle itself; obstacles can be objects within the visual detection range of the current vehicle, such as pedestrians, vehicles, etc. In this embodiment, the predicted trajectories of obstacles that interact with the current vehicle and those that do not can be obtained.

[0034] Specifically, an interaction between an obstacle and the current vehicle can be determined when the obstacle or the current vehicle intends to change lanes to the current lane of the current vehicle or the obstacle. For example, an interaction between an obstacle and the current vehicle can be determined in the following situations: the current vehicle intends to change lanes to the current lane of the obstacle; the obstacle intends to change lanes to the current lane of the current vehicle; the current vehicle intends to change lanes to the current lane of the obstacle, and the obstacle intends to change lanes to the current lane of the current vehicle.

[0035] The predicted trajectory of the obstacle can be provided by the prediction module of the current vehicle. In this embodiment, for each obstacle, the prediction module can provide a single predicted trajectory with the highest prediction probability; alternatively, the prediction module can provide multiple predicted trajectories with similar and relatively high prediction probabilities. The prediction probability can be the probability that the obstacle will travel along the predicted trajectory.

[0036] For example, the prediction module can sort all predicted trajectories in descending order of prediction probability. If the difference between the prediction probability of the first predicted trajectory and the prediction probability of the second predicted trajectory is greater than a set difference, then the first predicted trajectory, i.e., the predicted trajectory with the highest prediction probability, is output. If, among the top N predicted trajectories, the difference between the prediction probabilities of two adjacent predicted trajectories is less than a set difference, then the top N predicted trajectories are output.

[0037] In this embodiment, the predicted trajectory of each obstacle provided by the prediction module can be obtained, thereby obtaining multiple sets of candidate trajectories. For example, if there are multiple obstacles, and there is an obstacle with more than one predicted trajectory, one predicted trajectory can be selected from the predicted trajectories of each obstacle to construct a set of candidate trajectories; or, if there is a single obstacle, and there are multiple predicted trajectories for that obstacle, each predicted trajectory of that obstacle can be used as a set of candidate trajectories.

[0038] like Figure 2 As shown, Figure 2This is a schematic diagram of a set of alternative trajectories in an embodiment of this disclosure. Taking two obstacles as an example, the number of predicted trajectories for each obstacle is 2. The predicted trajectories A and B of obstacle 1 can be combined with the predicted trajectories C and D of obstacle 2 to obtain alternative trajectory sets {A,C}, {A,D}, {B,C}, and {B,D}.

[0039] In this embodiment, the purpose of obtaining multiple candidate trajectory sets is that the predicted trajectory of an obstacle can describe the lateral behavior of the obstacle during travel. For example, if the predicted trajectory is to change lanes first and then go straight, the lateral behavior includes changing lanes and going straight. If an obstacle has multiple predicted trajectories, it indicates that there is an obstacle with uncertain lateral behavior.

[0040] Therefore, in order to determine the lateral behavior of obstacles, multiple sets of alternative trajectories can be constructed to list all possible lateral behaviors of all obstacles. This allows for the analysis of longitudinal behavior for each set of alternative trajectories, and the final lateral behavior can be determined from all possible lateral behaviors by analyzing the optimal longitudinal behavior.

[0041] In addition to acquiring multiple alternative trajectory sets for obstacles, the system can also acquire the current vehicle's pre-decision trajectory. This pre-decision trajectory can be the driving trajectory output by the pre-decision module.

[0042] S120. For each candidate trajectory set, determine the corresponding benefit of each sampled longitudinal behavior scheme in the candidate trajectory set, and determine the target longitudinal behavior scheme of the candidate trajectory set based on each benefit.

[0043] The sampling longitudinal behavior scheme consists of the current vehicle's sampled longitudinal acceleration sequence and the obstacle's sampled longitudinal acceleration sequence. The target longitudinal behavior scheme consists of the current vehicle's target longitudinal acceleration sequence and the obstacle's target longitudinal acceleration sequence.

[0044] In this embodiment, for each set of candidate trajectories, multiple sampled longitudinal behavior schemes can be obtained by sampling the longitudinal behavior of the current vehicle and obstacles. Furthermore, through revenue analysis, the scheme with the optimal revenue is selected from the multiple sampled longitudinal behavior schemes as the target longitudinal behavior scheme.

[0045] It should be noted that the sampled longitudinal acceleration sequence can be composed of sampled longitudinal acceleration at different time points; similarly, the target longitudinal acceleration sequence can be composed of target longitudinal acceleration at different time points.

[0046] To avoid an explosion of computing power, the longitudinal acceleration can be kept constant at each time point in the same lateral behavior, thereby reducing the number of samples during the sampling of the current longitudinal behavior of the vehicle and obstacles.

[0047] For example, optionally, the longitudinal acceleration sequence of the current vehicle includes the longitudinal acceleration of the current vehicle at each time point, wherein all longitudinal accelerations are the same under the same lateral behavior of the current vehicle.

[0048] For example, assuming the lateral behavior of the current vehicle's pre-decision trajectory includes: going straight, changing lanes, going straight again, and the entire pre-decision trajectory includes 12 time points, then the longitudinal acceleration at each time point corresponding to the first straight movement is: 3 m / s². 2 3m / s 2 3m / s 2 3m / s 2 The longitudinal acceleration at each time point corresponding to the lane change is: -2 m / s² 2 -2m / s 2 -2m / s 2 -2m / s 2 The longitudinal acceleration at each time point corresponding to the second straight movement is: 1 m / s². 2 1m / s 2 1m / s 2 1m / s 2 .

[0049] Furthermore, after sampling multiple longitudinal behavior schemes under the candidate trajectory set, the corresponding revenue for each sampling longitudinal behavior scheme can be obtained. This revenue can consist of the cumulative revenue from obstacles and the cumulative revenue of the current vehicle at each individual time point. The cumulative revenue from obstacles can be the sum of the individual revenues of all obstacles at all time points when the current vehicle and obstacles travel according to the sampling longitudinal behavior scheme; the cumulative revenue of the current vehicle at each individual time point can be the sum of the individual revenues of the current vehicle at all time points when the current vehicle and obstacles travel according to the sampling longitudinal behavior scheme.

[0050] In one specific implementation, determining the benefit corresponding to each sampling longitudinal behavior scheme in the candidate trajectory set includes the following steps:

[0051] Step 1: For each sampled longitudinal behavior scheme in the candidate trajectory set, based on the longitudinal acceleration at each time point in the sampled longitudinal acceleration sequence of the current vehicle and the longitudinal acceleration at each time point in the sampled longitudinal acceleration sequence of the obstacle, determine the prediction information of all objects at each time point. Here, all objects include all obstacles and the current vehicle.

[0052] Step 2: Based on the prediction information at each time point, determine the single-point revenue of each object at each time point, and determine the individual cumulative revenue of each object based on the single-point revenue of each object at all time points.

[0053] Step 3: Determine the cumulative revenue of obstacles based on the cumulative revenue of each individual obstacle. Based on the cumulative revenue of obstacles and the cumulative revenue of the current vehicle, determine the revenue corresponding to the sampling longitudinal behavior scheme.

[0054] In step 1, the longitudinal acceleration sequences of the current vehicle and obstacles in the sampling longitudinal behavior scheme can be used to obtain the longitudinal acceleration of the current vehicle and obstacles at each time point. Combined with the position and velocity of the current vehicle and obstacles at the current moment, the predicted information of the current vehicle and all obstacles at each time point can be derived. The predicted information may include predicted longitudinal velocity, predicted lateral velocity, and predicted lateral distance and predicted longitudinal distance with other objects.

[0055] Furthermore, based on the predicted information of the current vehicles and obstacles at each time point, the single-point revenue of each object at each time point can be determined. Then, for each object, the single-point revenue at all time points can be accumulated to obtain the corresponding individual cumulative revenue.

[0056] In one optional implementation, the individual cumulative revenue of each object is determined based on the single-point revenue of each object at all time points, including: determining the weight corresponding to the single-point revenue of each object at each time point based on the order of each time point; and determining the individual cumulative revenue of each object based on the single-point revenue of each object at each time point and the corresponding weight.

[0057] Specifically, the individual returns at each time point can be weighted and summed. The closer a time point is to the current moment, the higher its weight, thus emphasizing the returns of time points close to the current moment throughout the prediction process. This method allows for the fusion of individual returns based on time sequence, further improving the accuracy of returns and facilitating the determination of the optimal sampling longitudinal behavior scheme based on the returns.

[0058] Furthermore, the cumulative individual rewards of all obstacles are summed to obtain the cumulative obstacle reward. This cumulative obstacle reward, along with the current vehicle's cumulative individual reward, is used as the reward corresponding to the sampled longitudinal behavior scheme. Through steps 1-3, the reward corresponding to each sampled longitudinal behavior scheme is accurately determined, facilitating the selection of the optimal longitudinal behavior scheme based on the rewards across the candidate trajectory set, thus obtaining the target longitudinal behavior scheme.

[0059] Regarding step 2 above, optionally, based on the prediction information at each time point, determine the single-point benefit of each object at each time point, including the following steps:

[0060] Step 21: Determine the objects in the candidate trajectory set that interact with other objects as the first object, and determine the objects in the candidate trajectory set that do not interact with other objects as the second object;

[0061] Step 22: Determine the safety gain and parking distance gain of each first object at each time point based on the prediction information at each time point, and determine the smoothness gain and decision consistency gain of each first object at each time point to obtain the single-point gain of each first object at each time point.

[0062] Step 23: Determine the lateral traffic violation penalty, lateral safety benefit, and braking benefit for each second object at each time point based on the prediction information at each time point, and obtain the single-point benefit for each second object at each time point.

[0063] In this embodiment, an object that interacts with other objects can be designated as the first object, and the single-point benefit of the interacting object can be determined through step 22. An object that does not interact with other objects can be designated as the second object, and the single-point benefit of the non-interacting object can be determined through step 23.

[0064] For the first subject, benefit analysis can be conducted from the perspectives of safety, parking distance, ride comfort, and decision consistency. Specifically, for each first subject, the safety benefit and parking distance benefit at each time point can be calculated based on the forecast information at each time point, and the ride comfort benefit and decision consistency benefit at each time point can also be determined.

[0065] Regarding step 22 above, in one example, determining the security gain of each first object at each time point based on the prediction information at each time point includes the following steps:

[0066] Step 221: For each first object, based on the predicted lateral velocity of all objects in the prediction information at each time point, determine the active braking distance or passive braking distance of the first object at each time point, and obtain the lateral safety distance of the first object at each time point based on the determined active braking distance or passive braking distance.

[0067] Step 222: Based on the predicted longitudinal velocity of all objects in the prediction information at each time point, determine the longitudinal safety distance of the first object at each time point. Based on the longitudinal safety distance of the first object at each time point and the predicted longitudinal distance between the first object and other objects in the prediction information at each time point, determine the maximum and minimum longitudinal safety velocity of the first object at each time point.

[0068] Step 223: For each time point, determine whether the predicted lateral distance between the first object and other objects in the prediction information of the time point is less than the lateral safety distance. If so, determine whether the predicted longitudinal velocity of the first object in the prediction information of the time point is between the maximum and minimum longitudinal safety velocity. If not, determine the safety gain of the first object at the time point based on the predicted longitudinal velocity, the maximum longitudinal safety velocity, and the minimum longitudinal safety velocity of the first object.

[0069] Specifically, safety benefits can be analyzed from both lateral and longitudinal perspectives. When calculating safe distances for obstacles, not only lateral but also longitudinal safe distances need to be considered. Absolute safety is impossible on public roads, so when considering safe distances, it's necessary to avoid deliberately engaging in unsafe behaviors with other vehicles.

[0070] Therefore, this embodiment introduces the RSS (Responsibility Sensitive Safety) model to calculate safety benefits using the criteria followed by the RSS model. The criteria followed by the RSS model include: 1. Avoid rear-end collisions with the vehicle in front; 2. Avoid dangerous cutting-in behavior; 3. Right-of-way is granted, not contested; 4. Be cautious of blind spots; 5. If an accident can be avoided safely, take actions to prevent it.

[0071] Specifically, when calculating the lateral safety distance, the safety distance generated by the velocity in the l direction is calculated in the sl coordinate system. In this embodiment, two distances can be introduced, namely the active braking distance and the passive braking distance.

[0072] For example, if the current vehicle is to the right of the obstacle and the obstacle moves in the same lateral direction as the current vehicle, the braking of the obstacle will affect the current vehicle, while the braking of the current vehicle will not affect the obstacle. Therefore, the obstacle is the active brake and the current vehicle is the passive brake.

[0073] In step 221 above, if the first object is the one actively braking, the active braking distance of the first object at each time point can be calculated based on the predicted lateral velocity of all objects in the prediction information. If the first object is the one passively braking, the passive braking distance of the first object at each time point can be calculated based on the predicted lateral velocity of all objects.

[0074] In this embodiment, the calculation of the lateral safety distance assumes that the active lateral braking party suddenly brakes, and the passive party, within its reaction time, intends to accelerate. Given the anticipated collision, the minimum deceleration required to avoid a collision is used for braking. An exemplary formula for calculating the active braking distance is as follows:

[0075]

[0076] Among them, D active The active braking distance, i.e., the distance traveled during active braking, v abs The absolute value of the predicted lateral velocity of the first object. This represents the maximum deceleration during lateral braking.

[0077] The formula for calculating passive braking distance is as follows:

[0078]

[0079] Among them, D passive ρ is the passive braking distance, and ρ is the reaction time. v1 is the minimum deceleration for lateral braking, v1 is the predicted lateral velocity of the first object, v 1，ρ The lateral velocity of the first object during the reaction time is as follows:

[0080]

[0081] in, This represents the maximum acceleration for lateral acceleration.

[0082] Furthermore, based on the calculated active or passive braking distance, the lateral safety distance of the first object can be calculated. Specifically, if the first object is the one actively braking, and the other interacting objects are the ones passively braking, the lateral safety distance of the first object can be determined by the difference between the passive braking distance of the other objects and the active braking distance of the first object, combined with the expected safety distance. If the first object is the one passively braking, and the other interacting objects are the ones actively braking, the lateral safety distance of the first object can be determined by the difference between the passive braking distance of the first object and the active braking distance of the other objects, combined with the expected safety distance. If both the first object and the other interacting objects are passively braking objects, the lateral safety distance of the first object can be determined by the sum of the passive braking distances of the first object and the other objects, combined with the expected safety distance.

[0083] For example, such as Figure 3 As shown, Figure 3 These are schematic diagrams illustrating various scenarios provided in the embodiments of this disclosure. Figure 3Several scenarios are illustrated where the obstacle is located to the right of the current vehicle. Taking the current vehicle as the first example, when the predicted lateral velocities of both the current vehicle and the obstacle are to the left, and the obstacle is located to the right of the current vehicle, the lateral safety distance of the current vehicle satisfies:

[0084] D safe =(D otherpassive -D egoactive )+μ;

[0085] Among them, D otherpassive D is the passive braking distance of the obstacle. egopassive denoted as the current active braking distance of the vehicle, and μ as the minimum expected safe distance;

[0086] When the predicted lateral velocity of the current vehicle is to the left, the predicted lateral velocity of the obstacle is to the right, and the obstacle is located to the right of the current vehicle, the lateral safety distance of the current vehicle satisfies:

[0087] D safe =μ;

[0088] When the predicted lateral velocities of both the current vehicle and the obstacle are to the right, and the obstacle is located to the right of the current vehicle, the current vehicle's lateral safe distance satisfies:

[0089] D safe =(D egopassive -D otheractive )+μ;

[0090] When the predicted lateral velocity of the current vehicle is to the right, the predicted lateral velocity of the obstacle is to the left, and the obstacle is located to the right of the current vehicle, the lateral safety distance of the current vehicle satisfies:

[0091] D safe =(D egopassive +D otheractive )+μ.

[0092] For example, such as Figure 4 As shown, Figure 4 These are schematic diagrams illustrating various scenarios provided in the embodiments of this disclosure. Figure 4 Several scenarios are illustrated where the obstacle is located to the left of the current vehicle. Taking the current vehicle as the first example, when the predicted lateral velocities of both the current vehicle and the obstacle are to the right, and the obstacle is located to the left of the current vehicle, the lateral safety distance of the current vehicle satisfies:

[0093] D safe =(D otherpassive -D egoactive )+μ;

[0094] When the predicted lateral velocity of the current vehicle is to the left, the predicted lateral velocity of the obstacle is to the right, and the obstacle is located to the left of the current vehicle, the lateral safety distance of the current vehicle satisfies:

[0095] D safe =(D egopassive +D otherpassive )+μ;

[0096] When the predicted lateral velocities of both the current vehicle and the obstacle are to the left, and the obstacle is located to the left of the current vehicle, the lateral safe distance of the current vehicle satisfies:

[0097] D safe =(D egopassive -D otheractive )+μ;

[0098] When the predicted lateral velocity of the current vehicle is to the left, the predicted lateral velocity of the obstacle is to the right, and the obstacle is located to the left of the current vehicle, the lateral safety distance of the current vehicle satisfies:

[0099] D safe =μ.

[0100] After calculating the lateral safety distance, the longitudinal safety distance of the first object is further calculated. The hypothetical scenario for the longitudinal safety distance is similar to that for the lateral safety distance: the vehicle in front suddenly brakes, and the vehicle behind intends to accelerate within the reaction time. Given the observed possibility of a collision, the vehicle brakes using the minimum deceleration to avoid a collision. In this embodiment, considering the different safety speeds and safety distances for obstacles traveling in the same direction and those traveling in the same direction, the longitudinal safety distances for these two types of obstacles can be distinguished.

[0101] Specifically, if the first object is traveling in the same direction as other objects, the longitudinal safety distance of the first object is calculated using the following formula:

[0102]

[0103] In the formula, D min1 For longitudinal safety distance, v r v is the predicted longitudinal velocity of the obstacle behind the first object. f Let ρ be the predicted longitudinal velocity of the obstacle in front of the first object, and a be the reaction time. maxacc For the maximum reaction acceleration, a minbrake For the minimum braking deceleration, a maxbrake This is the maximum braking deceleration;

[0104] If the first object is traveling in the opposite direction to other objects, the longitudinal safety distance of the first object is calculated using the following formula:

[0105]

[0106] In the formula, D min2 v1 is the longitudinal safety distance, v2 is the predicted longitudinal velocity of the first object, v 2，ρ v1 is the longitudinal velocity of the first object within the reaction time, v2 is the predicted longitudinal velocity of other obstacles traveling in the opposite direction, and v3 is the longitudinal velocity of the first object within the reaction time. 3，ρ For the longitudinal velocity of the other obstacle during the reaction time, specifically:

[0107]

[0108] In the formula, v′ abs The absolute value of the predicted longitudinal velocity of the first object. This represents the maximum acceleration in the longitudinal direction.

[0109] After obtaining the longitudinal safety distance of each first object, it is further possible to compare the longitudinal safety distance with the predicted longitudinal distance between the first object and other objects.

[0110] For example, if the first object and other objects are traveling in the same direction, and the predicted longitudinal distance is less than the longitudinal safety distance, and the first object is behind the other objects, then the maximum longitudinal safety speed is determined. The minimum longitudinal safe speed is a preset first speed (e.g., a very small value); if the predicted longitudinal distance is less than the longitudinal safe distance, and the first object is in front of other objects, then the minimum longitudinal safe speed is determined. The maximum longitudinal safe speed is the preset second speed (e.g., a very large value). Where D is the predicted longitudinal distance, and a... brakemax This is the maximum braking deceleration.

[0111] If the first object travels in the same direction as other objects, and the predicted longitudinal distance is greater than the longitudinal safety distance, then the minimum longitudinal safety speed is determined to be 0, and the maximum longitudinal safety speed is the preset second speed.

[0112] When the first object is traveling in the opposite direction to the other objects, the minimum longitudinal safe speed can be determined to be 0, and the maximum longitudinal safe speed is the preset second speed.

[0113] Furthermore, for each time point, we can first determine whether the predicted lateral distance between the first object and other objects is less than the lateral safety distance. If not, it means that the predicted lateral distance meets the lateral safety distance and is safe in the lateral direction. If yes, it means that the predicted lateral distance does not meet the lateral safety distance and we need to further determine whether the predicted longitudinal velocity of the first object meets the corresponding maximum and minimum longitudinal safety velocities.

[0114] Specifically, if the predicted longitudinal speed is between the maximum and minimum safe longitudinal speed, it indicates that there is safety in the longitudinal direction; otherwise, it indicates that there is no safety in the longitudinal direction. In this case, the safety benefit can be calculated further based on the predicted longitudinal speed, the maximum safe longitudinal speed, and the minimum safe longitudinal speed.

[0115] For example, if the predicted longitudinal velocity is greater than the maximum safe longitudinal velocity, it indicates that the longitudinal unsafety of the first object is due to excessive longitudinal velocity. In this case, the safety gain can be:

[0116]

[0117] If the predicted longitudinal velocity is less than the minimum safe longitudinal velocity, it indicates that the longitudinal unsafety of the first object is due to its excessively slow longitudinal velocity. In this case, the safety gain can be:

[0118]

[0119] Cost toofast Cost tooslow These represent the safety gains for excessively high and excessively low longitudinal speeds, respectively, ω. fastlinear ω slowlinear These are the first adjustment weights for excessively high and excessively low longitudinal velocities, ω. fastpower ω slowpower These represent the second adjustment weights corresponding to excessively high and excessively low longitudinal velocities, respectively, where v is the predicted longitudinal velocity. upper v lower These are the maximum and minimum longitudinal safe speeds, respectively.

[0120] Through steps 221-223 above, the security gains of the first object at each point in time are determined, thereby enabling security analysis of objects that interact with other objects.

[0121] Regarding step 22 above, in one example, determining the parking distance gain for each first object at each time point based on the prediction information at each time point includes the following steps:

[0122] Step 224: For each first object, determine whether the predicted parking distance of the first object in the prediction information at each time point is less than the parking distance threshold;

[0123] Step 225: If yes, then determine the parking distance benefit of the first object at a given time point based on the difference between the predicted parking distance and the parking distance threshold; wherein, the greater the difference between the predicted parking distance and the parking distance threshold, the greater the parking distance benefit.

[0124] In this embodiment, to encourage vehicles and obstacles to travel further away, a parking distance bonus can be applied to vehicles that stop quickly. Specifically, the prediction information for the first object may also include the predicted parking distance, that is, the distance traveled by the first object from the current time to the time of stopping.

[0125] In step 224 above, it can be determined first whether the predicted parking distance is less than the parking distance threshold; wherein, the parking distance threshold can be a pre-set critical distance for applying parking distance benefits, such as 10m.

[0126] If the predicted parking distance is less than a parking distance threshold, the parking distance revenue can be further determined based on the difference between the predicted parking distance and the parking distance threshold. The larger the difference between the predicted parking distance and the parking distance threshold, the greater the corresponding parking distance revenue. For example, a revenue correlation table can be pre-set, which includes the parking distance revenue corresponding to each difference, and then the corresponding parking distance revenue can be retrieved from the revenue correlation table based on the difference.

[0127] Through steps 224-225 above, the parking distance benefit is determined, which facilitates the subsequent selection of the optimal longitudinal behavior plan based on the parking distance benefit. This helps to avoid the current vehicle's planning module planning a plan that causes the current vehicle to suddenly stop with an obstacle, thereby further improving driving safety and ensuring the user's driving experience.

[0128] Regarding step 22 above, in one example, determining the smoothness gain and decision consistency gain for each first object at each time point includes the following steps:

[0129] Step 226: For each first object, determine the smoothness gain of the first object at each time point based on the acceleration difference between adjacent time points in the sampled longitudinal acceleration sequence of the first object;

[0130] Step 227: Based on the difference between the sampled longitudinal acceleration sequence of the first object and the target longitudinal acceleration sequence of the first object in the target longitudinal behavior scheme of the previous frame, determine the decision consistency benefit of the first object at each time point.

[0131] Specifically, smoothness gain can reflect the degree of change in the longitudinal acceleration of the first object. The greater the acceleration difference between adjacent time points in the sampled longitudinal acceleration sequence of the first object, the greater the corresponding smoothness gain.

[0132] The decision consistency benefit reflects the difference between the sampled longitudinal acceleration sequence of the first object and the optimal longitudinal acceleration sequence of the first object in the previous frame. The greater the difference between the sampled longitudinal acceleration sequence of the first object and the target longitudinal acceleration sequence of the first object in the target longitudinal behavior scheme of the previous frame, the greater the corresponding decision consistency benefit.

[0133] Through steps 226-227 above, the smoothness benefits and decision consistency benefits are determined, which facilitates the subsequent selection of the optimal longitudinal behavior scheme by combining the smoothness benefits and decision consistency benefits. This helps to avoid the current vehicle's planning module from planning a scheme with low driving smoothness, ensuring the driving smoothness of the current vehicle and obstacles, further improving driving safety and ensuring the user's driving experience.

[0134] In this embodiment, for a second object that does not interact with other objects, considering that such an object still has a safety impact on other objects in the lateral direction, the single-point benefit of the second object at each time point can be obtained by combining lateral traffic rule penalties, lateral safety benefits, and braking benefits.

[0135] In this context, lateral traffic rule penalties can be understood as negative benefits. Specifically, if a second party exhibits unreasonable behavior within the interaction timeframe, a negative benefit, namely a lateral traffic rule penalty, can be imposed.

[0136] In one example, determining the lateral traffic rule penalty for each second object at each time point based on the prediction information at each time point includes the following steps:

[0137] Step 231: For each second object, determine the collision prediction point or minimum distance prediction point between the second object and other objects based on the prediction information at each time point, and determine the interaction time range of the first object based on the collision prediction point or minimum distance prediction point.

[0138] Step 232: Based on the prediction information at each time point, determine whether the second object exhibits unreasonable behavior within the interaction time range. If so, determine the lateral traffic rule penalty for the second object within the interaction time range.

[0139] Among them, the collision prediction point can be the location point where the second object collides with other objects, and the minimum distance prediction point can be the location point where the distance between the second object and other objects is the shortest.

[0140] Specifically, when there is a collision between the second object and other objects, a collision prediction point can be determined, and the interaction time range can be determined based on the collision prediction point. For example, the time point corresponding to the collision prediction point, as well as the set time period before the collision prediction point and the set time period after the collision prediction point, can be determined as the interaction time range.

[0141] Alternatively, if there is no collision between the second object and other objects, the minimum distance prediction point can be determined, and the interaction time range can be determined based on the minimum distance prediction point. For example, the time point corresponding to the minimum distance prediction point, the set time period before the minimum distance prediction point, and the set time period after the minimum distance prediction point can be determined as the interaction time range.

[0142] Furthermore, the prediction information can be used to determine whether the second object exhibits unreasonable behavior within the interaction timeframe. Unreasonable behavior could include violations of traffic regulations, such as being at fault in a vehicle collision, or reversing.

[0143] Through steps 231-232 above, the lateral traffic rule penalty for the second object that does not have interaction with other objects is determined. This facilitates the subsequent selection of the optimal longitudinal behavior scheme by combining the lateral traffic rule penalty, so as to avoid the current vehicle's planning module planning a scheme with unreasonable behavior as much as possible.

[0144] In this embodiment, the lateral security benefit can reflect the lateral distance between the second object and other objects. The smaller the lateral distance, the greater the influence of other objects on the second object. That is, other objects can more easily influence the behavior of the second object, and therefore, the greater the corresponding lateral security benefit.

[0145] In one example, determining the horizontal security gain of each second object at each time point based on the prediction information at each time point includes the following steps:

[0146] Step 233: Based on the polygon radius corresponding to the second object and the polygon radii corresponding to other objects, determine the minimum safe distance between the second object and other objects;

[0147] Step 234: Based on the Cartesian predicted distance between the second object and other objects in the prediction information at each time point, and the minimum safe distance, determine the lateral safety gain of the second object at each time point.

[0148] The polygon radius can be the radius of the bounding box of the polygon centered on the object. Specifically, the minimum safe distance can be determined by the sum of the polygon radii corresponding to the second object and the polygon radii corresponding to the other objects.

[0149] Furthermore, it is determined whether the minimum safe distance is less than the Cartesian predicted distance. If so, the lateral safety gain is set to the preset gain (e.g., 1); otherwise, the lateral safety gain is calculated based on the minimum safe distance and the Cartesian predicted distance. For example, see the following formula:

[0150]

[0151] in, For horizontal security benefits, D k For the minimum safe distance, S k Predict distances for Descartes.

[0152] Through steps 233-234 above, the lateral safety benefits of a second object that does not interact with other objects are determined. This facilitates the subsequent selection of the optimal longitudinal behavior scheme based on the lateral safety benefits, so as to avoid the current vehicle's planning module planning a scheme that affects the driving safety of the current vehicle and obstacles.

[0153] In this embodiment, braking benefit can reflect the degree of unsolvability of the behavior of the second object. Specifically, if the behavior of other objects makes the behavior of the second object unsolvable, it means that the behavior of other objects forces the second object to need greater braking force. At this time, braking benefit can be assigned, which means that the influence of other objects on the second object is greater.

[0154] In one example, determining the braking benefit of each second object at each time point based on the prediction information at each time point includes: determining whether the second object is braking based on the prediction information at each time point; if so, determining the braking benefit of the second object at the braking time point based on the braking acceleration of the second object at the braking time point.

[0155] Specifically, the prediction information can be used to determine whether the second object is braking. If so, the braking acceleration of the second object at the braking time point can be obtained. The braking acceleration can reflect the braking force of the second object. The greater the braking acceleration, the stronger the braking force.

[0156] Furthermore, the braking benefit at the braking time point can be obtained based on the braking acceleration. The greater the braking acceleration, the greater the braking benefit.

[0157] Through the above implementation method, the braking benefits of a second object that does not interact with other objects are determined, which facilitates the subsequent selection of the optimal longitudinal behavior scheme based on the braking benefits, so as to avoid the current vehicle's planning module planning a scheme that forces the current vehicle to brake against obstacles.

[0158] It should be noted that in this embodiment, by defining the interactive object as the first object and the non-interactive object as the second object, different benefit analysis schemes are formulated for the first and second objects respectively, thus realizing benefit analysis for both interactive and non-interactive objects. For interactive objects, specific analyses are performed on safety benefits, parking distance benefits, ride comfort benefits, and decision consistency benefits. For non-interactive objects, specific analyses are performed on lateral traffic violation penalties, lateral safety benefits, and braking benefits, which can better mitigate potential future driving risks.

[0159] Furthermore, for each set of candidate trajectories, the sampling longitudinal action scheme with the best benefit can be selected from the set of candidate trajectories as the corresponding target longitudinal action scheme.

[0160] For example, based on the cumulative benefit of obstacles, the sampled longitudinal behavior scheme with the largest cumulative benefit of obstacles is selected as the target longitudinal behavior scheme. If there are multiple sampled longitudinal behavior schemes with the same cumulative benefit of obstacles, the target longitudinal behavior scheme can be determined by further combining the current vehicle's individual cumulative benefit.

[0161] S130. Based on the benefits corresponding to the target longitudinal behavior schemes of each candidate trajectory set, determine the target trajectory set from each candidate trajectory set, and determine decision information based on the target trajectory set and the target longitudinal behavior schemes of the target trajectory set.

[0162] Specifically, obtaining the target longitudinal behavior scheme for each set of candidate trajectories can be understood as determining the corresponding optimal longitudinal behavior scheme for each set of candidate trajectories, that is, the optimal longitudinal behavior of the current vehicle and obstacles.

[0163] Furthermore, it is also necessary to determine the optimal lateral behavior of the obstacle. In this embodiment, the set of candidate trajectories with the best return can be selected as the target trajectory set based on the returns corresponding to the target longitudinal behavior schemes of each set of candidate trajectories.

[0164] For example, Figure 5 This is a schematic diagram illustrating the determination of a target trajectory set in an embodiment of this disclosure. For example... Figure 5 As shown, P represents the cumulative benefit from obstacles, and Q represents the cumulative benefit of the current vehicle. "P:5, Q:3" represents the benefit corresponding to the target longitudinal behavior scheme of the candidate trajectory set AC. Based on the benefits corresponding to the target longitudinal behavior schemes of each candidate trajectory set, it can be seen that the candidate trajectory set BC has the best benefit. Therefore, the candidate trajectory set BC can be used as the target trajectory set.

[0165] Furthermore, the target trajectory set and the target longitudinal behavior plan of the target trajectory set can be used as decision information, and the decision information can be output to the planning module of the current vehicle.

[0166] The obstacle prediction trajectory decision-making method provided in this embodiment obtains a set of candidate trajectories consisting of a predicted trajectory corresponding to each obstacle. For each candidate trajectory set, the benefit corresponding to each sampled longitudinal behavior scheme is determined to obtain the target longitudinal behavior scheme. Then, the target trajectory set is determined by the benefit corresponding to the target longitudinal behavior scheme of each candidate trajectory set, thus realizing the decision on the lateral behavior of the obstacle. Furthermore, the decision information is determined by the target trajectory set and the corresponding target longitudinal behavior scheme, thus realizing the decision on the longitudinal behavior of the obstacle and the current vehicle. This method can perform intent analysis on all obstacles, solving the problem of low decision accuracy caused by only considering the trajectory of obstacles interacting with the vehicle in the prior art. Moreover, this method can determine the optimal lateral behavior and corresponding optimal longitudinal behavior of the obstacle, as well as the optimal longitudinal behavior of the current vehicle, when the probabilities of multiple predicted trajectories of the obstacle are similar, thus better solving the problem of difficulty in evaluating the similarity of predicted trajectory probabilities from a global perspective.

[0167] Furthermore, this method evaluates the uncertain behavior of obstacles and the current vehicle by considering benefits such as safety gains and traffic regulations penalties, and finds the trajectory that best matches the current vehicle's expectations. By using sampling and performing benefit analysis on each sampling scheme, it better evaluates the behavior between obstacles and the current vehicle, supports interactions of uncertain behaviors between multiple obstacles and the current vehicle, and better addresses the problem of difficulty in evaluating equal predicted trajectory probabilities from a global perspective.

[0168] Figure 6 This is a schematic diagram of the structure of a decision-making device for predicting obstacle trajectories according to an embodiment of this disclosure. Figure 6 As shown: The device includes: a set acquisition module 610, a longitudinal behavior determination module 620, and a trajectory decision module 630.

[0169] The set acquisition module 610 is used to acquire multiple candidate trajectory sets of obstacles, wherein each candidate trajectory set consists of a predicted trajectory corresponding to each obstacle, and the candidate trajectory sets are not completely identical.

[0170] The longitudinal behavior determination module 620 is used to determine the benefit corresponding to each sampled longitudinal behavior scheme in each candidate trajectory set for each candidate trajectory set, and to determine the target longitudinal behavior scheme of the candidate trajectory set based on each benefit, wherein the sampled longitudinal behavior scheme is composed of the sampled longitudinal acceleration sequence of the current vehicle and the sampled longitudinal acceleration sequence of each obstacle.

[0171] The trajectory decision module 630 is used to determine a target trajectory set from the candidate trajectory sets based on the benefits corresponding to the target longitudinal behavior schemes of each candidate trajectory set, and to determine decision information based on the target trajectory set and the target longitudinal behavior schemes of the target trajectory set.

[0172] The obstacle prediction trajectory decision-making device provided in this disclosure embodiment can execute the steps in the obstacle prediction trajectory decision-making method provided in this disclosure method embodiment, and has the execution steps and beneficial effects, which will not be repeated here.

[0173] Figure 7 This is a schematic diagram of the structure of an electronic device according to an embodiment of this disclosure. See below for details. Figure 7 It shows a schematic diagram of a structure suitable for implementing the electronic device 500 in the embodiments of this disclosure. Figure 7 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0174] like Figure 7 As shown, the electronic device 500 may include a processing device (e.g., a central processing unit, a graphics processor, etc.) 501, which can perform various appropriate actions and processes to implement the methods of the embodiments described herein, based on a program stored in a read-only memory (ROM) 502 or a program loaded from a storage device 508 into a random access memory (RAM) 503. The RAM 503 also stores various programs and data required for the operation of the electronic device 500. The processing device 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0175] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts, thereby implementing the obstacle prediction trajectory decision-making method as described above. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 509, or installed from a storage device 508, or installed from a ROM 502. When the computer program is executed by the processing device 501, it performs the functions defined in the methods of embodiments of this disclosure.

[0176] It should be noted that the computer-readable medium described in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0177] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs that, when executed by the electronic device, cause the electronic device to:

[0178] Obtain multiple sets of alternative trajectories for obstacles, wherein each set of alternative trajectories consists of a predicted trajectory corresponding to each obstacle, and the sets of alternative trajectories are not completely identical;

[0179] For each set of candidate trajectories, the benefit corresponding to each sampled longitudinal behavior scheme in the set of candidate trajectories is determined, and the target longitudinal behavior scheme of the set of candidate trajectories is determined based on each benefit. The sampled longitudinal behavior scheme is composed of the sampled longitudinal acceleration sequence of the current vehicle and the sampled longitudinal acceleration sequence of the obstacle.

[0180] Based on the benefits corresponding to the target longitudinal behavior schemes of each set of candidate trajectories, a target trajectory set is determined from each set of candidate trajectories, and decision information is determined based on the target trajectory set and the target longitudinal behavior schemes of the target trajectory set.

[0181] Optionally, when one or more of the above-described procedures are executed by the electronic device, the electronic device may also perform other steps described in the above embodiments.

[0182] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0183] Option 1: A decision-making method for predicting obstacle trajectories, the method comprising:

[0184] Obtain multiple sets of alternative trajectories for obstacles, wherein each set of alternative trajectories consists of a predicted trajectory corresponding to each obstacle, and the sets of alternative trajectories are not completely identical;

[0185] For each set of candidate trajectories, the benefit corresponding to each sampled longitudinal behavior scheme in the set of candidate trajectories is determined, and the target longitudinal behavior scheme of the set of candidate trajectories is determined based on each benefit. The sampled longitudinal behavior scheme is composed of the sampled longitudinal acceleration sequence of the current vehicle and the sampled longitudinal acceleration sequence of the obstacle.

[0186] Based on the benefits corresponding to the target longitudinal behavior schemes of each set of candidate trajectories, a target trajectory set is determined from each set of candidate trajectories, and decision information is determined based on the target trajectory set and the target longitudinal behavior schemes of the target trajectory set.

[0187] Option 2, according to the method described in Option 1, the step of determining the benefit corresponding to each sampling longitudinal behavior scheme in the candidate trajectory set includes:

[0188] For each sampled longitudinal behavior scheme in the candidate trajectory set, based on the longitudinal acceleration at each time point in the sampled longitudinal acceleration sequence of the current vehicle and the longitudinal acceleration at each time point in the sampled longitudinal acceleration sequence of the obstacle, the prediction information of all objects at each time point is determined, wherein all objects include all obstacles and the current vehicle.

[0189] Based on the forecast information at each time point, determine the single-point revenue of each object at each time point, and determine the individual cumulative revenue of each object based on the single-point revenue of each object at all time points.

[0190] The cumulative revenue of obstacles is determined based on the cumulative revenue of each individual obstacle. Based on the cumulative revenue of obstacles and the cumulative revenue of the current vehicle, the revenue corresponding to the sampling longitudinal behavior scheme is determined.

[0191] Option 3: Based on the method described in Option 2, determining the single-point benefit of each object at each time point based on the prediction information at each time point includes:

[0192] The object in the candidate trajectory set that interacts with other objects is determined as the first object, and the object in the candidate trajectory set that does not interact with other objects is determined as the second object;

[0193] Based on the prediction information at each time point, determine the safety gain and parking distance gain of each first object at each time point, and determine the smoothness gain and decision consistency gain of each first object at each time point, so as to obtain the single-point gain of each first object at each time point.

[0194] Based on the prediction information at each time point, determine the lateral traffic violation penalty, lateral safety benefit, and braking benefit for each second object at each time point, and obtain the single-point benefit for each second object at each time point.

[0195] Option 4: Based on the method described in Option 3, determine the security gain of each first object at each time point based on the prediction information at each time point, including:

[0196] For each of the first objects, based on the predicted lateral velocities of all objects in the prediction information at each time point, the active braking distance or passive braking distance of the first object at each time point is determined, and the lateral safety distance of the first object at each time point is obtained based on the determined active braking distance or passive braking distance.

[0197] Based on the predicted longitudinal velocity of all objects in the prediction information at each time point, the longitudinal safety distance of the first object at each time point is determined. Based on the longitudinal safety distance of the first object at each time point and the predicted longitudinal distance between the first object and other objects in the prediction information at each time point, the maximum and minimum longitudinal safety velocities of the first object at each time point are determined.

[0198] For each time point, it is determined whether the predicted lateral distance between the first object and other objects in the prediction information of the time point is less than the lateral safety distance. If so, it is determined whether the predicted longitudinal speed of the first object in the prediction information of the time point is between the maximum value of the longitudinal safety speed and the minimum value of the longitudinal safety speed. If not, the safety gain of the first object at the time point is determined based on the predicted longitudinal speed of the first object, the maximum value of the longitudinal safety speed, and the minimum value of the longitudinal safety speed.

[0199] Option 5: According to the method described in Option 3, determine the parking distance benefit for each of the first objects at each time point based on the prediction information at each time point, including:

[0200] For each of the first objects, determine whether the predicted parking distance of the first object in the prediction information at each time point is less than the parking distance threshold;

[0201] If so, then the parking distance gain of the first object at the time point is determined based on the difference between the predicted parking distance and the parking distance threshold;

[0202] The greater the difference between the predicted parking distance and the parking distance threshold, the greater the parking distance benefit.

[0203] Option 6: Based on the method described in Option 3, determine the smoothness benefit and decision consistency benefit for each of the first objects at each time point, including:

[0204] For each of the first objects, the smoothness gain of the first object at each time point is determined based on the acceleration difference between adjacent time points in the sampled longitudinal acceleration sequence of the first object.

[0205] Based on the difference between the sampled longitudinal acceleration sequence of the first object and the target longitudinal acceleration sequence of the first object in the target longitudinal behavior scheme of the previous frame, the decision consistency gain of the first object at each time point is determined.

[0206] Option 7: According to the method described in Option 3, determine the lateral traffic violation penalty for each second object at each time point based on the prediction information at each time point, including:

[0207] For each of the second objects, the collision prediction point or minimum distance prediction point between the second object and other objects is determined based on the prediction information at each time point, and the interaction time range of the first object is determined based on the collision prediction point or minimum distance prediction point.

[0208] Based on the prediction information at each time point, determine whether the second object exhibits unreasonable behavior within the interaction time range. If so, determine the lateral traffic rule penalty for the second object within the interaction time range.

[0209] Option 8: Based on the method described in Option 3, determine the horizontal security gain of each second object at each time point based on the prediction information at each time point, including:

[0210] Based on the polygon radius corresponding to the second object and the polygon radii corresponding to other objects, determine the minimum safe distance between the second object and other objects;

[0211] Based on the Cartesian predicted distances between the second object and other objects in the prediction information at each time point, and the minimum safe distance, the lateral safety gain of the second object at each time point is determined.

[0212] Option 9: According to the method described in Option 3, determine the braking benefit of each second object at each time point based on the prediction information at each time point, including:

[0213] Based on the prediction information at each time point, determine whether the second object is braking. If so, determine the braking benefit of the second object at the braking time point based on the braking acceleration of the second object at the braking time point.

[0214] Option 10: According to the method described in Option 2, the step of determining the individual cumulative revenue of each object based on the single-point revenue of each object at all time points includes:

[0215] The weight of each object's single-point gain at each time point is determined based on the order of the time points.

[0216] The cumulative individual return of each object is determined based on its single-point return at each time point and its corresponding weight.

[0217] Option 11: According to the method described in Option 1, the longitudinal acceleration sequence of the current vehicle includes the longitudinal acceleration of the current vehicle at each time point, wherein all longitudinal accelerations under the same lateral behavior of the current vehicle are the same.

[0218] Option 12: A decision-making device for predicting obstacle trajectories, comprising:

[0219] The set acquisition module is used to acquire multiple candidate trajectory sets for obstacles, wherein each candidate trajectory set consists of a predicted trajectory corresponding to each obstacle, and the candidate trajectory sets are not completely identical.

[0220] The longitudinal behavior determination module is used to determine the benefit corresponding to each sampled longitudinal behavior scheme in each candidate trajectory set for each candidate trajectory set, and to determine the target longitudinal behavior scheme of the candidate trajectory set based on each benefit, wherein the sampled longitudinal behavior scheme is composed of the sampled longitudinal acceleration sequence of the current vehicle and the sampled longitudinal acceleration sequence of each obstacle;

[0221] The trajectory decision module is used to determine the target trajectory set from the candidate trajectory sets based on the benefits corresponding to the target longitudinal behavior schemes of each candidate trajectory set, and to determine decision information based on the target trajectory set and the target longitudinal behavior schemes of the target trajectory set.

[0222] Option 13: An electronic device, the electronic device comprising:

[0223] One or more processors;

[0224] Storage device for storing one or more programs;

[0225] When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of schemes 1-11.

[0226] Option 14: A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method as described in any one of Options 1-11.

[0227] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features disclosed in this disclosure that have similar functions.

Claims

1. A decision-making method for predicting obstacle trajectories, characterized in that, The method includes: Obtain multiple sets of alternative trajectories for obstacles, wherein each set of alternative trajectories consists of a predicted trajectory corresponding to each obstacle, and the sets of alternative trajectories are not completely identical; For each set of candidate trajectories, the benefit corresponding to each sampled longitudinal behavior scheme in the set of candidate trajectories is determined, and the target longitudinal behavior scheme of the set of candidate trajectories is determined based on each benefit. The sampled longitudinal behavior scheme is composed of the sampled longitudinal acceleration sequence of the current vehicle and the sampled longitudinal acceleration sequence of the obstacle. Based on the benefits corresponding to the target longitudinal behavior schemes of each set of candidate trajectories, a target trajectory set is determined from each set of candidate trajectories, and decision information is determined based on the target trajectory set and the target longitudinal behavior schemes of the target trajectory set. Wherein, determining the benefit corresponding to each sampling longitudinal behavior scheme in the candidate trajectory set includes: For each sampled longitudinal behavior scheme in the candidate trajectory set, based on the longitudinal acceleration at each time point in the sampled longitudinal acceleration sequence of the current vehicle and the longitudinal acceleration at each time point in the sampled longitudinal acceleration sequence of the obstacle, the prediction information of all objects at each time point is determined, wherein all objects include all obstacles and the current vehicle. Based on the forecast information at each time point, determine the single-point revenue of each object at each time point, and determine the individual cumulative revenue of each object based on the single-point revenue of each object at all time points. The cumulative revenue of obstacles is determined based on the cumulative revenue of each individual obstacle. Based on the cumulative revenue of obstacles and the cumulative revenue of the current vehicle, the revenue corresponding to the sampling longitudinal behavior scheme is determined.

2. The method according to claim 1, characterized in that, The step of determining the single-point benefit of each object at each time point based on the prediction information at each time point includes: The object in the candidate trajectory set that interacts with other objects is determined as the first object, and the object in the candidate trajectory set that does not interact with other objects is determined as the second object; Based on the prediction information at each time point, determine the safety gain and parking distance gain of each first object at each time point, and determine the smoothness gain and decision consistency gain of each first object at each time point, so as to obtain the single-point gain of each first object at each time point. Based on the prediction information at each time point, determine the lateral traffic violation penalty, lateral safety benefit, and braking benefit for each second object at each time point, and obtain the single-point benefit for each second object at each time point.

3. The method according to claim 2, characterized in that, Based on the prediction information at each time point, determine the security gain of each of the first objects at each time point, including: For each of the first objects, based on the predicted lateral velocities of all objects in the prediction information at each time point, the active braking distance or passive braking distance of the first object at each time point is determined, and the lateral safety distance of the first object at each time point is obtained based on the determined active braking distance or passive braking distance. Based on the predicted longitudinal velocity of all objects in the prediction information at each time point, the longitudinal safety distance of the first object at each time point is determined. Based on the longitudinal safety distance of the first object at each time point and the predicted longitudinal distance between the first object and other objects in the prediction information at each time point, the maximum and minimum longitudinal safety velocities of the first object at each time point are determined. For each time point, it is determined whether the predicted lateral distance between the first object and other objects in the prediction information of the time point is less than the lateral safety distance. If so, it is determined whether the predicted longitudinal speed of the first object in the prediction information of the time point is between the maximum value of the longitudinal safety speed and the minimum value of the longitudinal safety speed. If not, the safety gain of the first object at the time point is determined based on the predicted longitudinal speed of the first object, the maximum value of the longitudinal safety speed, and the minimum value of the longitudinal safety speed.

4. The method according to claim 2, characterized in that, Based on the prediction information at each time point, determine the parking distance benefit for each of the first objects at each time point, including: For each of the first objects, determine whether the predicted parking distance of the first object in the prediction information at each time point is less than the parking distance threshold; If so, then the parking distance gain of the first object at the time point is determined based on the difference between the predicted parking distance and the parking distance threshold; The greater the difference between the predicted parking distance and the parking distance threshold, the greater the parking distance benefit.

5. The method according to claim 2, characterized in that, Determine the smoothness gain and decision consistency gain for each of the first objects at each time point, including: For each of the first objects, the smoothness gain of the first object at each time point is determined based on the acceleration difference between adjacent time points in the sampled longitudinal acceleration sequence of the first object. Based on the difference between the sampled longitudinal acceleration sequence of the first object and the target longitudinal acceleration sequence of the first object in the target longitudinal behavior scheme of the previous frame, the decision consistency gain of the first object at each time point is determined.

6. The method according to claim 2, characterized in that, Based on the prediction information at each time point, determine the lateral traffic violation penalty for each of the second objects at each time point, including: For each of the second objects, the collision prediction point or minimum distance prediction point between the second object and other objects is determined based on the prediction information at each time point, and the interaction time range of the first object is determined based on the collision prediction point or minimum distance prediction point. Based on the prediction information at each time point, determine whether the second object exhibits unreasonable behavior within the interaction time range. If so, determine the lateral traffic rule penalty for the second object within the interaction time range.

7. The method according to claim 2, characterized in that, Based on the prediction information at each time point, determine the horizontal security gain of each second object at each time point, including: Based on the polygon radius corresponding to the second object and the polygon radii corresponding to other objects, determine the minimum safe distance between the second object and other objects; Based on the Cartesian predicted distances between the second object and other objects in the prediction information at each time point, and the minimum safe distance, the lateral safety gain of the second object at each time point is determined.

8. The method according to claim 2, characterized in that, Based on the prediction information at each time point, determine the braking benefit of each second object at each time point, including: Based on the prediction information at each time point, determine whether the second object is braking. If so, determine the braking benefit of the second object at the braking time point based on the braking acceleration of the second object at the braking time point.

9. The method according to claim 1, characterized in that, The process of determining the individual cumulative revenue of each object based on its single-point revenue at all time points includes: The weight of each object's single-point gain at each time point is determined based on the order of the time points. The cumulative individual return of each object is determined based on its single-point return at each time point and its corresponding weight.

10. The method according to claim 1, characterized in that, The longitudinal acceleration sequence of the current vehicle includes the longitudinal acceleration of the current vehicle at each time point, wherein all longitudinal accelerations under the same lateral behavior of the current vehicle are the same.

11. A decision-making device for predicting obstacle trajectories, characterized in that, include: The set acquisition module is used to acquire multiple candidate trajectory sets for obstacles, wherein each candidate trajectory set consists of a predicted trajectory corresponding to each obstacle, and the candidate trajectory sets are not completely identical. The longitudinal behavior determination module is used to determine the benefit corresponding to each sampled longitudinal behavior scheme in each candidate trajectory set for each candidate trajectory set, and to determine the target longitudinal behavior scheme of the candidate trajectory set based on each benefit, wherein the sampled longitudinal behavior scheme is composed of the sampled longitudinal acceleration sequence of the current vehicle and the sampled longitudinal acceleration sequence of each obstacle; The trajectory decision module is used to determine the target trajectory set from the candidate trajectory sets based on the benefits corresponding to the target longitudinal behavior schemes of each candidate trajectory set, and to determine decision information based on the target trajectory set and the target longitudinal behavior schemes of the target trajectory set. The longitudinal behavior determination module is used for: For each sampled longitudinal behavior scheme in the candidate trajectory set, based on the longitudinal acceleration at each time point in the sampled longitudinal acceleration sequence of the current vehicle and the longitudinal acceleration at each time point in the sampled longitudinal acceleration sequence of the obstacle, the prediction information of all objects at each time point is determined, wherein all objects include all obstacles and the current vehicle. Based on the forecast information at each time point, determine the single-point revenue of each object at each time point, and determine the individual cumulative revenue of each object based on the single-point revenue of each object at all time points. The cumulative revenue of obstacles is determined based on the cumulative revenue of each individual obstacle. Based on the cumulative revenue of obstacles and the cumulative revenue of the current vehicle, the revenue corresponding to the sampling longitudinal behavior scheme is determined.

12. An electronic device, characterized in that, The electronic device includes: One or more processors; Storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-10.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-10.

Images