PROBABILISTIC SYSTEM FOR MODELING THE DRIVING BEHAVIOR OF A VEHICLE

Abstract

A hybrid, probabilistic system for modeling the weather-dependent driving behavior of a vehicle comprises one or more controllers that execute instructions to determine a probabilistic longitudinal speed of the vehicle relative to a current weather condition. This speed is determined using a longitudinal driving model stored by the one or more controllers, based on a car-following model, a semantic rule system, and a speed and visibility model. The current weather condition specifies adverse weather conditions that affect the vehicle's driving behavior.The one or more controllers determine, through a probabilistic lateral driving model, one or more lane choices for the vehicle with respect to the current weather condition, based on a route plan of the vehicle and perception data specifying the vehicle's environment, and select a final lane choice from the one or more lane choices.

Description

INTRODUCTION

[0001] The present disclosure relates to a hybrid probabilistic system for modeling the driving behavior of a vehicle that takes adverse weather conditions into account.

[0002] Autonomous vehicles are becoming increasingly ubiquitous, but they still share the roadway with vehicles manually controlled by humans. Driver behavior models can be used to predict driving maneuvers, driver intent, and vehicle state. One type of driver behavior model is a microscopic driver behavior model, which includes multiple submodels that describe driver behavior. Two examples of submodels are a car-following model and a lane-changing model. The car-following model simulates how a driver of a constrained vehicle reacts to changes in the relative position and speed of a leading vehicle in a continuous traffic flow. The lane-changing model refers to a vehicle changing from its current lane to an adjacent lane.It should be noted that the car-following model and the lane-change model are considered in isolation, without determining the effects of one on the other. Considering the models in isolation can lead to unrealistic behavior, as speed profiles and lane decisions are selected separately.

[0003] Adverse weather conditions such as rain, fog, or snow can impair visibility of objects around the vehicle. Adverse weather conditions can also affect road surface friction, as well as the vehicle's maneuverability and stability. Therefore, a driver may behave differently when driving a vehicle in adverse weather than when driving a vehicle in clear or non-harsh weather. However, current approaches typically ignore the impact of adverse weather conditions on driver behavior and typically only consider non-harsh weather conditions.

[0004] Although driver behavior models for autonomous vehicles serve their purpose, there is a need for improved driver behavior models that take adverse weather conditions into account. SUMMARY

[0005] According to several aspects, a hybrid probabilistic system for modeling the driving behavior of a vehicle depending on the weather is disclosed. The hybrid probabilistic driving behavior modeling system includes one or more controllers executing instructions to determine, through a longitudinal driving model stored by the one or more controllers, a probabilistic longitudinal speed of the vehicle with respect to a current weather condition based on a car-following model, a semantic control system, and a speed and visibility model, wherein the current weather condition indicates an adverse weather condition that affects the driving conditions for the vehicle.A probabilistic lateral driving model of the one or more controllers determines one or more lane choices for the vehicle relative to the current weather condition based on a route plan of the vehicle and perception data indicative of an environment around the vehicle. The controllers select a final lane choice from the one or more lane choices, the final lane choice having a maximum probability of being selected at the probabilistic longitudinal speed compared to the remaining lane choices that are part of the one or more lane choices. The controllers determine a next state of the vehicle based on the final lane choice and the probabilistic longitudinal speed.

[0006] In another aspect, the probabilistic longitudinal speed is determined by the car following model determining a speed of the vehicle by calculating a first maximum speed and a second maximum speed and selecting a minimum value between the first maximum speed and the second maximum speed.

[0007] In another aspect, the first maximum speed represents when the vehicle is driven to achieve a target speed and the second maximum speed represents when the vehicle is driven to maintain a recommended distance from a vehicle ahead.

[0008] In one aspect, the probabilistic longitudinal velocity is determined by determining, by the semantic rule system, a constrained velocity that limits the velocity of the vehicle determined by the car-following model based on a plurality of high-level rules.

[0009] In another aspect, the plurality of high-level rules limit the speed of the vehicle determined by the car-following model based on one or more driving conditions generated by a particular adverse weather condition indicated by the current weather condition.

[0010] In one aspect, the probabilistic longitudinal speed is determined by the speed and visibility model determining the probabilistic longitudinal speed of the vehicle with respect to the current weather condition by limiting the restricted speed based on a visibility range.

[0011] In another aspect, the speed and visibility model represents a relationship between the restricted speed and the visibility of a vehicle driver for a particular location, where the restricted speed is negatively correlated with the inverse of the visibility.

[0012] In another aspect, the probabilistic lateral driving model classifies a current road segment that is part of a road plan as containing either a required lane change or a free lane change.

[0013] In one aspect, if it is determined that a lane change on the current road segment is required to reach the next road segment that is part of the route plan, the current road segment is classified as including the required lane change.

[0014] In another aspect, the current road segment is classified as containing a clear lane change if the vehicle is determined to be performing one or more of the following: changing lanes to increase the distance between a preceding or following vehicle and to achieve a higher speed.

[0015] In another aspect, the one or more controllers store a plurality of joint probability distribution models indicating a probability that a particular lane choice will be selected at a particular longitudinal speed of the vehicle under a particular adverse weather condition.

[0016] In one aspect, the joint probability distribution models are determined by defining a set of random variables and a set of weather conditions, where a first random variable represents the longitudinal behavior and a second random variable represents the lateral behavior.

[0017] In another aspect, the adverse weather condition is one of the following: rain, fog, snow and reduced light levels.

[0018] In another aspect, a method for determining a next state of a vehicle by a hybrid probabilistic system for modeling driving behavior for a vehicle is disclosed. The method includes determining a probabilistic longitudinal speed of the vehicle with respect to a current weather condition based on a car-following model, a semantic control system, and a speed and visibility model by a longitudinal driving model stored in one or more controllers, wherein the current weather condition indicates an adverse weather condition affecting driving conditions for the vehicle.The method includes determining, by a probabilistic lateral driving model stored in the one or more controllers, one or more lane choices for the vehicle with respect to the current weather condition based on a route plan of the vehicle and perception data related to the vehicle's surroundings. The method includes selecting, by the one or more controllers, a final lane choice from the one or more lane choices, wherein the final lane choice has a maximum probability of being selected at the probabilistic longitudinal velocity compared to the remaining lane choices that are part of the one or more lane choices. The method includes determining a next state of the vehicle based on the final lane choice and the probabilistic longitudinal velocity.

[0019] In another aspect, the method includes determining a speed of the vehicle by the car-following model by calculating a first maximum speed and a second maximum speed, and selecting a minimum value between the first maximum speed and the second maximum speed.

[0020] In another aspect, the method includes determining, by the semantic rule system, a restricted speed that limits the speed of the vehicle determined by the car-following model based on a plurality of high-level rules.

[0021] In one aspect, the method includes determining the probabilistic longitudinal speed of the vehicle by the speed and visibility model with respect to the current weather condition by limiting the restricted speed based on visibility.

[0022] In another aspect, the method includes classifying a current road segment that is part of a road plan as containing either a required lane change or a free lane change.

[0023] In another aspect, a hybrid probabilistic driving behavior modeling system for a vehicle comprises one or more controllers executing instructions to determine, through a longitudinal driving model stored by the one or more controllers, a probabilistic longitudinal speed of the vehicle with respect to a current weather condition based on an auto-following model, a semantic control system, and a speed and visibility model, wherein the current weather condition indicates an adverse weather condition affecting driving conditions for the vehicle, and wherein the probabilistic longitudinal speed is determined by: determining a speed of the vehicle using the auto-following model by calculating a first maximum speed and a second maximum speed,Selecting the minimum value from the first maximum speed and the second maximum speed, determining a restricted speed by the semantic control system that limits the speed of the vehicle determined by the car-following model based on a plurality of high-level rules, and determining, by the speed and visibility model of the one or more controllers, the probabilistic longitudinal speed of the vehicle with respect to the current weather condition by limiting the restricted speed based on visibility. A probabilistic lateral travel model stored by the one or more controllers determines one or more lane choices for the vehicle with respect to the current weather condition based on a vehicle route plan and perception data.that indicate the environment of the vehicle. The one or more controllers select a final lane selection from the one or more lane selections, the final lane selection having a maximum probability of being selected at the probabilistic longitudinal velocity compared to the remaining lane selections that are part of the one or more lane selections. The one or more controllers determine a next state of the vehicle based on the final lane selection and the probabilistic longitudinal velocity.

[0024] In another aspect, the probabilistic lateral driving model classifies a current road segment that is part of a road plan as containing either a required lane change or a free lane change.

[0025] Further areas of applicability will become apparent from the description provided herein. It should be understood that the description and specific examples are for purposes of illustration only and are not intended to limit the scope of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The drawings described herein are for illustrative purposes only and are not intended to limit the scope of the present disclosure in any way. Fig. 1 is a schematic representation of a vehicle incorporating the disclosed hybrid probabilistic driving behavior modeling system according to an exemplary embodiment with one or more controllers; Fig. 2 represents the software architecture of one or more of the Fig. 1 according to an exemplary embodiment; Fig. 3 shows a method for determining a probabilistic longitudinal speed of the vehicle in a current weather condition by a longitudinal driving model block that is part of the one or more Fig. 2, according to an exemplary embodiment; Fig. Figure 4 shows a method for determining one or more lane choices at the longitudinal speed by a lateral driving model block that is part of the one or more lane choices shown in Fig. 2, according to an exemplary embodiment; and Fig. 5 is a process flow diagram illustrating a method for modeling the behavior of the Fig. 1 shows a hybrid probabilistic system for modeling driving behavior according to an exemplary embodiment. DETAILED DESCRIPTION

[0027] The following description is merely illustrative in nature and is not intended to limit the present disclosure, its application, or uses.

[0028] In Fig. 1, an exemplary hybrid probabilistic driving behavior modeling system 10 for a vehicle 12 is shown. The vehicle 12 may be any type of vehicle, such as a sedan, a truck, a sport utility vehicle (SUV), a van, or a motor home. The vehicle 12 includes either an automated driving system (ADS) or an advanced driver assistance system (ADAS) that assists the driver with steering, braking, and / or acceleration. The hybrid probabilistic driving behavior modeling system 10 includes one or more controllers 20 in electronic communication with a plurality of perception sensors 22 configured to collect perception data relating to the environment of the vehicle 12. The Fig. The perception sensors 22 shown in Figure 1 include one or more cameras 30, an inertial measurement unit (IMU) 32, a global positioning system (GPS) 34, radar 36, and LiDAR 38, but other sensors may also be used.

[0029] As in Fig. 1, the one or more controllers 20 of the hybrid probabilistic driving behavior modeling system 10 include a longitudinal driving model block 40, a lateral driving model block 42, a semantic driving model block 44, a model correlation module 46, a weather calibration module 48, and a driver behavior module 50. The one or more controllers 20 receive an input state X of the vehicle 12, where the input state X represents the current state of the vehicle 12 observed by the plurality of perception sensors 22 in combination with other sensors (not shown) that are part of the vehicle 12, such as wheel speed sensors. The input state X of the vehicle 12 includes a current weather condition W, a current kinematic state of the vehicle 12, and a road condition R.

[0030] It should be noted that the current weather condition W indicates an adverse weather condition that affects the driving conditions for the vehicle 12. Examples of adverse weather conditions include rain, fog, snow, and reduced illumination at dusk or at night. In a specific embodiment, rain and / or fog can be categorized into different intensities, e.g., very light rain, light rain, moderate rain, and heavy rain. The adverse weather conditions affect the driving conditions by limiting the driver's visibility or by affecting the road surface. For example, the surface of roads can be affected by rain, causing the pavement to become slippery. The road condition R is an indicator of the friction coefficient µ coefficient of the road surface. The friction coefficient µ coefficientof the road indicates whether the road surface is covered with ice or snow, if it is wet due to precipitation, or if it is dry. The friction coefficient µ coefficient of the road surface is related to and used to determine the uncertainty of the vehicle position caused by adverse weather conditions. Specifically, the deviation σ of a normal distribution of data values ​​modeling vehicle behavior is inversely proportional to the friction coefficient µ coefficient .

[0031] The current kinematic state of the vehicle 12 indicates variables such as the current position x, y of the vehicle 12, the current longitudinal velocity vx, the current lateral velocity vy, the current acceleration a, and the current heading angle θ. The one or more controllers 20 determine a next state X' of the vehicle 12 based on the input state X, where the next state X' of the vehicle 12 indicates the next kinematic state of the vehicle 12 and includes variables such as the next position x', y' of the vehicle 12, the next longitudinal velocity vx', the next lateral velocity vy', the next acceleration a', and the next heading angle θ'.

[0032] Fig. 2 illustrates the software architecture for the one or more controllers 20. The longitudinal driving model block 40 executes a longitudinal driving model that determines a probabilistic longitudinal speed of the vehicle 12 with respect to the current weather condition W, denoted as (Is)|W, wherein the driver behavior module 50 receives the probabilistic longitudinal speed (Is)|W as input. The longitudinal driving model includes a car-following model 60, a semantic control system 62, and a speed and visibility model 64. The lateral driving model block 42 determines one or more lane choices (Ic)|W for the vehicle 12 with respect to the current weather condition W, wherein the driver behavior module 50 receives the lane change maneuver (Ic)|W as input. The one or more lane choices (Ic)|W for vehicle 12 include turning right, turning left, or going straight ahead.

[0033] As explained further below, the model correlation module 46 stores a plurality of joint probability distribution models used by the driver behavior module 50. The joint probability distribution models indicate the probability that a particular lane choice will be selected at a particular longitudinal speed of the vehicle 12 under a particular adverse weather condition and are determined based on training data. The driver behavior module 50 receives the probabilistic longitudinal speed (Is)|W and the one or more lane choices (Ic)|W as input and selects a final lane choice from the one or more lane choices (Ic)|W, where the final lane choice includes the maximum probability of being selected at the longitudinal speed (Ic)|W compared to the remaining lane choices that are part of the one or more lane choices (Ic)|W.The driver behavior module 50 then outputs the final lane selection Ic and the probabilistic longitudinal velocity Is, wherein the one or more controllers 20 translate the final lane selection Ic and the probabilistic longitudinal velocity Is into the next state X' of the vehicle 12.

[0034] The longitudinal driving model block 40 of the one or more controllers 20 will now be described. The longitudinal driving model block 40 executes a longitudinal driving model that determines the probabilistic longitudinal speed (Is)|W of the vehicle 12 in the current weather condition W based on the car-following model 60, the semantic control system 62, and the speed and visibility model 64. Fig. 3 is a process flow diagram illustrating an exemplary method 300 for determining a probabilistic longitudinal velocity (Is)|W of the vehicle 12 with respect to the current weather condition W. As shown in the Fig. 2 and Fig. 3, the method 300 may begin with block 302. In block 302, the longitudinal driving model block 40 receives the input state X of the vehicle 12, which indicates the current weather condition W, the current kinematic state of the vehicle 12, and the road condition R. The longitudinal driving model block 40 also receives the perception data received from the plurality of perception sensors 22 (in Fig. 1) as input. The method 300 may then proceed to decision block 304.

[0035] At decision block 304, the longitudinal driving model block 40 determines the presence of a lead vehicle situation in front of the vehicle 12. If it is determined that no lead vehicle is present, the method 300 proceeds to block 306; otherwise, the method 300 proceeds to block 308. At block 306, the longitudinal driving model block 40 creates a virtual lead vehicle model that simulates a lead vehicle in front of the vehicle 12. The virtual lead vehicle model includes a standard deviation of lane position (SDLP) value that indicates how adverse weather conditions affect the lane-keeping ability of the simulated lead vehicle. The virtual lead vehicle model also indicates the speed of the lead vehicle.The SDLP value is determined based on equation 1, which is:. SPLD=0.306(Weathercoeff)−0.168(Speed ​​Limit)*(Weathercoeff) where Weather coeffrepresents a weather coefficient that indicates the effect of an adverse weather condition on the lane position deviation, whereby different adverse weather conditions (e.g., rain, snow, or fog) affect lane positions in different ways. The longitudinal driving model block 40 determines the speed of the leading vehicle based on a speed selection model that also takes adverse weather conditions into account. The speed selection model reduces the speed of the leading vehicle depending on the type of adverse weather conditions. For example, one type of speed selection model reduces the speed of the leading vehicle by 5 to 40% in snow, 3 to 13% in light rain, and 3 to 16% in heavy rain, based on data from the Federal Highway Administration (FHWA), although other data sources may also be used.The method 300 may then proceed to block 308.

[0036] In block 308, block 40 of the longitudinal driving model predicts a speed v nf (t + T) of the vehicle 12 in the current weather condition W based on the car following model 60, where the current weather condition W is an unfavorable weather condition. Predicting the speed v nf (t + T) of the vehicle 12 under the current weather conditions W is described below. The method 300 may then proceed to block 310.

[0037] In block 310, the semantic control system 62 of the longitudinal driving model block 40 determines a restricted speed v nf_constrained , which is the speed v determined by the car-following model 60 nf (t + T) of the vehicle 12 based on a plurality of high-level rules, which are described below. The plurality of high-level rules limit the speed v nf(t + T) of the vehicle 12, determined by the car-following model 60, based on one or more driving conditions caused by the adverse weather conditions indicated by the current weather condition W. The method 300 may then proceed to block 312.

[0038] In block 312, the speed and visibility model 64 of the longitudinal driving model block 40 determines the probabilistic longitudinal speed (Is)|W of the vehicle 12 with respect to the current weather condition W by limiting the restricted speed v nf_constrained based on visibility. In particular, the speed and visibility model 64 establishes a relationship between the restricted speed v nf_constrained and the visibility of a driver of the vehicle 12 for a particular location, wherein the restricted speed v nf_constrainedis negatively correlated with the inverse of the visibility range. The speed and visibility model 64 is described in more detail below. Once the probabilistic longitudinal speed (Is)|W of the vehicle 12 with respect to the current weather condition W is determined, the method 300 may be terminated.

[0039] The following describes the car-following model 60. The car-following model 60 is determined based on training data that models the adverse weather conditions. The training data may include real data or simulated data. In an exemplary embodiment, the car-following model is the Gipps car-following model, and the training data is the Wyoming Second Strategic Highway Research Program (SHRP2) Naturalistic Driving Study (NDS) dataset; however, other models and datasets may be used.

[0040] The speed v nf(t + T) is measured in a time interval (t + T), where t is the current time and T is an apparent reaction time that is constant for all vehicles in the environment (i.e., vehicle 12, a lead). The car-following model 60 determines the speed v nf (t + T) of the vehicle 12 by setting a first maximum speed vnfa and a second top speed vnfb calculated and then the minimum value of the first maximum speed vnfa and second top speed vnfb which is expressed in equation 2 as follows: vnf(t+T)=min{vnfa(t+T),vnfb(t+T)} where the first maximum speed vnfa when the vehicle 12 is driven to reach a target speed, and the second maximum speed vnfb when the vehicle is driven 12 to maintain a recommended distance from a vehicle in front. The first maximum speed vnfa is solved based on equation 3, and the second maximum speed vnfb is solved based on equation 4: vnfa(t+T)=vnf(t)+2.5×anfmax×T×(1−vnf(t)vnfdesired)×0.025+vnf(t)vnfdesired vnfb(t+T)=dnfmax×T+(dnfmax×T)2−dnfmax×[2×{xnl(t)−xnf(t)−(s)}−vnf(t)×T−vnl(t)2d^nlmax] where f is a subscript indicating a following vehicle (which is vehicle 12), I is a subscript indicating a preceding vehicle, x n indicates a location of either the preceding or following vehicle at time t, v n represents the speed of either the preceding or the following vehicle at time t, anmax specifies a maximum desired acceleration for vehicle n, desired specifies a maximum desired speed for a vehicle n (where vehicle n represents either the preceding or following vehicle), dnmax indicates a maximum desired deceleration for vehicle n based on the strongest braking applied by a driver of the following vehicle, d^nlmax represents an estimate of a maximum deceleration desired by leading to vehicle n and represents an estimate of the strongest braking capacity of the preceding vehicle, L represents the length of the preceding vehicle, S represents the distance between the vehicles at a stop or a minimum following distance at a stop, and s = (L l+ S) and is a sum of the size of the preceding vehicle and the minimum following distance at a stopping point. The values ​​of the Gipps parameters expressed in equations 3 and 4 (e.g. dnfmax,T,S) are stored by the weather calibration module 48 of the one or more controllers 20. For example, the Gipps parameter value for the vehicle separation s is 2.9 meters in non-harsh weather conditions, 4.4 meters in fog, 3.1 meters in very light rain, and 3.5 meters in snow.

[0041] The semantic rule system 62 of the longitudinal driving model block 40 will now be described. It can be seen that the semantic driving model block 44 provides the plurality of high-level rules for the semantic rule system 62 of the longitudinal driving model block 40. The plurality of high-level rules of the semantic rule system 62 limit the speed v nf(t + T) of the vehicle 12, which is determined by the car following model 60, based on one or more driving conditions generated by a particular adverse weather condition indicated by the current weather condition W. In one embodiment, the driving conditions include one or more of the following: road surface condition, visibility, road geometry, and construction areas. The road surface condition indicates when the coefficient of friction is altered by precipitation on the roadway. It should be noted that ice or snow may require the vehicle 12 to reduce its speed, and therefore, the plurality of high-level rules limit the speed v nf(t + T) of vehicle 12 accordingly. For example, if the road surface is dry, then the plurality of high-level rules specify that vehicle 12 should travel at its current speed with a probability of 0.95. However, if the road surface is icy and the current speed of vehicle 12 is below a threshold, vehicle 12 must reduce its speed by 80% with a probability of 0.9. Road geometry indicates curves in the road and when vehicle 12 is approaching an entry ramp or an exit ramp. The plurality of high-level rules specify that when entering a work zone and visibility is normal, the speed of vehicle 12 is reduced by 40%, but when visibility is poor, the speed of vehicle 12 is reduced by 70%.

[0042] The speed and visibility model 64 of the longitudinal driving model block 40 of the one or more controllers 20 will now be described. The speed and visibility model 64 determines the probabilistic longitudinal speed (Is)|W of the vehicle 12 with respect to the current weather condition W by determining the restricted speed v nf_constrained limited based on visibility. As already mentioned, the speed and visibility model 64 establishes a relationship between the restricted speed v nf_constrained and the visibility of the driver of the vehicle 12 for the particular location, with the restricted speed v nf_constrainedis negatively correlated with the inverse of visibility. In other words, the speed and visibility model 64 takes into account how drivers react and reduce the speed of the vehicle 12 when visibility decreases. The particular location refers to a specific road segment. For example, the particular location may refer to a section of a highway between a first and a second mile marker. It should be noted that a particular location contains various location-specific variables that cannot be directly extrapolated to another location; however, the speed and visibility model 64 may be used for a particular geographic region to set a speed limit based on visibility.An example of the speed and visibility model 64, representing a highway section between two kilometer markers, is expressed in Equation 5 as follows: . S=64.6−4204Vis+(1.13×DayNight)+6.07×SB6−2.67×SB7 where S stands for the average speed per 5 minutes, Vis for the visibility, DayNight for a day or night dummy variable, where 1 stands for day and 0 for night, SB6 for a variable where 1 stands for a first kilometer marker on the highway, SB7 for another variable where 1 stands for a second kilometer marker on the highway. In the event that the probabilistic longitudinal speed (Is)|W of the vehicle 12 with respect to the current weather condition W is less than the average speed S, the probabilistic longitudinal speed (Is)|W of the vehicle 12 is then set to the average speed S (S = Is), otherwise the probabilistic longitudinal speed (Is)|W is limited to the average speed S.

[0043] The lateral driving model block 42 of the one or more controllers 20 is described below. The lateral driving model block 42 determines one or more lane choices (Ic)|W for the vehicle 12 with respect to the current weather condition W based on a route plan of the vehicle 12 and the perception data indicating an environment of the vehicle 12 from the plurality of perception sensors 22 (in Fig. 1) based on a probabilistic lateral driving model. The probabilistic lateral driving model is a lane change model that determines lane choices when a vehicle changes from a current lane to an adjacent lane. The route plan for vehicle 12 includes a plurality of road segments, with the one or more lane choices (Ic)|W for vehicle 12 being determined for each road segment that is part of the route plan.

[0044] A method 400 for determining the one or more lane choices (Ic)|W for the vehicle 12 is described in Fig. 4. As shown in the Fig. 2 and Fig. 4, the method 400 begins with decision block 402. In block 402, the lateral travel model block 42 of the one or more controllers 20 classifies a current road segment that is part of a road plan as containing either a required lane change or a clear lane change. The current road segment is classified as containing the required lane change if it is determined that a lane change on the current road segment is required to reach the next road segment that is part of the route plan. The current road segment is classified as containing a clear lane change if it is determined that the vehicle 12 is performing one or more of the following actions: changing lanes to increase the distance between a leading or following vehicle, and to achieve a higher speed.The required lane change and the free lane change are described in more detail below. In response to classifying the current road segment as containing a free lane change, method 400 may proceed to decision block 404.

[0045] In decision block 404, the lateral driving model block 42 determines whether the conditions for performing the free lane change are met. The conditions for performing the free lane change include, in particular, determining whether a total stopping distance between the vehicle 12 and a following vehicle allows the vehicle 12 to come to a complete stop and whether there is sufficient separation in the direction of travel between the vehicle 12 and a following vehicle. The sufficient separation depends on the speed of the vehicle 12 and the speed of the following vehicle. The total stopping distance and the sufficient separation are calculated based on weather-dependent parameters stored in the weather calibration module 48 of the one or more controllers 20.In response to determining that the conditions for performing the free lane change are not met, method 400 may terminate, and a next road segment is considered. In response to determining that the conditions for performing the free lane change are met, method 400 proceeds to decision block 406.

[0046] In decision block 406, the lateral driving model block 42 of the one or more controllers 20 determines whether the values ​​for a maximum deceleration and an accepted deceleration corresponding to the necessary lane change are met. The maximum deceleration refers to an upper limit of the deceleration rate for the vehicle 12 and for a following vehicle, and the accepted deceleration refers to a lower limit of the deceleration for the vehicle 12 and for the following vehicle. The values ​​for the maximum deceleration and an accepted deceleration are adjusted to account for adverse weather conditions and in the weather calibration module 48 (see Fig. 2) is stored. In response to determining that the maximum deceleration and the accepted deceleration corresponding to the required lane change are not met, method 400 is terminated and the next road segment is considered. However, if it is determined that the maximum deceleration and the accepted deceleration corresponding to the required lane change are not met, method 400 proceeds to block 408.

[0047] In block 408, the lateral driving model block 42 of the one or more controllers 20 determines one or more potential lane choices based on the route plan and the perception data received from the plurality of Fig. 1. The method 400 may then proceed to decision block 410.

[0048] In decision block 410, the lateral travel model block 42 of the one or more controllers 20 evaluates each potential lane selection based on a first distance threshold measured between the vehicle 12 and a following vehicle in a target lane, a second distance threshold measured between the vehicle 12 and a trailing vehicle in the target lane, and a third distance threshold measured between the vehicle 12 and a following vehicle in the current lane. The first, second, and third distance thresholds are selected to allow sufficient travel space for the vehicle 12 during the lane change.

[0049] In response to determining that a specific lane choice does not satisfy the first distance threshold, the second distance threshold, or the third distance threshold, method 400 terminates, and the specific lane choice is removed from consideration. However, if a specific lane choice is determined to satisfy the first distance threshold, the second distance threshold, and the third distance threshold, the particular lane choice is added to the one or more lane choices (Ic)|W for vehicle 12. Once all specific lane choices have been considered, the lateral driving model block 42 of the one or more controllers 20 transmits the one or more lane choices (Ic)|W for vehicle 12 to the driver behavior module 50, and method 400 terminates.

[0050] In one non-limiting embodiment, the lane change model is simulated by a traffic simulation program such as VISSIM. The required lane change is defined by three parameters: the maximum deceleration, the accepted deceleration, and a reduction rate of acceleration. As previously mentioned, the maximum deceleration refers to an upper limit of the deceleration rate for the vehicle 12 and a following vehicle, and the accepted deceleration refers to a lower limit of the deceleration for the vehicle 12 and the following vehicle. The reduced rate of acceleration decreases the maximum deceleration linearly with increasing distance from an emergency stop position. Some examples of a required lane change include an entry ramp and an exit ramp.

[0051] Free lane changing is defined by three parameters: minimum follower distance, follower distance reduction factor, and maximum cooperative braking deceleration. The minimum follower distance is a minimum distance between vehicle 12 and a following or leading vehicle required after a lane change. During a lane change, a value of the minimum follower distance is reduced by the follower distance reduction factor. The maximum cooperative braking deceleration refers to the cooperative deceleration of the following vehicle that allows vehicle 12 to change lanes. Some examples of free lane changing include overtaking or passing a slow vehicle in the current lane, returning to a driver-preferred lane, and entering a new lane.The parameters for both the required lane change and the free lane change are adjusted taking into account the weather conditions and stored in the weather calibration module 48 (see . Fig. 2) stored. For example, the maximum deceleration of the vehicle is 12 -11.38 meters / second 2 (m / s 2 ) in non-rough weather, -8.32 m / s 2 in rain, -17.84 m / s 2 in snow and -17.43 m / s 2 in fog.

[0052] The driver behavior module 50 of the one or more controllers 20 is described below. As previously mentioned, the model correlation module 46 stores a plurality of joint probability distribution models indicating the probability that a particular lane choice will be selected at a particular longitudinal speed of the vehicle 12 under a particular adverse weather condition. The joint probability distribution models are determined by defining a set of random variables {Lo, La} and a set of weather conditions as {W}, where the first random variable Lo represents the longitudinal behavior and a second random variable La represents the lateral behavior. A joint probability distribution over the set of random variables {Lo, La}, conditioned on the set of weather conditions {W}, is defined as P(Lo, La | W).During training, a set of longitudinal probability distributions P(Lo|W) and a set of lateral probability distributions P(La|Lo,W) are defined. The joint probability distribution P(S, Lo, La | W) also takes into account the semantic behavior S, where the joint probability distribution P(S, Lo, La | W) is expressed in Equation 6 as follows: P(S,Lo,La|W)=P(S|Lo,La W)*P(S,La|Lo, W)*P(Lo|W)

[0053] The driver behavior module 50 selects a lateral behavior la that maximizes the random variable Lo for the longitudinal behavior and the random variable La for the lateral behavior based on the current weather conditions W, which is expressed in equation 7 as follows: la=argmaxla∈La log p(La,Lo|W)=argmaxla∈La log p(La|Lo,W)∗p(Lo|W) where p (La, Lo|W) represents a conditional probability of the random variable of the lateral behavior La given the random variable Lo of the longitudinal behavior and the current weather condition W, and p (Lo|W) represents a conditional probability of the random variable Lo of the longitudinal behavior given the current weather condition W. The driver behavior module 50 selects, from the one or more lane choices (Ic)|W received from the lateral driving model block 42, the final lane choice Ic that results in the highest probability compared to the remaining lane choices that are part of the one or more lane choices (Ic)|W and is expressed in Equation 8 as: lc=argmaxla∈La log p(Lc|Ls,W)∗p(Ls|W)

[0054] The driver behavior module 50 receives the probabilistic longitudinal velocity (Is)|W and the one or more lane choices (Ic)|W as input and selects a final lane choice from the one or more lane choices (Ic)|W based on the joint probability distributions, where the final lane choice contains the highest probability compared to the remaining lane choices that are part of the one or more lane choices (Ic)|W. The driver behavior module 50 then outputs the final lane choice Ic and the probabilistic longitudinal velocity Is, whereupon the one or more controllers 20 determine the next state X' of the vehicle 12 based on the final lane choice Ic and the probabilistic longitudinal velocity Is.

[0055] Fig. 5 shows a process flow diagram illustrating an exemplary method 500 for modeling the behavior of the Fig. 2 shows the hybrid probabilistic system 10 for modeling driving behavior. With reference to Fig. 1, Fig. 2 and Fig. 5, the method 500 may begin with block 502. In block 502, the one or more controllers 20 receive a road segment that is part of the route plan. The method 500 may then proceed to decision block 504.

[0056] If the road segment represents an intersection, the method proceeds to decision block 506 at decision block 504. Otherwise, the method 500 proceeds to decision block 510.

[0057] At decision block 506, the one or more controllers 20 continue to monitor the perception data collected by the plurality of perception sensors 22 to determine whether the vehicle 12 is permitted to perform a maneuver at the intersection. For example, the one or more controllers 20 may wait for a traffic signal indicating that the vehicle 12 is permitted to perform a maneuver. Once the one or more controllers 20 determine that the vehicle 12 is permitted to turn, the method 500 proceeds to block 508.

[0058] At block 508, vehicle 12 executes a maneuver at the intersection, selecting one of the following maneuvers: left turn, right turn, or straight ahead. The method 500 may then proceed to decision block 510 to evaluate the next road segment.

[0059] At decision block 510, the one or more controllers 20 determine whether the vehicle 12 is capable of traveling. If the vehicle 12 is capable of traveling, the method 500 continues to decision block 512. At decision block 512, the one or more controllers 20 determine whether the vehicle is traveling. If the vehicle 12 is not traveling, the one or more controllers 20 select a braking longitudinal model, and the method exits. The braking longitudinal model uses the same longitudinal traveling model as the longitudinal traveling model block 40 during normal operation of the vehicle 12 to determine a reduced target speed required when the vehicle 12 is braking and reducing speed. However, if the vehicle is not traveling, the method 500 returns to decision block 510.If the one or more controllers 20 determine in decision block 510 that the vehicle 12 can travel, the method 500 continues to decision block 512.

[0060] At decision block 512, the one or more controllers 20 determine the number of lanes on the roadway. In response to determining that only one lane is present on the roadway, the one or more controllers 20 select the lateral driving model block 42 to determine the one or more lane choices (Ic)|W for the vehicle 12. The method 500 may then exit. In response to determining that more than one lane is present on the roadway, the one or more controllers 20 select the longitudinal driving model block 40 to determine the probabilistic longitudinal velocity (Is)|W. The method 500 may then exit.

[0061] With general reference to the figures, the disclosed hybrid probabilistic driving behavior modeling system offers various technical effects and advantages. The hybrid probabilistic driving behavior modeling system integrates a car-following model, a lane-changing model, a driver's semantic behavior, and speed and visibility models to determine the vehicle's next state under adverse weather conditions, thereby solving the problem of inconsistent behaviors being determined by the models when considered independently. The hybrid probabilistic driving behavior modeling system maximizes the number of lane choices available to a vehicle on a given road segment of a route plan at a given longitudinal speed under the current weather conditions.

[0062] The controllers may refer to or be part of an electronic circuit, a combinational logic circuit, a field-programmable gate array (FPGA), a (shared, dedicated, or clustered) processor executing code, or a combination of some or all of the above, such as in a system-on-chip. Furthermore, the controllers may be microprocessor-based, such as a computer having at least one processor, memory (RAM and / or ROM), and associated input and output buses. The processor may operate under the control of an operating system residing in memory. The operating system may manage computer resources so that computer program code embodied as one or more computer software applications, e.g., an application residing in memory, can have instructions executed by the processor.In an alternative embodiment, the processor may execute the application directly; in this case, the operating system may be omitted.

[0063] The description of the present disclosure is merely exemplary, and variations that do not depart from the gist of the present disclosure are intended to be included within the scope of the present disclosure. Such variations are not to be regarded as a departure from the spirit and scope of the present disclosure.

Claims

[1] Hybrid probabilistic system for modeling the weather-dependent driving behavior of a vehicle, the hybrid probabilistic system for modeling the driving behavior comprising: one or more controllers that execute instructions to: Determining a probabilistic longitudinal speed of the vehicle with respect to a current weather condition based on a car-following model, a semantic control system, and a speed and visibility model by a longitudinal driving model stored by the one or more controllers, wherein the current weather condition indicates an adverse weather condition affecting driving conditions for the vehicle; Determining one or more lane choices for the vehicle through a probabilistic lateral driving model by the one or more controllers with respect to the current weather condition based on a route plan of the vehicle and perception data indicative of the vehicle's surroundings; Selecting a final lane choice from the one or more lane choices, the final lane choice having a maximum probability of being selected at the probabilistic longitudinal velocity compared to the remaining lane choices that are part of the one or more lane choices; and Determining a next state of the vehicle based on the final lane selection and the probabilistic longitudinal velocity. [2] A hybrid probabilistic system for modeling driving behavior according to claim 1, wherein the probabilistic longitudinal speed is determined by: Determining a speed of the vehicle using the car-following model by calculating a first maximum speed and a second maximum speed; and Select the minimum value from the first maximum speed and the second maximum speed. [3] A hybrid probabilistic driving behavior modeling system according to claim 2, wherein the first maximum speed indicates when the vehicle is driven to achieve a target speed and the second maximum speed indicates when the vehicle is driven to maintain a recommended distance from a preceding vehicle. [4] A hybrid probabilistic system for modeling driving behavior according to claim 2, wherein the probabilistic longitudinal speed is determined by: Determining a restricted speed by the semantic rule system that limits the speed of the vehicle determined by the car-following model based on a plurality of high-level rules. [5] The hybrid probabilistic driving behavior modeling system of claim 4, wherein the plurality of high-level rules limit the speed of the vehicle determined by the car-following model based on one or more driving conditions generated by a specific adverse weather condition indicated by the current weather condition. [6] A hybrid probabilistic system for modeling driving behavior according to claim 4, wherein the probabilistic longitudinal speed is determined by: Determining the probabilistic longitudinal speed of the vehicle by the speed and visibility model with respect to the current weather condition by limiting the restricted speed based on a visibility range. [7] A hybrid probabilistic driving behavior modeling system according to claim 5, wherein the speed and visibility model represents a relationship between the restricted speed and a visibility distance of a driver of the vehicle for a particular location, the restricted speed being negatively correlated with an inverse of the visibility distance. [8] A hybrid probabilistic driving behavior modeling system according to claim 1, wherein the probabilistic lateral driving model classifies a current road segment that is part of a road plan as containing either a required lane change or a free lane change. [9] A hybrid probabilistic driving behavior modeling system according to claim 8, wherein the current road segment is classified as containing the necessary lane change if it is determined that a lane change on the current road segment is required to reach the next road segment that is part of the route plan. [10] A hybrid probabilistic driving behavior modeling system according to claim 8, wherein the current road segment is classified as containing a free lane change if the vehicle is determined to perform one or more of the following actions: changing lanes to increase the distance between a preceding or following vehicle and to achieve a higher speed.

Images