CN113200054A

CN113200054A - Path planning method and system for automatic driving takeover

Info

Publication number: CN113200054A
Application number: CN202110684771.7A
Authority: CN
Inventors: 季学武; 储元
Original assignee: Tsinghua University
Current assignee: Tsinghua University
Priority date: 2021-06-21
Filing date: 2021-06-21
Publication date: 2021-08-03
Anticipated expiration: 2041-06-21
Also published as: CN113200054B

Abstract

The invention relates to a path planning method and a system for automatic driving takeover, wherein the method comprises the following steps: acquiring the acceptance of a user for taking over the vehicle; determining operation interference acceptance and control interference acceptance according to the takeover acceptance; determining an operation interference risk and a control risk according to the operation interference acceptance and the control interference acceptance; determining a takeover risk according to the operation interference risk and the control risk; acquiring the risk of the surrounding environment; determining a unified risk of the autonomous vehicle according to the takeover risk and the ambient risk; according to the unified risk, determining a set of points with the lowest expected risk strength in multiple risk evaluations of the automatic driving vehicle, and recording as an optimal safe path; and controlling the corner of the front wheel of the vehicle according to the optimal safety path to ensure that the vehicle road is attached to the optimal safety path. The method can combine the driver takeover risk and the road environment risk into a unified risk, and plans the safest path in the unified risk.

Description

Path planning method and system for automatic driving takeover

Technical Field

The invention relates to the field of automatic driving, in particular to a path planning method and system for automatic driving takeover.

Background

Motion planning is a core problem in the field of robotics (in particular unmanned aerial vehicles, autonomous vehicles) and can be divided into path planning and trajectory planning. Path planning generally refers to planning location points in space and forming a path from the location points independent of time constraints. Trajectory planning may take into account time constraints, such as planning a robot to pass a certain point at a certain time, so that a trajectory is a path with respect to time.

The path planning method mainly includes the following types: graph search based planning, sampling based planning, function curve based planning, optimization based planning, physical model based planning, and other planning methods.

The planning method based on graph search can simplify the complex structure on the space into a geometric figure in Euclidean space and can also simplify the environment into a topological map formed by topological nodes. The robot can plan the motion path of the robot by accessing the graph nodes. A common graph search algorithm is the a-x algorithm, which can obtain the node route with the lowest preset cost in the network. This type of algorithm is slow in computation speed for large number of nodes or large area and is not suitable for real-time path planning. Sampling-based planning methods are typically able to randomly sample a target area and generate specific curves on the nodes. A common sampling algorithm is a random search tree, which can randomly select nodes in a target area, and generate tree branches through the nodes, based on which new nodes can be further generated. Through the path generation mode, the robot can search the target area and finally can find a communicated path between the starting point and the target point. This type of algorithm can accommodate complex unstructured regions, however the paths it generates are not continuous. A function curve based planning method is able to generate a path by a specific curve function. A common algorithm is a fifth-order polynomial, when a path is generated by using the fifth-order polynomial, constraints are required to be designed for a starting point and an end point, lateral and longitudinal speeds and accelerations in a relative coordinate system or an absolute coordinate system of the robot are generally used as constraint conditions, and unknown parameters of the fifth-order polynomial are fitted through the constraint conditions. This type of algorithm is computationally inexpensive, but is not suitable for non-structural regions. The optimization-based planning method needs to model for actual problems, design a loss function for a target index, and solve the loss function so as to obtain a local or global optimal solution for the loss function, thereby enabling a path to meet the requirements of the design index. The method is wide in application range, but the modeling difficulty is high, and the situation that no optimal solution or no analytic solution exists in the optimization problem solution of the complex model. The planning method based on the physical model understands the motion of the robot as the change of the state of the robot after the robot is subjected to the action of external force. The artificial potential field method is a classical method in this type of planning. The artificial potential field method mainly aims at establishing a potential energy field for objects in the environment, and repulsion or attraction is formed among different objects. Thus, in this method a gravitational potential energy field is formed between the virtual object on the default target object or target point and the robot itself, which can provide an attractive interaction, and a repulsive potential energy field is formed between the robot and the obstacle, which can provide a repulsive interaction. On the other hand, the farther the robot position point and the target point are, the larger the potential energy of the attraction potential energy field is, and the closer the robot is to an obstacle within the effective range of the obstacle potential energy field, the larger the potential energy of the repulsion potential energy field is. The total potential energy field is the superposition of the repulsive and attractive potential energy fields. Meanwhile, when the method is used, the positive numerical value of the total potential energy field can be defined as a positive real number, the potential energy value of the repulsive potential energy field is a positive real number, and the potential energy value of the attractive potential field is a negative real number. Further, by adjusting parameters in the potential field expression, the potential field value of the position point of the robot can be the maximum value of the whole potential field, and the target point potential field is the minimum value point. Thus, the robot may generate a potential field force along a negative gradient direction of the total potential energy field function and thereby guide the machine to move, thereby forming a path. In addition, since the trajectory planning methods are numerous, they are not enumerated here.

Compared with path planning, trajectory planning describes the motion behavior of a robot from a time sequence perspective, and includes a specific description of how the robot moves from one space point to another space point. For example, the robot is required to pass a certain waypoint at a certain point in time. Therefore, trajectory planning is often tightly coupled with motion control. This means that the kinematic behavior constraints of the robot itself need to be considered in the design work of the trajectory planning, for example, the kinematics and dynamics constraints of the vehicle need to be considered in the trajectory planning of the unmanned vehicle. The actual motion state of the vehicle is controlled by comparing the current position information of the vehicle with preset (including time constraint) information. Therefore, the trajectory planning can also be based on the result of the path planning, that is, the vehicle is required to travel along the planned path, and the vehicle is controlled to continuously meet the time constraint, so as to realize the trajectory planning task of the whole vehicle.

Takeover behavior is an obvious feature of an autonomous driving scenario. Taking the SAEJ3016 standard as an example, the smart car may face situations such as system failure of a non-autonomous driving vehicle, system failure of an autonomous driving vehicle, and exceeding of a related operation domain at the level of L3. In all three cases, the autopilot system will not continue to operate effectively. At this time, the driver needs to re-touch the steering system and take over the operation. In an automated driving scenario at level L4, although the automated driving system does not require the driver to supervise and correct system failures, the smart car may still be exposed to system out of design operating domain. Therefore, in stage L4, there is still a scenario where the driver re-touches the steering system and takes over. It should be noted that for the stage L5, i.e. fully autonomous driving, the vehicle does not require the user to take over, but this does not mean that the user is not able to take over the vehicle. In other words, there is a possibility that automobile manufacturers continue to provide users with human-computer interaction interfaces for taking over vehicles, aiming at meeting the requirements and fun of users for manual driving and thus improving the competitiveness of their products in the market. Therefore, even in the full driving stage, there is still a possibility that the in-vehicle user takes over the vehicle.

On the other hand, in the future, the mechanical structure and the appearance of the control device for controlling the vehicle motion may be changed, that is, the control device is completely different from the current devices such as a steering wheel, an accelerator pedal and the like, and the situation that the control device does not exist may also occur. However, this does not completely negate the possibility that a tactile interactive motion control operator will remain in a human-computer interaction system in the future (this patent only discusses a tactile interactive steering gear). Furthermore, the steering gear is not always stationary during take-over, whether it is a conventional mechanically coupled steering system or a more advanced steer-by-wire system. Even if the steering control device is in a stationary state during the take-over process, this does not mean that the driver is in a relatively safe state in the face of the driving state and the environment at the take-over, after sensing, evaluating and manipulating actions.

Therefore, the risk of takeover is always present on a strict basis. In the course of a driver as a user interacting with a vehicle system (as an automated system), perfect cooperative cooperation between the two is difficult to achieve, and even take-over failure (vehicle rushing out of lane line) occurs, with consequent catastrophic consequences (such as impact on road infrastructure, entry into a reverse lane or rushing into a co-directional other lane). Such a decline in driving performance may be caused by the driver's situational awareness of the current driving state being incorrect or inappropriate. Inappropriate interactive feedback can create driver cognitive bias, further causing erroneous decisions and operations. Therefore, the driving hazard caused by improper information interaction to the management process cannot be ignored. Obviously, such inappropriate information interaction involves tactile interaction. During the take-over process, the tactile information transmitted to the driver by the vehicle through the steering control device influences the perception and judgment of the driver on the vehicle motion control. Based on human theory, improper tactile feedback may induce a collapse of driver trust in vehicle systems, which may be spread to the driver's trust in other well-functioning vehicle subsystems, further inducing a collapse of driver trust in more other subsystems. Obviously, such a trust breakdown may affect the driving performance of the driver, increasing the driving risk potential. Secondly, when the steering wheel is in a motion state in the process of taking over, the rotating steering wheel can pull/push the upper limbs of the driver and force the upper limbs to form a certain degree of uncontrollable stretch reflex, even become a strong stimulus, and then the driver emergency stress response is triggered. Therefore, the unreasonable tactile feedback not only increases the difficulty of taking over by the driver, but also has a possibility of inducing a driving accident.

At present, the research aiming at the intelligent vehicle taking-over process is not sufficient. Most research is limited to a driving scene containing a take-over request (which can be regarded as a passive take-over type), and mainly aims at the influence of factors such as take-over reaction time of a driver, take-over request signal types (including visual signals, auditory signals and touch signals), non-driving subtasks (reading, playing and the like), driver crowd difference (adolescence, the elderly, males and females) and the like on the take-over reaction time of the driver. The current relevant research does not model the risk in the take-over process, nor does it work on intelligent vehicle (autonomous) path planning based on this.

Therefore, the invention provides an automatic driving path planning method considering the driver takeover risk, which integrates the driver takeover risk (as a self-behavior risk) and the road environment risk (as an external environment risk), evaluates the unified risk comprising the self risk and the external risk, plans the safest path in the unified risk, and achieves the purpose of improving the driving safety including the takeover process.

Disclosure of Invention

The invention aims to provide a path planning method and system for automatic driving takeover, which integrates the driver takeover risk and the road environment risk into a unified risk and plans the safest path in the unified risk.

In order to achieve the purpose, the invention provides the following scheme:

a path planning method for automatic driving take-over, the path planning method comprising:

acquiring the acceptance of a user for taking over the vehicle;

determining operation interference acceptance and control interference acceptance according to the takeover acceptance;

determining an operation interference risk and a control risk according to the operation interference acceptance and the control interference acceptance;

determining a takeover risk according to the operation interference risk and the control risk;

acquiring the risk of the surrounding environment;

determining a unified risk of the autonomous vehicle according to the takeover risk and the ambient environment risk;

according to the unified risk, determining a set of points with the lowest expected risk strength in multiple risk evaluations of the automatic driving vehicle, and recording as an optimal safe path;

and controlling the corner of the front wheel of the vehicle according to the optimal safety path to ensure that the vehicle road is attached to the optimal safety path.

Optionally, the expression of the takeover risk is as follows:

wherein r is_disRepresenting the risk of operational disturbances, r_conIndicating control risk, D_LateralRepresents the deviation of the vehicle from the center of the lane at the present time, L_lanIndicating the lane width.

Optionally, the ambient risk includes two aspects, an external environment other than the autonomous vehicle and an autonomous vehicle environment other than the driver.

Optionally, the ambient risk includes six levels, the risk elements of the first level are traffic behaviors of the surrounding vehicles and pedestrians, the risk elements of the second level are vehicle subsystems, the vehicle subsystems are systems which do not generate direct interaction with users in the vehicles, the risk elements of the third level are traffic regulations, the risk elements of the fourth level are road facilities, the risk elements of the fifth level are road conditions, and the risk elements of the sixth level are climate conditions.

Optionally, the expression of the ambient risk element is as follows:

r_out＝∑K_lir_li

wherein r is_liIs the risk intensity, K, of all risk elements of layer i_liIs the weighting factor, r, of the layer_li＝∑re_ijWherein re_ijIs the risk strength of the jth risk element of the ith layer.

Optionally, the expression of the unified risk is as follows:

wherein the content of the first and second substances,

and

weight factors, r, for the takeover risk and the ambient risk, respectively_outFor the risk of the surrounding environment, r_inTo take over the risk.

Optionally, determining a set of points with the lowest expected risk strength in multiple risk evaluations of the autonomous vehicle according to the unified risk, and recording as an optimal safe path specifically includes:

set family P for determining planned points of bicycle path at any time_c，P_c＝{P_c1,P_c2,…,P_cn}; wherein the set P_c1The point in (1) is the current self-parking position X_cIs an origin and takes R_mThe boundary of the paradigm ball which is the distance is on the intersecting curve of the road surface;

according to the set P_c1Point determination set P of_clc(ii) a Wherein, P_c1c＝{X_P|||X-X_P| ═ vc }, and { road sueface points }, wherein v is the absolute value of the speed of the bicycle; c is a moving range coefficient for adjusting the observation P_clcSelecting range of medium elements;

in the first set P_clcSelecting a set with lower risk intensity (a suboptimal solution set) as P_c1，P_c1＝{argmin{rp_c1cAnd + -gamma, wherein gamma determines the value range of the suboptimal solution set.

With said set P_c1As a new self-parking position X_cRepeating the above steps to obtain a set P_c2Up to the Collection family P_cAll the sets in the vehicle are updated to obtain the final set family P of the planning points of the vehicle path_c＝{P_c1,P_c2,…,P_cn}；

Aiming at the problem that planned path points are unreachable, arbitrary values delta and N are given_PSo that

Retention of P_cnDiscarding P as a final planning result_cn+1Up to P_cn+NpThe planning result of (2);

determining the risk intensity corresponding to each point in the suboptimal solution set in the risk assessment result of the c1 th time

Wherein t is the total of t elements in the c1 th risk assessment result;

for a total of cn risk assessments, based on conditional probabilities in probability theory and expected numerical characteristics of discrete random variables, the mathematical expectation expression is as follows:

the non-reentrant problem is set to EventB, so the optimal path is

Optionally, the method further includes, after the step of controlling a front wheel rotation angle of the vehicle according to the optimal safe path to fit the vehicle path to the optimal safe path:

determining a path error;

obtaining a front wheel corner variation based on the path error and the vehicle front wheel corner;

and controlling the vehicle according to the front wheel corner variation to enable the vehicle to continuously approach the optimal safe path.

The invention further provides a system for planning a path for automatic driving takeover, comprising:

the system comprises a takeover acceptance acquiring module, a takeover acceptance acquiring module and a takeover acceptance acquiring module, wherein the takeover acceptance acquiring module is used for acquiring the takeover acceptance of a user on a vehicle;

an operation interference acceptance and control interference acceptance determining module for determining operation interference acceptance and control interference acceptance according to the take-over acceptance;

the operation interference risk and control risk determining module is used for determining operation interference risks and control risks according to the operation interference acceptance and the control interference acceptance;

a takeover risk determining module, configured to determine a takeover risk according to the operation interference risk and the control risk;

the ambient environment risk acquisition module is used for acquiring ambient environment risks;

a unified risk determination module for determining a unified risk of the autonomous vehicle according to the takeover risk and the ambient environment risk;

the optimal safety path determining module is used for determining a set of points with the lowest expected risk strength in multiple risk evaluations of the automatic driving vehicle according to the unified risk and recording the set as an optimal safety path;

and the control module is used for controlling the corner of the front wheel of the vehicle according to the optimal safe path so as to ensure that the track of the vehicle is attached to the optimal safe path.

The invention additionally provides a computer-readable storage medium storing program code for execution by a device, the program code comprising a method of path planning for autopilot take-over according to any of claims 1 to 8.

According to the specific embodiment provided by the invention, the invention discloses the following technical effects:

the method in the invention considers the driving risk elements of the automatic driving vehicle (even in the full automatic driving stage) more comprehensively, namely, considers the potential takeover risk, and plans the minimum risk path of the vehicle by combining the risk of the surrounding environment under the condition so as to improve the safety of automatic driving;

the method makes up the defect that the research work aiming at taking over risk and applied to intelligent vehicle motion planning is lacked in the existing taking over process research.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.

Fig. 1 is a flowchart of a path planning method for automatic driving take-over according to an embodiment of the present invention;

fig. 2 is a schematic structural diagram of a path planning system for automatic driving take-over according to an embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.

Fig. 1 is a flowchart of a path planning method for automatic driving take-over according to an embodiment of the present invention, and as shown in fig. 1, the method includes:

step 101: and acquiring the acceptance of the user for the vehicle.

The comfort recognition result is also called the acceptance of the take-over, using sr_taThe acceptance of take-over includes the acceptance of the user for two types of information during take-over, respectively the acceptance of the user for the disturbance of handling AC transmitted by the vehicle motion control device_disAnd the user's acceptance of the ability to control the movement of the vehicle by himself after the disturbance of the operation, i.e. the control disturbance acceptance AC_con. For single useThe subjective evaluation he/she can make is a limited set, denoted SR_taThe set does not exceed a limited set AC (of approvals of steering disturbances that the user himself can make)_disWith a limited set of AC's (for which the user can make approval for vehicle motion control himself)_conThe cartesian product of (a), which is noted as:

SR_ta＝AC_dis×AC_con

step 102: and determining operation interference acceptance and control interference acceptance according to the take-over acceptance.

Operating a disturbance acceptance AC_disA larger value (a higher subjective evaluation score) represents that the user considers that the operation disturbance (the tactile interaction type) is received in the take-over process to be larger. At the same time, a greater operational disturbance means a correspondingly higher risk of take-over, this type of risk being denoted r_dis(referred to as interference risk).

Controlling the interference acceptance AC_conThe larger or smaller value of (the higher or lower the subjective evaluation score) represents the higher or lower level of self-approval of the user's own ability to control the vehicle motion during take over. The higher (self-negative) or lower (self-inferior) level of self-recognition is considered as two extreme judgments for the actual ability of the user, and thus there is a cognitive bias. Thus, the greater the risk associated with these two extremes, this type of risk is denoted as r_con(referred to as controlling risk); controlling the risk r_conThe lowest value of (a) corresponds to the median of the elements in the user's subjective rating score (finite) set for that type of recognition.

Step 103: and determining operation interference risks and control risks according to the operation interference recognition and the control interference recognition.

Based on human factors engineering theory, the generation of misoperation is caused by error feedback formed under the condition that an operator realizes deviation or unsuitable situational awareness. However, the above reasons are subjective factors, cannot be directly measured, and therefore cannot be directly used for risk assessment. However, feedback of the motion state of the vehicle caused by a driver's misoperation can be madeAnd directly observing by a sensor. Therefore, the estimated value of the driver awareness information can be obtained by the observed vehicle motion state feedback data. The estimated values are divided into two categories, namely interference risk r_disAnd controlling the risk r_con. Further, based on the set theory knowledge, the risk assessment model has the following basic multiple mapping relationships, namely:

undetectable subjective intention information set { I } → set of corrected values for subjective evaluation { S }_dis,S_con} → set of estimated values of correction values

Interference risk and control risk set r_dis,r_con→ take over risk (internal risk) { r }_in}。

Further can be expressed as:

f:{I}→{S_dis,S_con}

j:{r_dis,r_con}→{r_in}

the subjective intention information set { I } belongs to information that cannot be directly measured, and is objectively expressed as a subjective evaluation result in the form of a clear positive integer value under the influence of factors such as biological variability. Based on any numerical subjective evaluation questionnaire evaluation scale, the maximum value vector and the minimum value vector of the evaluation scale are respectively

And

meanwhile, the evaluation value of any one driver for interference and control feelingThe result can be recorded as a two-dimensional vector, (S)_dis,S_con) The maximum and minimum numerical vectors corresponding thereto are

And

due to the individual difference influence, the pair (S) is required_dis,S_con) And (4) carrying out correction, and assigning an expression as follows:

by repeating the independent experiment method and based on the correction calculation method described above, a set of corrected numerical values { S ] for corrected numerical value estimation can be obtained_dis,S_conAnd a set of input values { D }_Lateral,L_lan}. Wherein D is_LateralIs the transverse line deviation of the vehicle and the central line of the lane at the current moment, L_lanIs the lane width. Then, the estimation model (before model parameter identification) is shown as follows according to the multivariate linear regression rationale:

(S_dis,S_con)＝β(D_Lateral,L_lan)

where β is the parameter matrix of the linear regression model.

Further, according to the least squares method, an estimate of β can be obtained:

then, the final estimation model is

(S_dis,S_con)＝(((D_Lateral,L_lan)^T(D_Lateral,L_lan))^-1(D_Lateral,L_lan)^T(S_dis,S_con))(D_Lateral,L_lan)

Step 104: and determining the takeover risk according to the operation interference risk and the control risk.

After the estimated value set of the correction value is obtained through a linear regression model and a least square method, the risk r is taken over by a user in the vehicle_inThe calculation formula of (a) is as follows:

wherein the content of the first and second substances,

is a harmonic mean of the control risk and the interference risk.

Step 105: and acquiring the risk of the surrounding environment.

Ambient (external) risk r_outTwo aspects are involved, namely the external environment other than the autonomous vehicle (own vehicle) and the autonomous vehicle (own vehicle) environment other than the driver.

Ambient (external) risk r_outExists within an effective autonomous vehicle viewing range R. In a cartesian coordinate system, the location of the autonomous vehicle is used as an origin to form a normal sphere (norm ball), i.e. the effective observation of the autonomous vehicle on the external environment is within the normal sphere and can be recorded as

Wherein, X_CFor the spatial position (Cartesian coordinate system) of the autonomous vehicle, R is a parameter constituting a paradigm sphere, which is a circle if it is on a two-dimensional plane, R is the radius of the circle, and X is the radius of the circle_CIs the center of the circle, and X is any point in the circle.

Ambient (external) risk r_outThe risk elements of (a) are divided into a hierarchy (hierarchy) structure having a total of six levels, with different levels being ordered causally. For example, rainfall makes the road slippery, reducing the road surfaceThe coefficient of adhesion. In this causal relationship, rainfall is the "cause" and the reduction in road adhesion coefficient is the "effect"; the vehicle can run in the traffic environment, and has the precondition of road besides the running capability. Road infrastructure can be set up also because of the preconditions for roads. If the existence of the road is a "cause", the elements such as the vehicle and the road arrangement can be the "effect" of the "cause". In the present invention, we consider the risk of "cause" first, and then the risk of "effect". Therefore, the temperature of the molten metal is controlled,

the lowest (sixth) risk elements are related to climatic conditions, including but not limited to elements of sunny, rain, fog, haze, and wind;

the next lower (fifth) risk element is related to road conditions and not to transportation means. Such as road surface adhesion coefficient, road curvature, number of lane lines, intersections, and ramps.

Similarly, the road condition has a certain influence on the existence and type of the road facility, such as the need of the road information prompting facility for the crossroads or the ramps. Furthermore, when considering climate, road conditions and road settings, attention is paid to the traffic environment and to the compliance with traffic regulations. Thus, traffic regulations may serve as a "cause" (thereby restricting the traffic behavior of the vehicle), and the traffic behavior of the vehicle may serve as the "effect" of this "cause".

The risk elements of the middle-lower middle layer (fourth layer) are road facilities such as traffic lights, isolation belts, lane markings and the like;

the middle and upper (third) risk elements are traffic regulations, such as rules that a vehicle needs to decelerate when approaching a crosswalk.

Similarly, even if we consider climate, road conditions, road facilities and traffic regulations, the driving of a vehicle is inevitably limited and affected by its own system. However, the vehicle's own system, as a risk factor, appears to have no direct causal relationship to climate, road conditions, road infrastructure, and traffic regulations. However, in this patent, the above-mentioned four types of elements are elements that can drive and restrict the vehicle from traveling, and it is considered that all of them have a certain degree of responsibility for the consequences of aging, failure, etc. that occur in the vehicle system due to traveling.

The next-to-upper (second) risk element is then a vehicle subsystem (a system that does not generate direct interaction with in-vehicle users).

Finally, the climate, road conditions, road facilities, traffic regulations and the vehicle system actually affect the traffic behaviors of the surrounding vehicles and pedestrians, and the traffic behaviors of the surrounding vehicles and the pedestrians affect the behaviors and decisions of the drivers of the vehicles, so that the traffic behaviors of the vehicles are affected.

Thus, the uppermost (first) risk element is the traffic behavior of the week and the pedestrian.

Any risk element has three risk element states (potentially indicating that the risk elements of different layers have equal positions), namely a "dangerous" state, a "caution" state and a "safe" state, which respectively correspond to risk strengths of 1 · M, 0.5 · M and 0, wherein M has two states, and one state, M, refers to any method capable of calculating the risk of the risk element (the invention considers the requirement of fusion with the existing risk assessment algorithm, and therefore, any method capable of calculating the risk of the risk element is set as M); another property is that M ═ 1, i.e., fusion is developed without regard to the risk calculation method with the existing risk elements.

The risk strength characterized by the risk elements can be calculated by two ways, namely by the operator xi (m1, m2) selecting the mode to be used when calculating the risk strength of the risk elements. Wherein M is present in two states, M1 ═ 1 and M1 ═ 0; m2 ∈ N. Let xi (1, m2), assume that there are three risk strengths for any one risk element, namely, a "dangerous" state, a "caution" state, and a "safe" state. The risk intensity for each state is fixed, 1, 0.5 and 0 respectively. The risk intensity for the three states can then be expressed again as 1 · m1, 0.5 · m1 and 0, combined with m1 being 1. In other words, this is a qualitative wind direction intensity calculation. In contrast, there is a quantitative way of calculating the risk intensity. However, quantitative calculation methods are numerous and non-enumeratable. In this patent, any existing quantitative calculation method of risk intensity can be fused through an operator xi, and the numerical calculation result is defined as an m2 value. Therefore, the risk intensity of any one risk element at Ξ ═ (0, m2) is m 2.

The judgment of the state of any risk element can be from the existing model (research), human engagement (stipulation) and related experience, namely, for different types of risk elements, the judgment of the state of the risk element can be carried out by combining related professional knowledge or mathematical models in the industry.

Any risk element is judged to be within the paradigm sphere formed by the own vehicle observation range R. Within this paradigm sphere, the spatial location of the risk strength provided by a risk element corresponds (determines) to the true spatial location of the risk element.

Ambient (external) risk r_outIs a superposition of the risk intensities provided by the risk elements of different layers, and is marked as

r_out＝∑K_lir_li

Wherein r is_liIs the risk intensity, K, of all risk elements of layer i_liIs the weighting factor for that layer.

The risk intensity provided for any one risk layer is the superposition of the risk intensities of all risk elements within that layer, denoted as

r_li＝∑re_ij

Wherein re_ijIs the risk strength of the jth risk element of the ith layer.

Step 106: determining a unified risk of the autonomous vehicle according to the takeover risk and the ambient risk.

Unified risk R of self-vehicle at any moment_driTake over (internal) risks r for in-vehicle users, respectively_inRisk to the surroundings (outside) r_outLinear superposition of

Wherein the content of the first and second substances,

and

are weighting factors for take-over (internal) risks and ambient (external) risks, respectively.

Step 107: and determining a set of points with the lowest expected risk strength in the multiple risk assessment of the automatic driving vehicle according to the unified risk, and recording as an optimal safe path.

At any time, the set family of the path planning points of the self vehicle is marked as P_cCan be represented as

P_c＝{P_c1,P_c2,…,P_cn}

Wherein, P_cnRefers to the nth set of path planning points, the whole set P_cIs generated from an initial set P_c1Initially, newly appearing elements (collections) are sorted in a sequence of corner-labeled numbers and are individually grouped into the family of collections. The whole planning process consists of the results of single planning, and the plans of different times correspond to the sequence of numbers { c1, c2, …, cn }, namely the first plan is called c1, and the total number of the plans is n.

Set P_c1The middle point is the original point with the current self-vehicle position X as the origin and R as the origin_mThe boundary of the paradigm sphere, which is the distance, is on the intersection curve with the road surface. Meanwhile, provision is made for:

R_m＝v·c

where c is a moving range coefficient and v is a scalar of the current vehicle speed. Then, a reference P is obtained_c1Set P of_c1cAnd is recorded as:

P_c1c＝{X_P|||X-X_P||＝vc}∩{road sueface points}

needs to be in the set P_c1cTo find the set (sub-optimal solution set) with lower risk intensity, and the point (set) is P_c1Is marked as

P_c1＝{argmin{rp_c1c}±γ}

Wherein, { rp_c1cIs the set P_c1cThe set of risk strengths possessed by each element, note here that P_c1It is possible that not a single point set, but a set of points.

When P is present_c1When not a single point set, the points within the set can be defined as:

wherein

Refer to in the set P_c1The m-th element in (b).

Draw out P_c1Then, P is added_c1Repeating the above steps as a new X point (set) to obtain P_c2(modified here to

) In this way, iterative calculations are performed to obtain a new point (set).

For a single path, it is a set of single coordinate points, denoted

Due to the "turn-around" routes that cannot be completed by the vehicle, it is necessary to screen points whose spatial positions satisfy the vehicle's motion ability in each risk assessment result, for example, in a set

To the selected point

Ratio collection

To the selected point

The values in the vehicle coordinate system are all large, whichThe relationship is expressed as:

(Note that the unreachable problem of the target point in the conventional manual potential field method may not reach the target point because the repulsive potential field in front of the target point is larger than the attractive potential field, in addition to the fact that the planned point may fall into the local minimum point; a commonly used solution, for example, provides an artificial disturbance to the current planned point to make it jump out of the local minimum point. however, in the risk assessment work, the action of adding the artificial disturbance is also a dangerous action because the artificial disturbance may make the planned point jump into a high-risk region, which is unreasonable; secondly, the repulsive potential field provided by the risk element forms a risk barrier to the target point, which indicates that the region is at a high risk and the vehicle should stop moving ahead, rather than forcibly crossing the risk aggregation region to reach the target point; in this patent, if the target point is not set, the path planning of the vehicle should be along a low-risk path, if the area where the current risk element is located cannot be crossed, the vehicle should stop advancing. )

The calculation is iterated until the path points are spatially "stagnated" (even if the distances between the planned path points are small, the position of the vehicle in the space is not significantly changed), namely, any values delta and N are given_PSuch that (where collective symbols are used instead of existing single location points)

At this time, only P is reserved_cnAs a result of the final planning, i.e. discarding P_cn+1Up to P_cn+NpThe planning result of (2).

In the whole planning process, an optimal solution set is not obtained every time, but a suboptimal solution set, namely a point with smaller risk is obtained, and then a path with the minimum risk expectation is selected from all risk evaluations. Thus, for any one evaluationIt is estimated that the risk intensity of all points (elements) in the suboptimal solution set is different, that is, the risk intensity corresponding to each point in the suboptimal solution set in the c1 th risk assessment result is:

where t means that there are t elements in the c1 th risk assessment result (suboptimal solution set).

For risk assessment of a total of cn times, the conditional probabilities in the probability theory and the expected numerical characteristics of the discrete random variables are based. The mathematical expectation expression is as follows

Then, consider the non-reentrant problem as event B (event B), so the optimal path is

Step 108: and controlling the corner of the front wheel of the vehicle according to the optimal safety path to ensure that the vehicle road is attached to the optimal safety path.

Step 109: a path error is determined.

Step 110: and obtaining the front wheel rotation angle variable quantity based on the path error and the vehicle front wheel rotation angle.

Step 111: and controlling the vehicle according to the front wheel corner variation to enable the vehicle to continuously approach the optimal safe path.

At the actuator, denoted as ζ, and redefining a limited set of path points constituting this path ζ, denoted as

Meanwhile, a path formed by the motion of the vehicle at present is defined and is marked as xi. However, the path is actually composed of finite points, so we need to fit the finite set by using quintic polynomial method to obtain a continuous path functionNumber, is marked as xi_c。

Determining an optimal safe path, denoted as ζ, and redefining a finite set of path points constituting the path, denoted as ζ

Meanwhile, a path formed by the motion of the vehicle at present is defined and is marked as xi. However, since the path is actually composed of finite points, it is necessary to fit the finite set using quintic polynomial method to obtain a continuous path function, which is expressed as ξ_c。

Regarding the planned optimal safe path zeta as a desired path, the front wheel turning angle theta needs to be controlled (assuming that the vehicle is a front wheel steering vehicle), and the vehicle path zeta is attached to the optimal safe path zeta. Here, a path error e, e ═ ζ - ξ | is defined, and the step size per error observation is denoted as T.

We consider the planned optimal safe path ζ as a desired path and we need to control the front wheel steering angle θ (assuming our vehicle is a front-wheel-steering vehicle) and fit the own vehicle path ξ toward the optimal safe path ζ. Here we define the path error ε, and the step size for each error observation is denoted as T.

After the error epsilon is observed, a PID controller is adopted, and the front wheel steering angle of the vehicle is used as a control quantity. (within a step T, it is desirable for the vehicle to be able to derive a front wheel angle change Δ θ from the error ε observed during the beginning of the step T by a PID controller, the front wheel angle change Δ θ causing the vehicle to begin to decrease by a step-wise amount within the T, thereby causing the vehicle to approach the optimal safe path ζ, the control relationship being

Δθ＝K_p·ε+K_i·∫ε·dt+K_d·(dε/dt)

Wherein K_pIs the proportional term coefficient, K, of the PID controller_iIs the integral term coefficient, K, of the PID controller_dIs the PID controller differential term coefficient.

In this example we consider a special (extreme) case, i.e. we are falseThe method is characterized in that the weather is sunny, the road is a straight road with a single lane (the gradient is zero), the state of the road surface is good, no road facilities are provided, the vehicle collision is not allowed by regulations, the vehicle system is always kept normal and no pedestrian is provided, and only one vehicle (side vehicle) (the coordinate is x) is arranged in front of the self vehicle (the coordinate is x)₀). Then, at this point:

rl₆0 (no risk is considered on sunny days);

rl₅0 (good road surface condition is considered as no risk);

rl₄0 (no asset is considered as no risk);

rl₃0 (non-crashable legislation is considered as no risk);

rl₂0 (vehicle systems are always normally considered risk-free);

rl₁1 m1 (no pedestrian only having a week risk)

M selects the Artificial potential field method (Artificial potential field). Therefore, the risk intensity of the side vehicle to the self vehicle is obtained by the formula of the artificial potential field method and is recorded as

rl₁＝1/2k₀r^2(x,x₀)

Wherein k is₀Is the potential field coefficient, r ^2(x, x)₀) Is a distance function based on the coordinates of the own vehicle and the side vehicle;

ambient (external) risk r_outIs composed of

r_out＝∑kl_irl_j＝kl₁[1/2k₀r^2(x,x₀)]

Designing a subjective evaluation questionnaire according to the SAEJ1441 standard to obtain subjective evaluation scores of the in-vehicle user about the takeover process, wherein the subjective evaluation scores are interference evaluation sr_tadisAnd control evaluation sr_tacon；

Evaluation of sr for interference_tadisAnd control evaluation sr_tacon(two evaluation types) an offline training dataset is constructed that is used to train the predictive model for its evaluation results. Specifically, first, different subjective evaluation grades (for the two evaluation types) are set as different evaluation typesAnd (4) a label. Secondly, the vehicle speed v (own vehicle) and the road curvature rho data corresponding to different subjective evaluation levels are used as input data under the subjective evaluation levels, so that (two, aiming at different evaluation types) subjective evaluation data sets are constructed.

In addition, based on two established subjective evaluation data sets, two neural network models are respectively established (corresponding to the two data sets one by one), the two data sets are used, and an error reverse propagation algorithm is adopted to complete the off-line training work of the two neural network models, so that two subjective evaluation prediction models of the taking-over process of the vehicle user based on the vehicle speed v and the road curvature rho data are obtained;

in the actual driving process of the self-vehicle, the automatic driving system can predict the subjective evaluation result of the driver in the potential taking-over process according to the current vehicle speed and the road curvature and by combining the trained neural network model, wherein the prediction result is sr_ta，sr_taIncluding interference assessment sr_tadisAnd control evaluation sr_tacon；

Based on the predicted subjective evaluation result sr_tadisThe interference risk r can be designed_disIs composed of

r_dis＝kr_dissr_tadis

Wherein, kr_disThe correction factor is evaluated for interference. After the information of the current self-parking position and the week position is introduced, the formula can be further rewritten as:

r_dis＝(kr_dissr_tadis·r^2(x,x₀))/2

in the same way, based on the predicted subjective evaluation result sr_taconAnd considering the own car position and the week car position information, the interference risk can be designed to be

r_con＝kr_con·(1/(1+e^(sr_tacon-sr_tacon-mid)))sig(sr_tacon-sr_tacon-mid)

Wherein, kr_disFor the interference evaluation of the correction factor, sr_tacon-midFor subjective evaluation of median in the ruler table (sr when using SAEJ 1441)_tacon-mid5), the formula can be further rewritten as the current own vehicle position and week position information is introduced

r_con＝[kr_con·(1/(1+e^(sr_tacon-sr_tacon-mid)))sig(sr_tacon-sr_tacon-mid)r^2(x,x₀)]/2

Then an (internal) risk r can be obtained_in，

r_in＝||(kr_dissr_tadis·r^2(x,x₀))/2[kr_con·(1/(1+e^(sr_tacon-sr_tacon-mid)))sig(sr_tacon-sr_tacon-mid)r^2(x,x₀)]/2||₁

A uniform risk R can be obtained_dri：

R_dri＝(kr_dissr_tadis·r^2(x,x₀))/2+kr_in||(kr_dissr_tadis·r^2(x,x₀))/2

r_con＝kr_con·(1/(1+e^(sr_tacon-sr_tacon-mid)))sig(sr_tacon-sr_tacon-mid)r^2(x,x₀)]/2||₁

Unified risk R_driActually representing the risk intensity in the current observation range, and further, developing the track path point generation work based on the risk intensity result.

Finally, we can get an optimal safe path, denoted as ζ, and redefine a finite set of path points, denoted as P, that make up this path ζ_ζ(P₁,…,P_m). Meanwhile, a path formed by the motion of the vehicle at present is defined and is marked as xi. However, the path is actually composed of finite points, so we need to fit the finite set by using quintic polynomial method to get a continuous path function, which is expressed as ξ_c；

We consider the planned optimal safe path ζ as a desired path and we need to control the front wheel steering angle θ (assuming our vehicle is a front-wheel-steering vehicle) and fit the own vehicle path ξ toward the optimal safe path ζ. Here, we define the path error epsilon, and the step length of each error observation is denoted as T; finally, we use the PID controller to take the vehicle front wheel steering angle as a control quantity after observing the error epsilon. (within a step T, it is desirable that the vehicle be able to derive a front wheel angle change Δ θ from the error ε observed during the beginning of the step T by a PID controller, the front wheel angle change Δ θ causing the vehicle to start decreasing in steps within the T by ε, thereby causing the vehicle to approach the optimal safe path ζ continuously, the control relationship being

Δθ＝K_p·ε+K_i·∫ε·dt+K_d·(dε/dt)

Wherein K_pIs the proportional term coefficient, K, of the PID controller_iIs the integral term coefficient, K, of the PID controller_dIs a PID controller differential term coefficient;

based on the ideas and the methods, the subjective evaluation of the users in the vehicle is predicted in the taking-over process, a risk evaluation model is built based on the prediction result, then a continuous optimal path is planned according to the risk evaluation model, the front wheel turning angle of the vehicle is controlled according to the optimal path result, and finally the intelligent vehicle path planning work based on the automatic driving taking-over risk is achieved.

Fig. 2 is a schematic structural diagram of a path planning system for automatic driving take-over according to an embodiment of the present invention, and as shown in fig. 2, the system includes:

a takeover acceptance acquiring module 201, configured to acquire a takeover acceptance of the vehicle by the user;

an operation interference acceptance and control interference acceptance determining module 202, configured to determine an operation interference acceptance and a control interference acceptance according to the takeover acceptance;

an operation interference risk and control risk determining module 203, configured to determine an operation interference risk and a control risk according to the operation interference approval and the control interference approval;

a takeover risk determining module 204, configured to determine a takeover risk according to the operation interference risk and the control risk;

an ambient risk obtaining module 205, configured to obtain an ambient risk;

a unified risk determination module 206 for determining a unified risk of the autonomous vehicle based on the takeover risk and the ambient risk;

the optimal safety path determining module 207 is used for determining a set of points with the lowest expected risk strength in multiple risk evaluations of the automatic driving vehicle according to the unified risk, and recording the set as an optimal safety path;

and the control module 208 is used for controlling the corner of the front wheel of the vehicle according to the optimal safe path track so that the vehicle path is attached to the optimal safe path.

The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.

The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.

Claims

1. A path planning method for automatic driving takeover is characterized by comprising the following steps:

acquiring the acceptance of a user for taking over the vehicle;

acquiring the risk of the surrounding environment;

determining a unified risk of the autonomous vehicle according to the takeover risk and the ambient risk;

2. The method for path planning for autonomous driving take-over according to claim 1, wherein the expression of the take-over risk is as follows:

3. The method for path planning for autonomous driving take-over according to claim 1 wherein the ambient risk includes two aspects, an external environment other than an autonomous vehicle and an autonomous vehicle environment other than a driver.

4. The method of claim 1, wherein the environmental risk comprises six levels, the first level risk elements are traffic behaviors of the week and pedestrians, the second level risk elements are vehicle subsystems, the vehicle subsystems are systems which do not directly interact with users in the vehicles, the third level risk elements are traffic regulations, the fourth level risk elements are road facilities, the fifth level risk elements are road conditions, and the sixth level risk elements are climate conditions.

5. The method for path planning for autonomous driving take-over according to claim 1, wherein the expression of the ambient risk element is as follows:

r_out＝∑K_lir_li

6. The method for path planning for autonomous driving take-over according to claim 1, wherein the expression of the unified risk is as follows:

wherein the content of the first and second substances,

and

7. The method for planning a path taken over by automatic driving according to claim 1, wherein according to the unified risk, determining a set of points with the lowest expected risk intensity in multiple risk evaluations of the automatic driving vehicle, and recording as an optimal safe path specifically comprises:

set family P for determining planned points of bicycle path at any time_c，P_c＝{P_c1,P_c2,…,P_cn}; wherein the set P_c1The middle point is the original point with the current self-vehicle position X as the origin and R as the origin_mIs a distanceThe boundary of the paradigm ball is on the intersecting curve of the road surface;

according to the set P_c1Point determination set P of_clc(ii) a Wherein, P_c1c＝{X_P|||X-X_P||＝vc}∩{road sueface points}；

In the first set P_clcSelecting the set with the lowest risk intensity as P_c1，P_c1＝{argmin{rp_c1c}±γ}；

With said set P_c1Repeating the above steps to obtain P as a new own vehicle position X_c2Up to the Collection family P_cAll the sets in the vehicle are updated to obtain the final set family P of the planning points of the vehicle path_c＝{P_c1,P_c2,…,P_cn}；

Given arbitrary values of δ and N_PSo that P_cn,P_cn+Np||₁< delta, retention P_cnDiscarding P as a final planning result_cn+1Up to P_cn+NpThe planning result of (2);

Wherein t is the total of t elements in the c1 th risk assessment result;

the problem of non-foldback is considered to be EventB, so the optimal path is

8. The method for planning the path taken over by the automatic driver according to claim 1, wherein the method further comprises, after the step of controlling the turning angle of the front wheel of the vehicle according to the optimal safety path so as to fit the self-vehicle path to the optimal safety path:

determining a path error;

9. A system for path planning for automated driving takeover, the system comprising:

and the control module is used for controlling the corner of the front wheel of the vehicle according to the optimal safety path so that the vehicle road is attached to the optimal safety path.

10. A computer-readable storage medium, characterized in that the computer-readable medium stores program code for execution by a device, the program code comprising the method for path planning for autopilot take-over according to any of claims 1-8.