A human-machine co-driving decision and control system for autonomous vehicles

Abstract

The present invention relates to the field of autonomous driving technology, and in particular to a human-machine co-driving decision and control system applied to autonomous driving vehicles. It includes a strategy planning layer and a decision execution layer; the strategy planning layer includes: a driver state unit; an intention recognition unit; a risk assessment unit; a decision arbitration unit; the decision execution layer includes: a local path unit; a PMP control unit. The human-machine co-driving decision control system based on online observation of the driver's neuromuscular state in the present invention enables the autonomous driving vehicle to reflect the actual driver's neuromuscular state, thereby realizing online adjustment and closed-loop feedback of the human-machine control rights allocation according to different driver characteristics. The present invention can effectively reduce the safety risks brought by human-machine co-driving in autonomous driving, and is conducive to realizing personalized driving needs.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving technology, and in particular to a human-machine co-driving decision-making and control system applied to an autonomous driving vehicle. Background Art

[0002] As people's living standards continue to improve, cars have become an indispensable means of transportation for people to travel, and self-driving cars and the safety issues they bring are increasingly gaining widespread attention. At present, the industry generally believes that the maturity of self-driving car technology itself, the safety compatibility of the existing road traffic system, the degree of acceptability of human society, and the degree of perfection of relevant laws and regulations are still far from the large-scale commercial operation of fully self-driving cars. Therefore, the industry generally believes that promoting SAE L3 self-driving cars in closed areas such as airports and docks and dedicated roads is a more feasible application form for a long time in the future. Unlike fully self-driving cars where drivers can leave the control loop, L3 self-driving cars need to face the problem of shared control between people and cars, that is, human-machine co-driving. At this time, the human-machine system cooperates with each other, and the driver needs to respond to the intervention request issued by the system in a timely manner and be ready to take over control at any time. The takeover, allocation and integration of control rights have an important impact on the stability of the "human-vehicle-road" closed-loop system, reducing the cognitive load and fatigue intensity of drivers, and meeting personalized needs. It is an urgent problem to be solved in the development of smart cars. Summary of the invention

[0003] The purpose of the present invention is to provide a human-machine co-driving decision-making and control system applied to an autonomous vehicle, which is used to solve the problem of the prior art: the control rights takeover, allocation and integration methods cannot be well solved.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions:

[0005] A human-machine co-driving decision and control system applied to autonomous vehicles, including a strategy planning layer and a decision execution layer;

[0006] The strategic planning layer includes:

[0007] A driver state unit is used to collect driving operation signals; it is used to perform online identification of the driver during the execution of actions based on the acquired driver's dynamic state parameters, combined with the driver's neuromuscular dynamics model and the unscented Kalman filter observer algorithm;

[0008] The intention recognition unit uses historical prior information and real-time vehicle information to establish a Bayesian classifier to recognize the driver's intentions for straight-line acceleration, deceleration, lane change, and steering.

[0009] The risk assessment unit is used to calculate the minimum lane change time and the minimum collision time based on the recognition result of the intention recognition unit and in combination with the environmental information; the risk of speeding, lane departure and collision is estimated according to the driver's neuromuscular state parameters obtained by the driver state unit online observation, and the driving intention risk index considering the driver's state is established;

[0010] A decision arbitration unit is used to make a decision arbitration on the human-machine control right based on the risk index obtained by the risk assessment unit, so as to determine whether the vehicle is taken over by the driver and the automatic control system exits, or the automatic control system takes over and the driver exits, or the driver and the automatic control system jointly allocate control rights to jointly control the vehicle, and thereby determine the final decision trajectory;

[0011] The decision execution layer includes:

[0012] The local path unit is used to determine whether the vehicle needs to change lanes or follow the target vehicle ahead based on the decision trajectory determined by the strategic planning layer and combined with vehicle and road information;

[0013] The PMP control unit is used to establish a Pontryagin minimum control strategy with the decision trajectory tracking position deviation as the target, the steering motor torque and the electronic throttle opening as the control variables, and the driver's neuromuscular characteristic parameters as the state variables, to control the longitudinal and lateral motion states of the vehicle and realize the adaptive allocation of human-machine co-driving rights.

[0014] Furthermore, the driving operation signal includes: steering wheel angle, pedal opening, and steering switch.

[0015] Furthermore, the driver's dynamic state parameters include: driver's neuromuscular reference torque, contraction torque, and muscle stiffness.

[0016] Furthermore, the online observation method of the driver's dynamic state parameters is:

[0017] Taking the driver's dynamic state parameters as state variables, a dynamic model of the human-machine co-driving system expressed in the form of state equation is established;

[0018] Taking the torque and speed signals actually exerted by the driver on the steering wheel output by the steering wheel torque sensor as the actual measurement values, a state observer based on UKF unscented Kalman filter is established to observe the key state variables of the driver's neuromuscular system online.

[0019] Furthermore, the historical prior information includes: vehicle running trajectory, road curvature, road width, longitudinal and lateral position, speed, acceleration;

[0020] The vehicle real-time information includes steering wheel angle, pedal opening, steering switch signal, relative distance and relative vehicle speed to the preceding vehicle, and lateral distance.

[0021] Furthermore, the environmental information includes: road geometry features, adjacent lanes and vehicle status of the own lane.

[0022] Furthermore, the vehicle and road information includes: vehicle yaw angle, distance from lane line, and obstacle position.

[0023] Furthermore, the intention recognition method includes:

[0024] The key characteristic parameters reflecting lane changing and acceleration and deceleration are selected as variables of the Bayesian classifier model, including the vehicle speed, the acceleration of the vehicle, the time distance between the front and rear vehicles in the same lane, and the time distance between the front and rear vehicles in the target lane. The Bayesian classifier is constructed. The frequency of each variable in the training samples and the conditional probability estimate of each category for each characteristic attribute division are calculated, that is, the posterior probability is calculated using the conditional probability density parameter and the prior probability, and the maximum value of the posterior probability is used as the output result to predict the vehicle's current driving intention.

[0025] The present invention has at least the following beneficial effects:

[0026] The human-machine co-driving decision control system based on online observation of the driver's neuromuscular state in the present invention enables the autonomous driving car to reflect the actual driver's neuromuscular state, thereby realizing online adjustment and closed-loop feedback of the human-machine control rights distribution according to different driver characteristics. The present invention can effectively reduce the safety risks brought by autonomous driving human-machine co-driving, and is conducive to realizing personalized driving needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0028] Figure 1 It is a schematic diagram of the system of the present invention;

[0029] Figure 2 This is the state variable block diagram of the dynamics model of the human-machine co-driving system;

[0030] Figure 3 This is the dynamic model diagram of the human-machine co-driving system after dimensionality reduction;

[0031] Figure 4 This is the logic block diagram of the dynamic state of the human-machine co-driving system after dimensionality reduction. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0033] Please refer to Figure 1 The present invention provides an automatic driving human-machine co-driving decision and control system based on the observation of the driver's neuromuscular state, which is mainly composed of two subsystems: a strategy planning layer and a decision execution layer.

[0034] (1) Strategic Planning Layer

[0035] The subsystem is mainly composed of four units: driver status, intention recognition, risk assessment, and decision arbitration.

[0036] 1.1 Driver status

[0037] The main task of the driver status unit is to collect signals such as steering wheel angle, pedal opening, steering switch, etc. In addition, based on the driver's neuromuscular dynamics model and combined with the unscented Kalman filter observer algorithm, the unit can perform online identification of the driver's neuromuscular reference torque, contraction torque, muscle stiffness and other dynamic state parameters during the driver's action execution process;

[0038] The driver's neuromuscular dynamics model is introduced as follows:

[0039] The driver neuromuscular dynamics model is to convert the transfer function form of the driver's NMS neuromuscular dynamics model into a state space expression, and on this basis, combine it with the vehicle's two-degree-of-freedom model to form a human-machine co-driving dynamics model. According to modern control theory, the nth-order transfer function can be converted into the product of multiple first-order transfer functions, and several first-order transfer function links are connected in series to form a state logic block diagram of the system, such as Figure 2 Shown is the first-order transfer function logic block diagram of the human-machine co-driving dynamics model.

[0040] Figure 2 The state variable block diagram in the figure consists of 14 state variables, including: steering wheel angle δ sw , steering wheel angular velocity Steering wheel torque T m , Arm contraction torque T a , arm reflection torque T r , Reference contraction moment T rm , ankle-foot complex displacement θ limb , ankle-foot complex movement speed θ li ' mb, contact torque T c , internal contraction moment T int , Leg contraction torque T l , activation torque T act , Golgi tendon organ feedback torque T gto , muscle spindle feedback torque T sp .

[0041] Tendons and muscle fibers together constitute the driver's muscles, and the muscles are equivalent to three parts: parallel elastic units, series elastic units, and active contraction units. The passive traction of muscle fibers and the surrounding soft components are represented by parallel elastic units, the tendons and the connection between tendons and surrounding elastic tissues are represented by series elastic units, and the muscle fibers are represented by active contraction units. Therefore, the muscle composition can be equivalently replaced by the above three parts. Therefore, the generation process of muscle force can be represented by the spring damping system, and the intrinsic state of the muscle can be represented by the electromyography (EMG) signal on the driver's skin surface:

[0042]

[0043] Where F is the intrinsic muscle force; F pe and F ce They represent muscle contraction force and muscle elasticity respectively; F max is the maximum tension of the muscle; a(i) represents the degree of muscle activation, represented by the EMG signal; f a (ε) and f b (ε) is the passive and active contraction coefficient of the muscle; ε is the degree of deformation when the muscle contracts.

[0044] The torque output by the muscles is not the result of the action of a single muscle group, but is generated by the joint action of multiple muscle groups. Therefore, a weighted method is used to assign different weights to muscle groups with different contributions. The active contraction torque T acting on the pedals and steering wheel by the driver is defined according to the weighted principle. m and T l .

[0045]

[0046] In the above formula, n i Represents the weight coefficient, R m and R l Indicates the turning radius of the steering wheel and pedals.

[0047] Reflected torque T in the driver's arm NMS model r The auxiliary torque acting on the steering wheel and the arm contraction torque T a It is deduced that:

[0048] T r=T a -T as (3)

[0049] Reference model moment T rm and the torque T acting on the steering wheel m The relationship between can be expressed as:

[0050]

[0051] T sw =T m -T dis (5)

[0052] In the above formula, K a is the active stiffness; T m is the torque acting on the steering wheel; T sw is the total torque of the steering wheel after considering the interference torque; T dis is the external interference torque.

[0053] When the vehicle is turning, the road surface transmits force to the steering system through the tires, and the driver feels the force feedback from the road surface. The feedback torque at this time can be expressed by the steering wheel angle, steering wheel angular velocity and tire slip angle:

[0054]

[0055] In the formula, K f Indicates the steering wheel angle gain coefficient; B f Represents the steering wheel angular velocity gain coefficient; G f Represents the tire slip angle gain coefficient.

[0056] The driver's leg NMS model is responsible for the longitudinal control of the vehicle. Among them, the leg contraction torque T l is the main torque exerted by the driver on the pedal, which can be expressed by the pedal torque T ped and disturbance torque T dis express:

[0057] T l =T ped -T dis (7)

[0058] Driver's leg internal torque characteristics T int Describing the intrinsic dynamic characteristics of the muscle, it can be expressed as:

[0059] T int =T act +K tend θ tend (8)

[0060] Where, T actis the muscle activation torque; K tend is tendon stiffness; θ tend For tendon displacement, the spinal cord receives muscle torque feedback mainly through the Golgi tendon organs; it receives muscle velocity and displacement feedback mainly through the muscle spindle organs, and also obtains the position of the pedal through feedback dynamics.

[0061]

[0062] In the formula, is the desired pedal travel and τ is the time delay.

[0063] In summary, the expression of the dynamic model of the human-machine co-driving system is converted into the state space equation form represented by the augmented matrix:

[0064]

[0065] Where:

[0066]

[0067]

[0068] B 1 =[0 b 2 b 3 0 b 5 b 6 b 7 b 8 ] T

[0069]

[0070] D 1 =D 2 =0

[0071]

[0072] a 54 =-35.71,a 55 =-2.96,

[0073]

[0074] a 111 =I ped ,

[0075] a 121 =k tend , a 146=-2βω 0 ,

[0076]

[0077]

[0078]

[0079]

[0080] In the above formula, X 4 , X 6 , X 14 is the intermediate state quantity, and the input is the desired steering wheel angle and pedal travel

[0081] exist Figure 2 The human-machine co-driving dynamics model after dimensionality reduction and simplification based on the model represented by Figure 3 shown.

[0082] Among them, the arm NMS model after dimensionality reduction:

[0083] The present invention mainly reduces the dimension of the reference model part in the arm NMS model. The reference model simulates the driver's learning process and outputs torque according to the expected steering wheel angle. But in fact, the essence of the driver's output torque is the coordinated contraction of the muscles. The muscle fiber nerve endings receive the excitation of the α motor neurons, thereby outputting muscle torque. The driver's arm NMS model structure after dimensionality reduction simplifies the reference model part, but retains the steering system dynamics, reflex dynamics and active stiffness part. The steering system consists of a steering wheel, gears and racks, a steering column, tires and wheels, and can be expressed as an inertia, damping and spring system. The additional torque feedback term represents the torque generated by the lateral force and the self-alignment torque.

[0084] Assume that the driver holds the steering wheel at the "3-9" o'clock position with both hands without rotation, the arm muscles and skin surface soft tissue are in a relaxed state, the driver's arm and the steering system are dynamically coupled, and the transfer function equation and motion equation are as follows:

[0085]

[0086]

[0087] δ sw is the steering wheel angle, T d is the torque applied by the driver on the steering wheel, M t is the moment generated on the tire due to the lateral force, n rswis the steering ratio of the steering system.

[0088] The reflex dynamics remain the same as before dimensionality reduction. Alpha motor neurons can be stimulated in two ways: directly from the central nervous system and through muscle spindle feedback. Muscle spindles sense the position and speed of muscle movement and feed this information back to alpha motor neurons in real time, which in turn controls muscle movement through alpha motor neurons. It can be seen that muscle spindles control muscles through closed-loop reflexes, and the magnitude of the reflex gain changes with the intensity of the muscle movement pattern. The greater the reflex gain, the greater the additional stiffness and damping of the muscle. The reflex control link can be described by equation 13:

[0089]

[0090]

[0091] In formula 13, ω c is the cut-off frequency, τ is the time delay between the α motor neuron receiving feedback information and sending out the excitation signal, K r Represents reflex stiffness, which varies with the driver's muscle state and driving task.

[0092] It is worth noting that there is a time delay in muscle activation. -sτ , then the time delay of the activation part can be approximately expressed as a first-order transfer function.

[0093] Among them, the leg NMS model after dimensionality reduction:

[0094] The leg NMS model is mainly used to reduce the dimension of contact dynamics and muscle spindle feedback dynamics. Figure 2 The contact dynamics model is included. Contact dynamics characterizes the tiny displacements of the skin and superficial soft tissues of the skin when the driver's ankle-foot complex steps on the pedal. These tiny quantities play a very limited role in the model and will increase the complexity and computational complexity of the model. Similarly, the main function of muscle spindle feedback dynamics is to feedback muscle force, but the feedback effect is not obvious, which will also increase the complexity and computational complexity of the model, and the feedback effect can be replaced by GTO feedback. The leg NMS model describes the dynamic interaction between the ankle-foot complex and the pedal. The input is the desired pedal stroke, and the output is the pedal force and the actual pedal stroke. The reduced-dimensional leg NMS model includes: activation dynamics, intrinsic characteristic dynamics, inertial dynamics, tendon dynamics, and GTO feedback dynamics. The transfer function and motion equation of the reduced-dimensional leg NMS model are as follows:

[0095] The transfer function for inertial dynamics is shown below:

[0096]

[0097] I seg is the moment of inertia of the ankle-foot complex, pedal and their connections.

[0098] The output process of muscle torque is characterized by muscle activation dynamics and muscle intrinsic dynamics, and its transfer function can be expressed as:

[0099]

[0100] T mus It is the muscle co-contraction torque, which is affected by the degree of muscle activation and the intrinsic state of the muscle.

[0101] Muscle intrinsic dynamics describes the changes in intrinsic stiffness and damping of muscles after being stimulated by activation signals. It is essentially a coordinated contraction process, and its transfer function is:

[0102] H int (s) = k int +b int s (17)

[0103] In the formula, k int and b int It is the intrinsic stiffness and damping of the muscle after receiving the excitation signal.

[0104] Muscle activation dynamics describes the process of muscle force generation after the muscle is acted upon by an excitation signal, which can be expressed by a second-order transfer function:

[0105]

[0106] Where β is the relative damping.

[0107] GTO dynamics has an stimulating or inhibiting effect on muscle movement. The process of GTO feedback muscle force can be expressed as:

[0108]

[0109] Tendons are continuous elastic units that connect muscles to bones. Muscle fibers move bones through tendons, which can be expressed by tendon dynamics and tendon stiffness:

[0110]

[0111] The movement of the ankle-foot complex is the result of muscle forces, while the torque is generated by all muscle groups working together. The relevant transfer function is shown below:

[0112] θ limb (s)=H seg (s)·T mus (s) (21)

[0113] Among them, the human-machine co-driving dynamics model after dimensionality reduction

[0114] The driver's NMS model after dimensionality reduction is added with the vehicle's two-degree-of-freedom model and the MPC controller to form the human-machine co-driving dynamics model after dimensionality reduction. Similarly, the n-order transfer function is converted into the product of several first-order transfer functions, and several first-order transfer functions are connected in series to form a human-machine co-driving system, and its state logic block diagram is shown in Figure 4.

[0115] In summary, according to the model conversion theory in modern control theory, the state logic block diagram of the human-machine co-driving dynamics model after dimensionality reduction is converted into a state space expression in the form of an augmented matrix:

[0116]

[0117] In the formula,

[0118]

[0119]

[0120] B 1 =[0 b 2 b 3 b 4 b 5 b 6 ] T

[0121]

[0122] C 1 =[1 1 1 0 1 1]

[0123] C 2 =[1 1 01 0 1]

[0124] D 1 =D 2 =0

[0125] In the formula, a 41 =-ω c k r +ω c 2 B r ,a 44 =-ω c , a 91 =k tend , a 115 =-2βω 0 , b4 =ω c k r -ω c 2 B,

[0126]

[0127]

[0128]

[0129] In the above formula, X 4 , X 7 , X 9 is the intermediate state quantity, and the input is the desired steering wheel angle and pedal travel

[0130] 1.2 Intent Recognition

[0131] The main task of the intention recognition unit is to use historical prior information (including vehicle trajectory, road curvature, road width, longitudinal and lateral position, speed, acceleration, etc.) and establish a Bayesian classifier based on real-time vehicle information such as steering wheel angle, pedal opening, steering switch signal, relative distance and relative speed to the vehicle in front, and lateral distance to recognize the driver's intention to accelerate or decelerate in a straight line, change lanes, and turn.

[0132] The key characteristic parameters reflecting lane changing, acceleration and deceleration, such as the vehicle speed, acceleration, front and rear vehicle distance in the same lane, and front and rear vehicle distance in the target lane, are selected as variables of the Bayesian classifier model. The Bayesian classifier is constructed to calculate the frequency of each variable in the training samples and the conditional probability estimate of each category for each characteristic attribute classification. That is, the posterior probability is calculated using the conditional probability density parameters and the prior probability, and the maximum value of the posterior probability is used as the output result, thereby predicting the vehicle's current driving intention.

[0133] 1.3 Risk Assessment

[0134] The main task of the risk assessment unit is to calculate the minimum collision time based on the driver intention recognition results, combined with environmental information such as road geometry, adjacent lanes and vehicle status in the own lane, and establish a comprehensive evaluation index function of driving risk considering the driver's state based on the driver's neuromuscular state parameters obtained by the driver state unit online observation, combined with the collision energy loss function and lane change time, so as to conduct risk assessment on the driver's behavior;

[0135] 1.3.1 Collision Energy Loss Function

[0136] The collision energy loss can be derived according to the law of conservation of energy:

[0137]

[0138] ΔE=ΔE 0 +ΔE i (twenty three)

[0139] In predicting collision behavior, the collision energy loss of the ego vehicle ΔE 0 Collision energy loss with obstacle vehicle ΔE i It can be expressed as:

[0140]

[0141]

[0142] In the formula, m 0 、v 0 and v 0 'represent the vehicle mass, initial velocity and velocity after collision respectively; m i 、v i 、v i 'represent the mass, speed and post-collision speed of the i-th obstacle vehicle; ΔE 0 , ΔE i is the energy loss of the ego vehicle and the i-th obstacle vehicle, and ΔE represents the total energy loss.

[0143] The energy loss rate can be obtained by taking partial derivative of the above formula:

[0144]

[0145] To obtain the maximum collision energy loss, assume that d(ΔE) / d(v 0 ′)=0, then the energy loss relationship is expressed as:

[0146] v 0 ′=v i ′=(m 0 v 0 +m i v i ) / (m 0 +m i ) (27)

[0147] According to formula 27, when the masses of the two vehicles are equal, the post-collision velocity can be obtained. If the mass of the obstacle vehicle is larger than that of the ego vehicle, the energy loss caused by the collision with the ego vehicle is also greater. The final collision loss can be obtained by defining the loss characteristics η of different vehicles 0 , η j get:

[0148] I i=η 0 ΔE 0 +η i ΔE i (28)

[0149] 1.3.2 Collision Time TTC

[0150] The vehicle collision time TTC can be calculated by the relative distance R between the vehicle and the obstacle vehicle. dis And the relative speed calculation:

[0151]

[0152] 1.3.3 Driving risk comprehensive evaluation index function

[0153] A driving risk comprehensive evaluation index function R expressed as follows is established to evaluate the multiple generated feasible trajectories to obtain the optimal trajectory that meets the evaluation index:

[0154]

[0155] R al (i) = k 1 R ld (i)+k 2 R cl (i)+k 3 R s (i) (31)

[0156] I i =η 0 ΔE 0 +η i ΔE i (32)

[0157] In the formula, R c represents the normalized risk coefficient; R ld Indicates lane-changing risk; R cl Indicates the risk of collision; R s Indicates speed risk; c 1 and c 2 represents the weight factor of different trajectories; τ represents the actual collision distance; τ thr represents the collision distance threshold; k represents the weight factor of different risk selection weights; I i Indicates collision energy loss; E 0 and E i Represent the energy loss of the ego vehicle and the i-th obstacle vehicle respectively;

[0158] 1.4 Decision-making Arbitration

[0159] The main task of the decision arbitration unit is to make decisions on human-machine control rights according to the safety risk index, so as to decide whether the driver takes over the automatic control system to push the vehicle out, the automatic control system takes over the driver to exit, or the driver and the automatic control system jointly allocate control rights to control the vehicle, and thus determine the final decision trajectory;

[0160] According to the risk level of the predicted comprehensive evaluation index function and real-time environment information, the driving system makes an initial allocation of human-machine co-driving control rights. As shown in Table 1, the response value is the lowest when there is no co-driving control. The risk level at this time is the lowest, and the driver only needs to assume the supervision responsibility. Therefore, the automatic driving system occupies 100% of the control authority; correspondingly, when in active shared control, the vehicle may face low-risk scenarios such as meeting, but the driver lacks trust in the automatic driving system and needs to take precautions and be ready to take over control at any time. At this time, it is most appropriate for the automatic driving system to occupy 80% of the control rights and the driver to occupy 20% of the control rights. According to this definition rule, during the automatic driving process of the vehicle, the higher the risk level, the greater the control rights occupied by the driver when a dangerous accident occurs on one side.

[0161] Table 1 Initial allocation of shared control rights

[0162]

[0163] Note: The driver takes over the automatic control system and exits C6; the automatic control system takes over and the driver exits, which is C1; the driver and the automatic control system jointly allocate control rights, which are C2, C3, C4, and C5.

[0164] It should be noted that the method for the driving muscle state unit to observe the driver's neuromuscular characteristic parameters online is: taking the neuromuscular reference torque, contraction torque, muscle stiffness and other dynamic parameters as state variables, and establishing a human-machine co-driving system dynamics model expressed in the form of a state equation; taking the torque and speed signals actually applied to the steering wheel by the driver output by the steering wheel torque sensor as actual measurement values, and establishing a state observer based on the UKF (Unscented Kalman Filter) to observe the driver's key neuromuscular state variables online.

[0165] (2) Decision-making and execution layer

[0166] The main task is to establish the Pontryagin Minimum Principle (PMP) control strategy with the decision trajectory tracking position deviation as the target, the steering motor torque and electronic throttle opening as the control variables, and the driver's neuromuscular characteristic parameters as the state variables, to control the longitudinal and lateral motion states of the vehicle and realize the adaptive allocation of human-machine co-driving rights.

[0167] It should be noted that the calculation method of the PMP control unit is to take the desired trajectory of the vehicle as the center, and perform expansion processing according to the size constraints of the obstacle vehicle and the size constraints of the self-vehicle to generate a safe virtual area where the vehicle is allowed to travel; then, based on the conformal mapping theory, the symmetrically convergent velocity field of the regular area of ​​the unit circle is mapped to the safe virtual area, so as to generate a virtual velocity field inside and outside the irregular area to prevent the vehicle from deviating from the expected trajectory; finally, according to the actual needs of the vehicle for the desired trajectory tracking position and speed, on the basis of the dynamic equations of the human-machine co-driving system, the steering motor torque and pedal opening are used as control variables to establish the PMP minimum control strategy, and obtain the human-machine action torque and pedal travel control law.

[0168] Thus, the present invention produces at least the following beneficial effects:

[0169] The human-machine co-driving decision control system based on online observation of the driver's neuromuscular state in the present invention enables the autonomous driving car to reflect the actual driver's neuromuscular state, thereby realizing online adjustment and closed-loop feedback of the human-machine control rights distribution according to different driver characteristics. The present invention has important practical significance for reducing the safety risks brought by autonomous driving human-machine co-driving and realizing personalized driving needs.

[0170] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. The present invention may be subject to various changes and improvements without departing from the spirit and scope of the present invention. These changes and improvements fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the attached claims and their equivalents.

Claims

1. A human-machine co-driving decision and control system applied to an autonomous vehicle, characterized in that: Includes strategic planning layer and decision-making execution layer; The strategic planning layer includes: A driver state unit is used to collect driving operation signals; it is used to perform online identification of the driver during the execution of actions based on the acquired driver's dynamic state parameters, combined with the driver's neuromuscular dynamics model and the unscented Kalman filter observer algorithm; The intention recognition unit uses historical prior information and real-time vehicle information to establish a Bayesian classifier to recognize the driver's intentions for straight-line acceleration, deceleration, lane change, and steering. The risk assessment unit is used to calculate the minimum lane change time and the minimum collision time based on the recognition result of the intention recognition unit and in combination with the environmental information; the risk of speeding, lane departure and collision is estimated according to the driver's neuromuscular state parameters obtained by the driver state unit online observation, and the driving intention risk index considering the driver's state is established; Assume that the speeding risk, lane departure risk and collision risk are Rs, Rld and Rcl respectively, establish a driving intention risk index R considering the driver's state, and evaluate the generated multiple feasible trajectories; The expression of the driving intention risk index R(i) of each feasible trajectory is as follows: R al (i)=k1R ld (i)+k2R cl (i)+k3R s (i) I i =η0ΔE0+η i DE i In the formula, R al represents the normalized risk coefficient; c1 and c2 represent the weight factors of different trajectories; τ represents the actual collision distance; τ thr represents the collision distance threshold; k represents the weight factor of different risk selection weights; I i represents the collision energy loss; E0 and E i Represent the energy loss of the ego vehicle and the i-th obstacle vehicle respectively; A decision arbitration unit is used to make a decision arbitration on the human-machine control right based on the risk index obtained by the risk assessment unit, so as to determine whether the vehicle is taken over by the driver and the automatic control system exits, or the automatic control system takes over and the driver exits, or the driver and the automatic control system jointly allocate control rights to jointly control the vehicle, and thereby determine the final decision trajectory; The decision execution layer includes: The local path unit is used to determine whether the vehicle needs to change lanes or follow the target vehicle ahead based on the decision trajectory determined by the strategic planning layer and combined with vehicle and road information; The PMP control unit is used to establish a Pontryagin minimum control strategy with the decision trajectory tracking position deviation as the target, the steering motor torque and the electronic throttle opening as the control variables, and the driver's neuromuscular characteristic parameters as the state variables, to control the longitudinal and lateral motion states of the vehicle and realize the adaptive allocation of human-machine co-driving rights; The calculation method of the PMP control unit to realize the adaptive allocation of human-machine co-driving rights includes: Taking the expected trajectory of the vehicle as the center, the expansion process is performed according to the size constraints of the obstacle vehicle and the vehicle size constraints to generate a safe virtual area where the vehicle is allowed to travel; According to the conformal mapping theory, the symmetrically convergent velocity field of the regular region of the unit circle is mapped to the safe virtual region, so as to generate a virtual velocity field inside and outside the irregular region to prevent the vehicle from deviating from the expected trajectory. According to the actual needs of the vehicle for the desired trajectory tracking position and speed, based on the dynamic equation of the human-machine co-driving system, the PMP minimum control strategy is established with the steering motor torque and pedal opening as the control variables to obtain the human-machine action torque and pedal travel control law; The driving operation signal includes: steering wheel angle, pedal opening, steering switch; The driver's dynamic state parameters include: driver's neuromuscular reference torque, contraction torque, and muscle stiffness.

2. The human-machine co-driving decision and control system for an autonomous vehicle according to claim 1, characterized in that: The online observation method of the driver's dynamic state parameters is: Taking the driver's dynamic state parameters as state variables, a dynamic model of the human-machine co-driving system expressed in the form of state equation is established; Taking the torque and speed signals actually exerted by the driver on the steering wheel output by the steering wheel torque sensor as the actual measurement values, a state observer based on UKF unscented Kalman filter is established to observe the key state variables of the driver's neuromuscular system online.

3. The human-machine co-driving decision and control system for an autonomous driving vehicle according to claim 1, characterized in that: The historical prior information includes: vehicle running trajectory, road curvature, road width, longitudinal and lateral position, speed, acceleration; The vehicle real-time information includes steering wheel angle, pedal opening, steering switch signal, relative distance and relative vehicle speed to the preceding vehicle, and lateral distance.

4. The human-machine co-driving decision and control system for an autonomous vehicle according to claim 1, characterized in that: The environmental information includes: road geometry, adjacent lanes and vehicle status in the own lane.

5. The human-machine co-driving decision and control system for an autonomous vehicle according to claim 1, characterized in that: The vehicle and road information includes: vehicle yaw angle, distance from lane line, and obstacle position.

6. The human-machine co-driving decision and control system for an autonomous vehicle according to claim 1, characterized in that: The intention recognition method includes: The key characteristic parameters reflecting lane changing and acceleration and deceleration are selected as variables of the Bayesian classifier model, including the vehicle speed, the acceleration of the vehicle, the time distance between the front and rear vehicles in the same lane, and the time distance between the front and rear vehicles in the target lane. The Bayesian classifier is constructed. The frequency of each variable in the training samples and the conditional probability estimate of each category for each characteristic attribute division are calculated, that is, the posterior probability is calculated using the conditional probability density parameter and the prior probability, and the maximum value of the posterior probability is used as the output result to predict the vehicle's current driving intention.

Images