Automatic emergency obstacle avoidance system for unmanned vehicle and control method thereof

Abstract

The present invention belongs to the field of safe driving of intelligent vehicles, and discloses an automatic emergency obstacle avoidance control method (Consensus Behavior Migration Algorithm, CBMA) for unmanned vehicles, which includes the following steps: first, judging whether a collision can be avoided by braking in the lane; if it is judged that the self-vehicle and the front vehicle can avoid collision, automatic braking is adopted; if it is judged that the self-vehicle cannot avoid collision with the front vehicle after braking, the critical headway of the self-vehicle to the obstacle vehicle in front is adopted, and the front collision warning coefficient is calculated to make an obstacle avoidance strategy, and the self-vehicle and the surrounding vehicles are circularly clustered, and the consensus behavior migration algorithm is introduced to calculate the fitness function based on the planned safe obstacle avoidance trajectory point, and automatic emergency steering is performed to avoid obstacles. The CBMA algorithm solves the problem of mobile trajectory point planning for automatic emergency obstacle avoidance of unmanned vehicles, and reduces the risk of collision with the front vehicle.

Description

Technical Field

[0001] The present invention relates to the field of safe driving of intelligent vehicles, and in particular to an automatic emergency obstacle avoidance system for unmanned vehicles and a control method thereof. Background Art

[0002] With convenient transportation today, smart driving has gradually been recognized by passengers, and many complex traffic scenes are also filled with them. If autonomous driving replaces human drivers, it requires more intelligent equipment to cope with more complex working conditions. For example, during driving, if the vehicle in front suddenly brakes, it is particularly important for the vehicle to have a collision avoidance function. At present, the obstacle avoidance functions on the market include forward collision warning, active emergency braking, obstacle avoidance, and automatic emergency steering. These two systems are used to avoid collisions between the vehicle and the vehicle in front. Forward warning function: Use cameras, radars and other perception sensors to detect the status information of the vehicle in front, and make decisions to remind the driver to pre-move or brake. In order to avoid collisions with obstacles in the lane, an automatic emergency obstacle avoidance control method is proposed by using the forward collision risk coefficient to identify potential dangers as a judgment condition. Through the above analysis, the problems and defects existing in the prior art are:

[0003] 1. In collision avoidance conditions, identifying potentially dangerous obstacles as one of the key technologies for obstacle avoidance cannot guarantee that all vehicles around the ego vehicle are located within the sensor range.

[0004] 2. In complex scene obstacle avoidance systems, a system that only has longitudinal collisions but lacks comprehensive coordination will have poor obstacle avoidance effects.

[0005] 3. In the existing steering obstacle avoidance system, the design variables of the optimal path are limited, thereby reducing the protective effect of the vehicle's passive safety constraints. Summary of the invention

[0006] In view of the problems existing in the prior art, one of the objectives of the present invention is to provide an automatic emergency obstacle avoidance control method for an unmanned vehicle based on a circular clustering consensus behavior migration algorithm.

[0007] Since the dynamic obstacles faced by the vehicle under high-speed driving conditions are cars, the present invention performs circular clustering on the self-vehicle and other vehicles, sets the path estimation radius according to parameters such as vehicle speed, vehicle width, steering angle, yaw rate, and calculates the critical headway; on this basis, when the self-vehicle cannot avoid the obstacle vehicle in front, calculates the forward collision warning coefficient to identify potential dangerous behaviors and activates the obstacle avoidance system; by performing circular clustering on the self-vehicle and the obstacle vehicle in front, using the consensus behavior migration algorithm, calculates the fitness function based on the principle of gravity of the lane change target position and repulsion of the front vehicle, selects the guidance point with the lowest fitness to help the self-vehicle find the next trajectory point, and performs automatic emergency collision avoidance, otherwise automatic emergency braking is still used, that is, the consensus behavior migration algorithm.

[0008] The automatic emergency obstacle avoidance control method for unmanned vehicles based on the circular clustering consensus behavior migration algorithm of the present invention first determines whether the self-vehicle can avoid a collision with the vehicle in front through automatic braking. If it is determined that automatic braking can avoid the collision and there is no danger of rear-end collision between the rear vehicle and the self-vehicle, automatic braking is adopted; if it is determined that automatic braking cannot avoid a collision with the vehicle in front, circular clustering is performed on the self-vehicle and the surrounding obstacle vehicles, and a set of obstacle avoidance trajectory points is created using the consensus behavior migration algorithm to help the self-vehicle find a movable emergency obstacle avoidance trajectory.

[0009] The specific steps are as follows:

[0010] Step 1: Collect real-time data from millimeter-wave radar, binocular camera and vehicle-mounted sensors to obtain the longitudinal distance between the vehicle and the front vehicle, the speed of the front vehicle, the lateral distance between the vehicle and the target vehicle in the adjacent lane, the longitudinal distance between the vehicle and the target vehicle, and the geometric size information of the front vehicle;

[0011] Step 2: Calculate the path curvature radius based on vehicle speed, vehicle width, steering angle, yaw rate and other parameters, calculate the critical headway based on the real-time data obtained by the binocular cameras, compare the critical headway between the vehicle and the preceding vehicle with the actual headway, and determine whether braking can be used to avoid a collision in the lane;

[0012] The step 2 specifically includes the following steps:

[0013] (1) Take the actual distance difference to get the relative speed V r , combined with the vehicle speed to obtain the front vehicle speed; select a smaller curvature radius and combine it with the data collected by the ring sensing unit to obtain the expected critical headway L0;

[0014] (2) If L0>D1, and D1=V r ·TTC A +D0,C b =1, start automatic braking to avoid obstacles; TTC Ais the first-level alarm threshold, D1 is the corresponding first-level braking critical distance, and D0 is the fixed distance value, that is, the minimum distance must be maintained when the relative speed is zero;

[0015] If L0≤D2, and D2=V r ·TTC B +D0,C b =2, when automatic braking fails to avoid obstacles, the automatic emergency obstacle avoidance system is activated;

[0016] Among them, TTC B is the secondary obstacle avoidance threshold; C b =2 is the automatic emergency obstacle avoidance state.

[0017] Step three: Based on the safety time distance TTC required for emergency obstacle avoidance, the forward collision warning coefficient FCW is introduced as a prerequisite for taking obstacle avoidance measures. If FCW>0.3, the scene is an emergency and the CBMA algorithm is activated for obstacle avoidance.

[0018] Step 4: Based on the circular clustering consensus behavior migration method, the obstacle avoidance trajectory points of the car are initialized, and the original trajectory points of the car are iterated through cross-behavior interference to create a set of new moving trajectory points.

[0019] The fourth step comprises:

[0020] (1) Initialize the actual position of the ego vehicle, which includes two parameters on the x-axis and y-axis, define the CBMA parameters and their boundaries, and iteratively select a guide point for the current trajectory point;

[0021] (2) To ensure the diversity of trajectory points, CBMA defines a trajectory point interference vector PRT. This parameter determines whether the initialized trajectory point will move directly to the next guide point. The random number generated by [0.1] is compared with the defined PRT vector. The parameter of each trajectory point must be randomly selected from the given range [lo,hi].

[0022] initial P=X act +rand[0,1]×(x (hi) -x (lo) )

[0023] Among them, P is the initial position of the trajectory point, X act is the actual position of the vehicle, rand[0,1] is a random number between 0 and 1, and x (hi) 、x (lo) is the maximum position range;

[0024] (3) In CBMA, the intersection behavior simulation of intelligent organisms is used to obtain new moving trajectory points through intersection behavior; each generated trajectory point is evaluated through a cost function, and a fitness function based on the principle of the gravitational force of the lane change target position and the repulsive force of surrounding vehicles is calculated. For the current iteration, a guide point with the lowest fitness is selected as the position where the vehicle will move until it reaches the target point.

[0025] Step 5: Use the constraint function to introduce the moving step of the ego vehicle into a distance metric of the feasible area, calculate the fitness function and evaluate the generated new trajectory points, and select the trajectory point with the lowest fitness value as the new position where the ego vehicle will move.

[0026] The step five comprises the following steps:

[0027] (1) The salient feature of CBMA is that the PRT vector is changed at each intersection so that the distance from the actual position of the ego vehicle to the new guidance point must be less than the ego vehicle's moving step length. The soft constraint (penalty) method is used to process the moving step length constraint function and bring the ego vehicle's moving trajectory into a feasible region.

[0028] (2) Calculate the fitness function to evaluate the generated obstacle avoidance trajectory points. The fitness function consists of two parts: the attraction element generated by the target and the repulsion element generated by the vehicle in front and the vehicles in the adjacent lanes. Select the trajectory point with the lowest fitness as the new position to which the vehicle is to move, and then pass through the obstacle gap to reach the target position.

[0029] Another object of the present invention is to provide an automatic emergency obstacle avoidance system for an unmanned vehicle, comprising an ambient sensing unit, a decision unit and a control unit;

[0030] a. Surround sensing unit, which obtains information about the vehicle and surrounding vehicles through two millimeter-wave radars, a binocular camera and the vehicle's sensor data, including the speed, acceleration and distance of the vehicle in front, the vehicle in the adjacent lane and the vehicle behind in the adjacent lane;

[0031] b. The decision-making unit determines whether the vehicle should apply longitudinal braking to avoid obstacles based on the data uploaded by the ambient sensing unit and makes an obstacle avoidance decision;

[0032] c. A control unit, which is used to respond to the decision instructions issued by the decision unit to achieve obstacle avoidance for the vehicle. The control unit includes a wire-controlled brake and wire-controlled steering system, which automatically brakes according to the obstacle avoidance decision made by the consensus behavior migration algorithm or controls the vehicle to automatically steer according to the planned path.

[0033] The surrounding sensing unit includes two millimeter-wave radars and a binocular camera. The millimeter-wave radar is a long-range millimeter-wave radar with a ranging range of 0-300m, an azimuth angle of ±10°, and a pitch angle of 13°. It is installed on the tail fin of the car and the logo on the front bumper; the binocular camera can record information such as the vehicle's speed, acceleration, braking operation, steering operation, GPS, and images in front of the vehicle within a period of time before and after the emergency occurs by utilizing pixel width and focal length. It is installed near the central rearview mirror on the inside of the front windshield of the vehicle.

[0034] The vehicle sensor includes a vehicle speed sensor, a vehicle steering wheel angle sensor and a yaw angle sensor.

[0035] The control device involved in the control unit includes a central processing unit and a memory. The memory has a high-precision map and a computer program. The computer program can execute the automatic emergency obstacle avoidance control method for the unmanned vehicle.

[0036] Compared with the prior art, the solution of the present invention has the following beneficial technical effects:

[0037] The vehicle obstacle avoidance control method based on the circular clustering consensus behavior migration algorithm ensures that the driverless car can accurately identify the distance between cars even in complex road shapes and when there are multiple cars nearby, which not only improves the steering accuracy, but also avoids accidents caused by path errors due to simplification. The CBMA algorithm solves the problem of mobile trajectory point planning for automatic emergency obstacle avoidance of driverless vehicles, reducing the risk of collision with the vehicle in front. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments of the present application. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0039] Figure 1 It is a flow chart of an automatic emergency obstacle avoidance control method for an unmanned vehicle based on a circular clustering consensus behavior migration algorithm provided by an embodiment of the present invention;

[0040] Figure 2 is a schematic diagram of an automatic emergency obstacle avoidance system for a vehicle provided by an embodiment of the present invention;

[0041] Figure 3 is a schematic diagram of the working principle of a binocular camera provided by an embodiment of the present invention and a schematic bird's-eye view of a vehicle in a circular clustering working state;

[0042] Figure 4is a TTC collision time variation process provided by an embodiment of the present invention;

[0043] Figure 5 This is a trajectory point tracking curve effect diagram provided by an embodiment of the present invention; DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of this application more clear, the present invention is further described below in conjunction with the accompanying drawings and embodiments. It should be understood that various equivalent modifications of the present invention by those skilled in the art fall within the scope of the claims attached to this application.

[0045] The present application is further described below in conjunction with specific embodiments and drawings.

[0046] like Figure 2 As shown, an embodiment of the present invention provides an automatic emergency obstacle avoidance system for an unmanned vehicle, including an ambient sensing unit, a decision unit and a control unit;

[0047] a. Surround sensing unit, which uses two millimeter-wave radars, a binocular camera and self-vehicle sensors to obtain information about the self-vehicle and surrounding vehicles, including the speed, acceleration and distance of the front vehicle, the front vehicle in the adjacent lane and the rear vehicle in the adjacent lane;

[0048] b. The decision-making unit determines that the longitudinal braking of the vehicle cannot avoid the obstacle based on the data from the vehicle's sensors and the data uploaded by the surrounding sensing unit, and makes an obstacle avoidance decision;

[0049] c. A control unit, which is used to respond to the decision instructions issued by the decision unit to achieve obstacle avoidance for the vehicle. The control unit includes a wire-controlled brake and wire-controlled steering system, which automatically brakes according to the obstacle avoidance decision made by the consensus behavior migration algorithm or controls the vehicle to automatically steer according to the planned path.

[0050] The two millimeter-wave radars are long-range millimeter-wave radars with a ranging range of 0-300m, an azimuth angle of ±10°, and a pitch angle of 13°. They are installed on the tail fin of the car and the logo on the front bumper; the two cameras can record information such as the vehicle's speed, acceleration, braking operation, steering operation, GPS, and images in front of the vehicle within a period of time before and after an emergency occurs by utilizing pixel width and focal length. They are installed near the central rearview mirror on the inside of the front windshield of the vehicle.

[0051] The vehicle sensor includes a vehicle speed sensor, a vehicle steering wheel angle sensor and a yaw angle sensor.

[0052] The control device involved in the control unit includes a central processing unit and a memory. The memory has a high-precision map and a computer program. The computer program can execute the automatic emergency obstacle avoidance control method for the unmanned vehicle.

[0053] like Figure 1 As shown, the automatic emergency obstacle avoidance control method for unmanned vehicles based on the circular clustering consensus behavior migration algorithm provided by the embodiment of the present invention has the following basic procedures:

[0054] Use binocular cameras and millimeter-wave radars to collect dynamic data of the ego vehicle, obtain the relative speed between the ego vehicle and the front vehicle, the speed of the front vehicle, the speed of the ego vehicle, and the geometric dimensions of the front vehicle, calculate the critical headway through the path curvature radius combined with the data obtained by the radar, and determine whether the ego vehicle brakes in its lane to avoid obstacles. When longitudinal braking cannot avoid obstacles, calculate the forward collision warning coefficient to determine the warning level, and then activate the obstacle avoidance system.

[0055] The automatic emergency obstacle avoidance control method for unmanned vehicles based on the circular clustering consensus behavior migration algorithm specifically includes the following steps:

[0056] Step 1: Collect real-time data from millimeter-wave radar, binocular camera and vehicle-mounted sensors to obtain the longitudinal distance between the vehicle and the front vehicle, the speed of the front vehicle, the lateral distance between the vehicle and the target vehicle in the adjacent lane, the longitudinal distance between the vehicle and the target vehicle, and the geometric size information of the front vehicle.

[0057] Step 2: Calculate the path curvature radius based on vehicle speed, vehicle width, steering angle, yaw rate and other parameters, calculate the critical headway with the real-time data obtained by the binocular cameras, compare the critical headway with the front vehicle and the actual headway, and determine whether braking can be used to avoid a collision in the lane.

[0058] Step 3: Calculate the required safety time distance TTC for emergency obstacle avoidance based on the longitudinal critical headway, introduce the forward collision warning coefficient FCW, confirm that the scene is an emergency, and the system adopts an automatic emergency obstacle avoidance strategy;

[0059] Step 4: Based on the circular clustering consensus behavior migration method, the vehicle obstacle avoidance trajectory points are initialized, and the original trajectory points of the vehicle are iterated through cross-behavior interference to create a set of new moving trajectory points;

[0060] Step 5: Use the constraint function to introduce the moving step of the ego vehicle into a distance metric of the feasible area, calculate the fitness function and evaluate the generated new trajectory points, and select the trajectory point with the lowest fitness value as the new position where the ego vehicle will move.

[0061] The specific operations in this embodiment are as follows:

[0062] Combination Figure 3The working principle diagram of the binocular camera in working state is used to illustrate that the vehicle external surrounding sensing unit includes two millimeter wave radars, which are long-range millimeter wave radars with a ranging range of 0-300m, an azimuth angle of ±10°, and a pitch angle of 13°. They are installed on the tail fin part of the car and the logo of the front bumper; a binocular camera is installed near the central rearview mirror on the inside of the front windshield of the car. The pixel width and focal length of the binocular camera can record the vehicle speed, acceleration, braking operation, steering operation, GPS and images in front of the vehicle within a period of time before and after the emergency occurs. In order to obtain information such as the distance between vehicles and the speed of the vehicle in front during the following process, the present invention adopts a ranging method based on the principle of pinhole imaging to calculate the speed of the vehicle in front:

[0063] PD=WF,D=WF / P (1)

[0064] Among them, P is the pixel width, D is the actual distance, F is the focal length, and W is the actual object width. The relative speed is obtained by differentiating the actual distance, and the speed of the leading vehicle is obtained by combining the speed of the vehicle.

[0065] Combination Figure 3 A schematic bird's-eye view of the vehicle in the circular clustering working state, the vehicle automatic emergency obstacle avoidance control method is divided into two objectives:

[0066] The first is the path estimation algorithm, which collects the target vehicle data obtained by binocular cameras and millimeter-wave radars; combined with vehicle dynamics, it obtains the critical headway L0. This algorithm can be applied to conditions where the vehicle speed is as high as 80-100km / h, and the road environment is a straight highway with a curvature radius of more than 30 meters. The performance will decrease as the distance to the previous vehicle becomes farther. The specific steps are as follows:

[0067] The estimated curvature radius R of the path is determined by two estimated curvature radii R1 and R2. The first estimated curvature radius R1 is obtained by the vehicle speed and steering angle, and the second estimated curvature radius R2 is obtained by the vehicle speed and yaw rate. The vehicle automatic emergency obstacle avoidance control system uses the vehicle distance as the judgment variable, performs the first-level braking system according to the following vehicle status, and activates the obstacle avoidance system at the second level.

[0068]

[0069]

[0070] Among them, S is the stability coefficient, L is the wheel base, δ is the steering angle, γ is the yaw rate, Q is the transmission ratio, and V is the vehicle speed. Considering the vehicle's working conditions, the minimum curvature radius is selected and combined with radar data (distance, direction) to obtain the expected critical headway.

[0071] The second is the safety / danger decision algorithm. The critical headway is calculated by the road condition (wet / dry), the speed of the following vehicle and the relative speed. The relative speed is the differential value of the headway specified by the path estimation algorithm. The actual headway is calculated by the road condition, the speed of the self-vehicle and the relative speed. The critical headway is compared with the measured headway to process the safety / danger decision. The critical headway is expressed as:

[0072]

[0073] Among them, L0 is the critical headway, α1 is the automatic braking control deceleration, α2 is the front vehicle deceleration, t1 is the deceleration time, t2 is the delay time, d is the headway offset, and v is the relative speed.

[0074] The safe time distance TTC required for emergency obstacle avoidance reflects the vehicle's judgment mechanism of rear-end collision danger and the urgency of potential rear-end collision danger. Figure 4 Based on this conclusion, the critical vehicle distance L0 is used as the judgment variable. Considering that the relative speed is zero and the distance between vehicles is very close, a fixed distance value D0 is added to the algorithm, that is, the minimum distance must be maintained when the relative speed is zero. This paper takes 3m. r , calculate the safe vehicle distance based on TTC and compare it with the current vehicle distance L0 as the basis for system state switching.

[0075] The system is in the first-level braking state. In this state, the system starts automatic braking to avoid collision. The conditions are:

[0076] If L0>D1,

[0077] Then D1=V r ·TTC A +D0 (5)

[0078] C b =1

[0079] Among them, TTC A is the first-level alarm threshold, and D1 is the corresponding first-level braking critical distance. When the distance is greater than D1, braking is implemented, C b =1 is the automatic braking status level.

[0080] The system is in the secondary obstacle avoidance state. In this state, the system activates the obstacle avoidance system as the danger level continues to increase. The conditions are:

[0081] if L0≤D2,

[0082] Then D2=V r ·TTC B +D0 (6)

[0083] Cb =2

[0084] Among them, TTC B is the secondary obstacle avoidance threshold; C b =2 is the automatic emergency obstacle avoidance state.

[0085] When the system triggers the automatic emergency obstacle avoidance system, the control unit controls the actuator to perform the automatic emergency obstacle avoidance action, and the safety time required for steering to avoid obstacles should meet the following requirements:

[0086]

[0087] Based on the concept of collision time TTC, this paper introduces the forward collision warning coefficient (FCW) as a prerequisite for whether to take obstacle avoidance measures:

[0088]

[0089] Among them, V R is the relative speed between the vehicle and the obstacle, and L0 is the critical headway. When FCR>0.3, the coefficient rises to level III, and its application scenario is an emergency.

[0090] In the process of obstacle avoidance, the consensus behavior migration algorithm is used to move from the starting position to the target position without colliding with surrounding vehicles, and its movement trajectory is found in real time; this trajectory is considered to be a collection of points with a small distance that the car must pass through so that the car does not hit obstacles during the movement, thereby generating a safe trajectory to move to the target, including the following steps:

[0091] In the first step, the trajectory point is initialized, assuming the ego vehicle position P, which contains two parameters on the two axes x and y to which the ego vehicle will move. At the beginning of the algorithm, the CBMA algorithm parameters and their boundaries are defined (as shown in Table 1).

[0092] Table 1 CBMA algorithm definition parameters

[0093] <![CDATA[S d ]]> remove PRT WPL MinDiv total <![CDATA[r sensor ]]> <![CDATA[r mobilestep ]]>

[0094] The trajectory points are initialized around the actual position of the vehicle. There is also a so-called cost value associated with each trajectory point. The cost value is used to evaluate the fitness of the trajectory point and select a guide point for the current iteration.

[0095] The second step is qualitative change. To ensure the diversity of trajectory points, the CBMA algorithm defines a trajectory point interference vector PRT. This parameter determines whether the initialized trajectory point will move directly to the next guide point. The random number generated by [0.1] is compared with the defined PRT vector. The parameters of each trajectory point must be randomly selected from the given range [lo, hi]:

[0096] initialP=X act +rand[0,1]×(x (hi) -x (lo) ) (9)

[0097] Among them, P is the initial position of the trajectory point, X act is the actual position of the vehicle, rand[0,1] is a random number between 0 and 1, and x (hi) 、x (lo) is the maximum position range.

[0098] The third step is crossover. The PRT vector is the trajectory point before performing crossover motion in the search space. The PRT vector is changed every time it jumps. The trajectory of the trajectory point is no longer a straight line, but a step-like shape. The PRT vector value is:

[0099] if rand <PRT;PRTVector=1;else,PRT Vector=0 (10)

[0100] Here, PRT stands for Perturbation Movement.

[0101] The salient feature of the CBMA algorithm is to change the PRT vector before crossing so that the original trajectory point moves toward the guide point. The crossing process of the CBMA algorithm is to disturb the original trajectory point in the search space to create a set of new moving trajectory points. The new moving position is:

[0102]

[0103] in, is the new move position of the iteration, is the initial trajectory point position in the iteration, is the guide point position in the iteration, and k is the qualitative change step size.

[0104] The fourth step is the constraint function. The distance from the original trajectory point to the new guide point must be less than the moving step length of the ego vehicle. The soft constraint (penalty) method is used to process the moving step length constraint function and introduce the ego vehicle moving trajectory point into a feasible area, which is expressed as follows:

[0105]

[0106]

[0107] Among them, β is used to appropriately scale the constraint function value, σ is used to modify the shape of the optimization plane, and a is a constant. When the value of a is set high enough, it will not affect the trajectory point search process.

[0108] Step 5, CBMA algorithm and cost function evaluation, the number of cost function evaluations during a single trajectory point iteration:

[0109]

[0110] Where WPL is the distance between the initialization trajectory point and the guidance point, and Sd is the interval distance of the trajectory point distribution.

[0111] Each iteration will have a guide point, so there are a total of (total-1) points running in the iteration, and in each iteration, the quality of its cost function is evaluated:

[0112]

[0113] Among them, total-1 is the trajectory point sample (excluding the guide point).

[0114] Furthermore, the total number of cost functions given in all migration processes is

[0115]

[0116] Among them, remove represents the number of iterations of the trajectory point, which means that the trajectory point moves in the plane to find the best solution. If the difference between the guide point and the worst trajectory point is not less than the maximum allowable difference MinDiv, and the maximum number of migrations is reached, the best solution trajectory is located during the search process.

[0117] The sixth step is fitness function. The fitness function is calculated to evaluate the generated obstacle avoidance trajectory points. The fitness function consists of two parts: the attraction element generated by the target and the repulsion element generated by the surrounding obstacle vehicles. The trajectory point with low fitness is selected as the new position to which the vehicle is to move, and then pass through the obstacle gap to reach the target position.

[0118] The CBMA algorithm considers two indicators: the construction of the fitness function and moving to the target point and avoiding obstacles. Moving to the target requires two forces: target attraction T attract and the obstacle repulsion force O repul The attraction and repulsion are adjusted to have appropriate weights. The fitness function consists of two parts: an attraction element generated by the target and a repulsion element generated by the obstacle; the target attraction depends on the distance from the ego vehicle to the target, and the obstacle repulsion elements have three: the distance from the ego vehicle to the obstacle, the obstacle size, and the obstacle influence coefficient C, which determines the gap between the ego vehicle and the obstacle. When the ego vehicle is trapped between the local minimum obstacles, the influence coefficient C will be changed, creating a relative balance between the two components of the force.

[0119]

[0120] Among them, T attract , O repul They represent the balance coefficients of the target gravity and obstacle repulsion, f value represents the fitness value, d target is the distance from the vehicle to the target point, n represents the number of obstacles, c represents the influence coefficient of obstacles, r obs Represents the obstacle radius, the distance from the vehicle to the obstacle.

[0121] Simulation Results

[0122] The simulation was established in Matlab using a map: the ego vehicle was initially placed at (1, 1) m and the target was at (17, 27) m; the CBMA algorithm parameters, sensor activity range, and maximum moving step length of the ego vehicle are shown in Table 3.

[0123] Table 2 Simulation parameters

[0124] Sd remove <![CDATA[r mobilestep ]]> PRT WPL <![CDATA[r sensor ]]> MinDiv 0.21 20 1m 0.1 3 5m 1

[0125] Combined with the above constraints, assuming the same expected lateral displacement y e = 1m, draw the trajectory point curve effect diagram as follows Figure 5 As shown, the fit is good, but there is a certain time lag.

[0126] The above description is only a specific implementation of the present invention, which can be improved according to the above description, and all these improvements within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. An automatic emergency obstacle avoidance control method for an unmanned vehicle, characterized in that: The following steps are involved: The system first determines whether the vehicle can avoid a collision with the vehicle in front by automatic braking. If it is determined that automatic braking can avoid a collision and there is no danger of rear-end collision between the vehicle in front and the vehicle in front, automatic braking is used. If it is determined that automatic braking cannot avoid a collision with the vehicle in front, the system clusters the vehicle and the obstacle vehicle in a circle, uses the consensus behavior migration algorithm, calculates the fitness function based on the planned safe obstacle avoidance trajectory points, and performs automatic emergency steering to avoid obstacles. The specific steps include: Step 1: Collect real-time data from millimeter-wave radar, binocular camera and vehicle-mounted sensors to obtain the longitudinal distance between the vehicle and the front vehicle, the speed of the front vehicle, the lateral distance between the vehicle and the target vehicle in the adjacent lane, the longitudinal distance between the vehicle and the target vehicle, and the geometric size information of the front vehicle; Step 2: Calculate the longitudinal critical headway 𝐿0 by the vehicle speed and the relative speed of the two vehicles, compare the expected headway between the vehicle and the front vehicle with the actual headway, and determine whether the automatic braking function can avoid obstacles; Step 3: Calculate the required safety time distance TTC for emergency obstacle avoidance based on the longitudinal critical headway, introduce the forward collision warning coefficient FCW, confirm that the scene is an emergency, and the system adopts an automatic emergency obstacle avoidance strategy; Step 4: Based on the circular clustering consensus behavior migration method, the vehicle obstacle avoidance trajectory points are initialized, and the original trajectory points of the vehicle are iterated through cross-behavior interference to create a set of new moving trajectory points; Step 5: Use the constraint function to introduce the moving step of the ego vehicle into a distance metric of a feasible area, calculate the fitness function and evaluate the generated new trajectory points, and select the trajectory point with the lowest fitness value as the new position where the ego vehicle will move; The step three comprises: Based on the TTC required for emergency obstacle avoidance, the forward collision warning coefficient FCW is introduced as a prerequisite for taking obstacle avoidance measures. If FCW>0.3, the scene is an emergency and the CBMA algorithm is activated for obstacle avoidance. The fourth step comprises: Initialize around the actual position of the ego vehicle, which contains two parameters on the x-axis and y-axis, define the CBMA parameters and their boundaries, and iteratively select a guide point for the current trajectory point; To ensure the diversity of trajectory points, CBMA defines a trajectory point interference vector PRT. This parameter determines whether the initialized trajectory point will move directly to the next guide point. The random number generated by [0.1] is compared with the defined PRT vector. The parameters of each trajectory point must be randomly selected from the given range [lo,hi]. Initial P=X act +rand[0,1]×(x (hi) −x (lo) ) Among them, P is the initial position of the trajectory point, X act is the actual position of the vehicle, rand[0,1] is a random number between 0 and 1, and x (hi) 、x (lo) is the maximum position range; In CBMA, the intersection behavior simulation of intelligent organisms is referenced to obtain new moving trajectory points through the intersection behavior; each generated trajectory point is evaluated through the cost function, and the fitness function based on the principle of the gravitational force of the lane change target position and the repulsive force of the surrounding vehicles is calculated. For the current iteration, a guide point with the lowest fitness is selected as the position where the vehicle will move until it reaches the target point; The step five comprises: The salient feature of CBMA is that the PRT vector is changed at each intersection so that the distance from the actual position of the ego vehicle to the new guide point must be less than the ego vehicle's moving step length. The soft constraint method is used to process the moving step length constraint function and introduce the ego vehicle's moving trajectory into a feasible region. Calculate the fitness function to evaluate the generated obstacle avoidance trajectory points. The fitness function consists of two parts: the attraction element generated by the target and the repulsion element generated by the front vehicle and the vehicles in the adjacent lane. Select the trajectory point with the lowest fitness as the new position to be moved by the vehicle, and then pass through the obstacle gap to reach the target position; Specifically: the first step, crossover, the PRT vector is performed before the trajectory point performs crossover motion in the search space, and the PRT vector is changed every time it jumps; the trajectory of the trajectory point is no longer a straight line, but a step-shaped one; the PRT vector value is: Among them, PRT stands for perturbation motion; The salient feature of the CBMA algorithm is to change the PRT vector before crossing so that the original trajectory point moves toward the guide point. The crossing process of the CBMA algorithm is to disturb the original trajectory point in the search space to create a set of new moving trajectory points. The new moving position is: in, is the new move position of the iteration, is the initial trajectory point position in the iteration, is the position of the guiding point in the iteration, k is the qualitative change step size; The second step is the constraint function. The distance from the original trajectory point to the new guide point must be less than the moving step length of the ego vehicle. The moving step length constraint function is processed using the soft constraint method to introduce the ego vehicle moving trajectory point into a feasible area, which is expressed as follows: Among them, β is used to appropriately scale the constraint function value, σ is used to modify the shape of the optimization plane, and a is a constant. When the value of a is set high enough, it will not affect the trajectory point search process; The third step is the CBMA algorithm and the cost function evaluation. The number of cost function evaluations during a single trajectory point iteration is: Where WPL is the distance between the initialization trajectory point and the guidance point, and Sd is the interval distance of the trajectory point distribution; Each iteration will have a guide point, so there are a total of (total−1) points running in the iteration, and in each iteration, the quality of its cost function is evaluated: Among them, total−1 is the trajectory point sample, excluding the guide point; Furthermore, the total number of cost functions given in all migration processes is Among them, remove represents the number of iterations of the trajectory point, which means that the trajectory point moves in the plane to find the best solution; if the difference between the guide point and the worst trajectory point is not less than the maximum allowable difference MinDiv, and the maximum number of migrations is reached, the best solution trajectory is located during the search process; The fourth step is the fitness function. The fitness function is calculated to evaluate the generated obstacle avoidance trajectory points. The fitness function consists of two parts: the attraction element generated by the target and the repulsion element generated by the surrounding obstacle vehicles. The trajectory point with low fitness is selected as the new position to be moved by the vehicle, and then the vehicle passes through the obstacle gap and reaches the target position. The CBMA algorithm considers two indicators: the construction of the fitness function and moving to the target point and avoiding obstacles; moving to the target requires two forces: the target attraction T attract and the obstacle repulsion force O repul ; The attraction and repulsion are adjusted to have appropriate weights; The fitness function consists of two parts: an attraction element generated by the target and a repulsion element generated by the obstacle; The target attraction depends on the distance from the vehicle to the target, and the obstacle repulsion elements have three: the distance from the vehicle to the obstacle, the obstacle size, and the obstacle influence coefficient C, which determines the gap between the two obstacles; When the vehicle is trapped between the local minimum obstacles, the influence coefficient C will be changed, creating a relative balance between the two components of the force Among them, T attract , O repul They represent the balance coefficients of the target gravity and obstacle repulsion, 𝑓 value represents the fitness value, d target is the distance from the vehicle to the target point, 𝑛 represents the number of obstacles, c represents the influence coefficient of the obstacles, 𝑟 obs Represents the obstacle radius, the distance from the vehicle to the obstacle.

2. A system for implementing the automatic emergency obstacle avoidance control method for an unmanned vehicle according to claim 1, characterized in that: include: Environmental sensing unit, decision-making unit and control unit; The surrounding sensing unit obtains information about the vehicle and surrounding vehicles through two millimeter-wave radars, a binocular camera and the vehicle's sensors, including the speed, acceleration, distance and geometric size information of the vehicle in front, the vehicle in front of the adjacent lane and the vehicle behind in the adjacent lane; The decision-making unit determines that the vehicle's longitudinal braking cannot avoid the obstacle based on the data uploaded by the ambient sensing unit, and makes an obstacle avoidance decision; A control unit controls the actuator to perform braking / steering actions based on the data analyzed by the decision unit and uploaded by the ambient sensing unit, so as to avoid obstacles of the vehicle in front; the control unit includes a wire-controlled brake and wire-controlled steering system, which automatically brakes according to the obstacle avoidance decision made by the consensus behavior migration algorithm or controls the vehicle to automatically steer according to the planned path; The control device involved in the control unit includes a central processing unit and a memory. The memory has a high-precision map and a computer program. The computer program can execute the automatic emergency obstacle avoidance control method for the unmanned vehicle as described in claim 1.

3. The system for implementing the automatic emergency obstacle avoidance control method for an unmanned vehicle as claimed in claim 2, characterized in that: The two millimeter-wave radars are long-range millimeter-wave radars with a range of 0-300m, an azimuth angle of ±10°, and a pitch angle of 13°. They are installed at the tail fin and the front bumper logo of the car. The two cameras can record information such as the vehicle speed, acceleration, braking operation, steering operation, GPS, and images in front of the vehicle within a period of time before and after the emergency occurs by using pixel width and focal length. They are installed near the central rearview mirror on the inside of the front windshield of the car; The vehicle sensor includes a vehicle speed sensor, a vehicle steering wheel angle sensor and a yaw angle sensor.

Images