Automatic driving lane changing method fusing intention risk field and state machine decision reversal mechanism

Abstract

The invention discloses an automatic driving behavior decision-making method based on an intention-driven risk field. The method comprises the following steps: 1, eliminating geometric distortion by constructing a generalized bounding box; 2, introducing a momentum correction and intention coupling mechanism, and dynamically remodeling a risk field topology to simulate defensive driving psychology; and 3, scoring by using a space-time most disable principle, carrying out decision by combining with a dual risk threshold state machine, and introducing a decision reversal mechanism to break deadlock. According to the method, the problem of driving behavior decision oscillation is effectively solved, and the traffic efficiency and safety in a complex interaction scene are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of autonomous driving technology, specifically relating to an autonomous driving behavior decision-making method based on an intent-driven risk field. Background Technology

[0002] In the layered architecture of autonomous driving, the behavior decision layer plays a crucial role in bridging the gap between the upper and lower layers. However, existing methods still face three major theoretical bottlenecks when dealing with complex dynamic interactions: First, geometric representation distortion leads to collision risks. Since the mass model cannot cover the vehicle's true physical contours, the vehicle is prone to a lack of safety envelope when operating in narrow spaces or interacting closely with large vehicles, increasing the risk of collision. Second, the decision logic lacks defensiveness and human-likeness. Because the system cannot simulate the human psychology of looking before acting, the timing of lane changing is too mechanical, making it difficult to reserve sufficient safety margin in the game interaction, and it is prone to high-frequency oscillation of commands under slight environmental fluctuations. Third, rigid decision-making leads to low traffic efficiency. Traditional state machines often fall into deadlock logic in complex conditions, where once stopped, they must completely back off, failing to flexibly capture dynamically changing traffic windows, severely limiting the vehicle's ability to pass through complex scenarios. Summary of the Invention

[0003] To overcome the shortcomings of existing technologies, such as geometric distortion in environmental representation and lack of characterization of human drivers' defensive cognitive mechanisms, this invention provides an autonomous driving method based on an intent-driven risk field and a decision reversal mechanism. By constructing an asymmetric risk field that conforms to the vehicle's true profile and introducing an intent-driven field strength reshaping mechanism and a dual threshold adjudication logic, this method aims to achieve autonomous driving behavior decision-making that combines geometric accuracy, cognitive anthropomorphism, and decision stability. This effectively solves the problem of oscillations in driving behavior decisions and significantly improves traffic efficiency and safety in complex interactive scenarios.

[0004] To achieve the above-mentioned objectives, the present invention adopts the following technical solution: The autonomous driving behavior decision-making method based on an intent-driven risk field, as described in this invention, is characterized by the following steps: Step 1: Establish a local Cartesian coordinate system with the vehicle's dynamic center as the origin and a two-dimensional coordinate system with the obstacle vehicle's dynamic center as the origin; In a two-dimensional coordinate system, the generalized bounding box distance of the obstacle vehicle relative to the vehicle is calculated, thereby constructing an intent-driven generalized bounding box risk field; Step 2: Introduce a momentum correction factor to construct a longitudinal risk field based on the Sigmoid function, thereby obtaining the longitudinal normalized distance of the vehicle in the two-dimensional coordinate system. Step 3: Introduce an intent-driven correction mechanism and adjust it according to the vehicle's lateral driving intent commands. We construct an intention-driven lateral risk field to obtain the lateral normalized distance in a two-dimensional coordinate system. Step 4: Based on the candidate trajectory set generated by the decision layer The worst-case scenario principle in space and time is used to calculate the planning time domain. Any candidate trajectory Risk score; Step 5: Set the expected risk threshold As a condition for state admission, a maximum acceptable risk threshold is set. As a condition for maintaining the state, the vehicle's position in the planning time domain is determined based on the risk score and using a dual risk threshold. The driving decision-making instructions issued by the vehicle cause the vehicle to switch between different driving states.

[0005] The characteristic of the autonomous driving behavior decision-making method based on intention-driven risk field described in this invention is that step 1 uses equation (1) to construct a generalized bounding box risk field: (1) In equation (1), This represents the position of obstacle vehicle i in the two-dimensional coordinate system at time t. The potential energy of risk generated at the location; This represents the position of vehicle i in the obstacle system at time t in a two-dimensional coordinate system. The generalized bounding box distance at the location, and we have: (2) In equation (2), This represents the longitudinal coordinate of vehicle i (the obstacle) at time t in a two-dimensional coordinate system. longitudinal velocity The generalized bounding box distance below, This represents the longitudinal coordinate of vehicle i (the obstacle) at time t in a two-dimensional coordinate system. lateral velocity The generalized bounding box distance below.

[0006] Furthermore, step 2 includes: Step 2.1: Use equation (3) to obtain the longitudinal position of obstacle vehicle i and the vehicle at time t in the two-dimensional coordinate system. Basic physical distance : (3) In equation (3), Let i be the length of the vehicle with the obstacle. Let t be the longitudinal coordinate of the vehicle in the two-dimensional coordinate system at time t; Step 2.2: Define the momentum correction factor, including: the longitudinal velocity of the vehicle as shown in equation (4). forward momentum parameters and the backward buffer parameter shown in equation (5) : (4) In equation (4), This is the momentum gain coefficient. This is the minimum safe distance factor; Step 2.3: Use equation (5) to obtain the longitudinal coordinate of the vehicle at time t. longitudinal velocity Dynamic longitudinal shape parameters : (5) In equation (5), Vertical axis The standard Sigmoid function, It is a smoothing factor; Step 2.4: Use equation (6) to obtain the longitudinal coordinates of the vehicle in the two-dimensional coordinate system at time t. longitudinal velocity The vertical normalized distance below : (6).

[0007] Furthermore, step 3 includes: Step 3.1: Use equation (7) to obtain the lateral positions of obstacle vehicle i and the vehicle in the two-dimensional coordinate system. Basic physical distance : (7) In equation (7), The width of obstacle vehicle i; Step 3.2: Define the intention-driven correction mechanism using equations (8) and (9), including: the lateral position y of the vehicle in the two-dimensional coordinate system and the lateral velocity. The physical layer velocity correction term The vehicle's lateral position y in a two-dimensional coordinate system and the driver's intention command Cognitive layer intention modification item : (8) (9) In equations (8) and (9), Let be the lateral velocity of the vehicle relative to obstacle vehicle i in a two-dimensional coordinate system. This is the gain coefficient; The cognitive inflation coefficient, Let be the orientation symbol of the vehicle relative to the center of obstacle vehicle i. If the vehicle is to the left of obstacle vehicle i, then let . =+1, if the vehicle is to the right of obstacle vehicle i, then let =-1, This indicates the lateral position of the vehicle in the Cartesian coordinate system. Let represent the position of the obstacle vehicle in the Cartesian coordinate system. If the obstacle vehicle is to the left of the vehicle, then let . =-1, otherwise, let =+1, Let be the cognitive shearing indicator function, and we have: like and Then let =1; like and Then let =1; Otherwise, let =0; Step 3.3: Use equation (11) to obtain the lateral coordinate y and lateral velocity of the vehicle in the two-dimensional coordinate system. Dynamic lateral shape parameters : (10) In equation (10), Basic risk width; Step 3.4: Obtain the driving intention command using equation (11) The lateral coordinate y and lateral velocity of the vehicle in the two-dimensional coordinate system. Horizontal normalized distance : (11).

[0008] Furthermore, in step 4, candidate trajectories are obtained using equation (12). Final risk score : (12) In equation (13), This indicates that the vehicle is on the candidate trajectory at time t. The position above; The generalized bounding box risk field represents the obstacle vehicle i. Let t represent the set of obstacle vehicles surrounding the vehicle at time t, where t∈[0,T].

[0009] Furthermore, step 5 includes: Step 5.1: Define the vehicle's driving status, including: following status. Lane change execution status Lane change stopped ; Step 5.2: Set the expected risk threshold as follows. Set the maximum acceptable risk threshold as ; Step 5.3, if < Then the jump instruction is output, making the planning time domain... The car is in a following state Jump to lane change execution state ;like > Then the termination instruction is output, causing the planning time domain to be terminated. The internal bicycle is in lane change execution state Jump to lane change termination status ; If the jump reaches the lane change termination state During the process < If the output is positive, a return command will be sent, causing the vehicle to exit the lane-changing abort state. Jump back to the lane change execution state ; If the lateral position deviation of the vehicle converges to the allowable range of the target lane during the lane change process, it indicates that the lane change is complete, and a follow command is output, causing the vehicle's driving state to switch to follow mode. ; If the vehicle completely returns to the center line of the original lane during the lane change abort, it indicates that the hazard avoidance is complete, and a reset command is output, resetting the vehicle's driving status to follow-the-car mode. .

[0010] The present invention provides an electronic device, including a memory and a processor, characterized in that the memory is used to store a program supporting the processor in performing the method described therein, and the processor is configured to execute the program stored in the memory.

[0011] The present invention discloses a computer-readable storage medium storing a computer program, characterized in that the computer program is executed by a processor to perform the steps of the method described thereon.

[0012] Compared with existing technologies, the beneficial technical effects of this invention are reflected in: 1. This invention uses generalized bounding box distance to replace traditional Euclidean distance, eliminating geometric distortion of the mass model during close-range interaction, ensuring that the risk field can cover the real physical contour of the vehicle, improving the geometric accuracy of environmental representation, and effectively solving the problem of missing safety envelope in narrow spaces or around large vehicles.

[0013] 2. The intent-field strong coupling mechanism proposed in this invention can simulate the "defensive attention" psychology (i.e., "look before moving") of a human driver when generating the intention to change lanes. By subjectively amplifying the risk perception of the target area, the system is forced to trigger lane changes only when the safety buffer is significantly sufficient, enhancing the anthropomorphism and defensiveness of the decision-making, thereby significantly improving the safety and ride comfort of autonomous vehicles in complex interaction scenarios.

[0014] 3. The dual risk threshold mechanism established in this invention forms a decision lag interval, effectively eliminating high-frequency oscillations in decision instructions caused by minor environmental fluctuations (Zeno's phenomenon). Simultaneously, the unique decision reversal mechanism breaks the rigid logic of traditional state machines that "must completely revert once stopped," endowing vehicles with the ability to flexibly capture fleeting time windows in game-theoretic interactions, significantly improving traffic efficiency and ensuring the stability and efficiency of decision-making. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the overall process of the present invention; Figure 2 This is a schematic diagram of the vehicle dynamics coordinate system and the generalized bounding box risk field representation of the present invention; Figure 3 This is a schematic diagram of the longitudinal asymmetric risk field distribution based on momentum-driven principles, according to the present invention. Figure 4 This is a schematic diagram of the cognitive elastic deformation of the lateral risk field based on driving intention in this invention; Figure 5 This is a schematic diagram of the hysteresis discrimination mechanism based on dual risk thresholds in this invention; Figure 6 This is a schematic diagram of the dynamic decision-making and flow logic of the finite state machine of the present invention. Detailed Implementation

[0016] In this embodiment, an autonomous driving behavior decision-making method based on an intent-driven risk field and decision reversal mechanism introduces a generalized bounding box and asymmetric potential energy at the physical level, and an "intent-field strength" coupling mechanism at the cognitive level. Combined with a finite state machine (FSM) based on dual thresholds, it achieves a general decision-making framework that combines geometric accuracy and cognitive anthropomorphism. Specifically, as... Figure 1 As shown, the method includes the following steps: Step 1: First, establish a local Cartesian coordinate system with the vehicle's dynamic center as the origin. Following the ISO 8855 vehicle dynamics standard, define the longitudinal extension direction of the vehicle's front as... The axis is positive, perpendicular to the vehicle body and pointing to the left. Positive axis. Acquiring vehicle status information: Through the perception and positioning modules of the autonomous driving system, real-time motion status information of the vehicle and surrounding obstacle vehicles is obtained, including the vehicle's geometric position. Heading angle Longitudinal velocity and lateral speed Meanwhile, it obtains the vehicle's driving intention commands from the upper-level behavior planning module. (such as changing lanes to the left, changing lanes to the right, or keeping lanes open). For example... Figure 2 As shown, in this reference frame, a generalized bounding box distance is introduced. As a geometric benchmark for field strength attenuation, the risk field function is constructed using a normalized reciprocal form, where the first... The potential energy of the obstacle vehicle at its own location : (1) In equation (1), This represents the position of obstacle vehicle i in the two-dimensional coordinate system at time t. The potential energy of risk generated at the location; This represents the position of vehicle i in the obstacle system at time t in a two-dimensional coordinate system. The generalized bounding box distance at the location, and we have: (2) In equation (2), This represents the longitudinal coordinate of vehicle i (the obstacle) at time t in a two-dimensional coordinate system. longitudinal velocity The generalized bounding box distance below, This represents the longitudinal coordinate of vehicle i (the obstacle) at time t in a two-dimensional coordinate system. lateral velocity The generalized bounding box distance below.

[0017] At the same time, it is important to note the risk potential. Satisfying boundedness and convergence, when the vehicle interacts with an obstacle... Approaching infinity At that time, risk value Characterized as obstacles In a state of absolute safety; when the vehicle touches or intrudes into the interior of the obstacle enclosure. At that time, the risk value strictly converges to its maximum value. Characterized by the vehicle and obstacles It is in a critical collision state. Unlike the isotropic characteristic of Euclidean distance, The growth rate is decoupled in the longitudinal and lateral directions. Subsequently, by introducing shape parameters, the distance field can exhibit a longer extension in the frontal direction while remaining compact in the lateral direction, thus accurately reproducing the dynamic risk distribution characteristics of the vehicle at high speeds.

[0018] Step 2: Construct a longitudinal asymmetric risk field. Specifically, to reflect the dynamic influence of velocity on the risk field distribution, this embodiment introduces a momentum correction factor and synthesizes forward momentum parameters through smoothing. With back buffer parameters To reshape the vertical risk distribution, a schematic diagram of the vertical asymmetric risk field distribution is shown below. Figure 3 As shown, the specific steps are as follows: Step 2.1: Use equation (3) to obtain the longitudinal position of obstacle vehicle i and the vehicle at time t in the two-dimensional coordinate system. Basic physical distance : (3) In equation (3), Let i be the length of the vehicle with the obstacle. Let t be the longitudinal coordinate of the vehicle in the two-dimensional coordinate system at time t; Step 2.2: Define the momentum correction factor, including: the longitudinal velocity of the vehicle as shown in equation (4). forward momentum parameters and the backward buffer parameter shown in equation (5) : (4) In equation (4), where This is the momentum gain coefficient. This is the minimum safe distance coefficient.

[0019] Step 2.3: Use equation (5) to obtain the longitudinal coordinate of the vehicle at time t. longitudinal velocity Dynamic longitudinal shape parameters : (5) In equation (5), For the standard Sigmoid function, A smoothing factor controls the steepness of the transition between the foreground and background. Dynamic adjustment of momentum-driven parameters. Its core lies in its adaptive response to the vehicle's motion state: (1) Forward momentum parameter : with relative longitudinal velocity The correlation is positive, which represents the dynamic characteristic that "the faster the speed, the longer the braking distance, and the farther the forward risk radiation."

[0020] (2) Back buffer parameters Set as a constant The purpose is to maintain a basic rearward safety buffer and prevent following vehicles from getting too close.

[0021] Step 2.4: Use equation (6) to obtain the longitudinal coordinates of the vehicle in the two-dimensional coordinate system at time t. longitudinal velocity The vertical normalized distance below : (6) Step 3: In this embodiment, step 3 dynamically reshapes the topology of the lateral risk field by introducing driving intention commands to simulate the "defensive attention" psychology of a human driver when generating a lane-changing intention. Specifically, this process achieves asymmetric elastic reshaping of the risk distribution of surrounding obstacles through the product coupling of the physical layer speed correction term and the cognitive layer intention correction term. A schematic diagram of the cognitive elastic deformation of the risk field is shown below. Figure 4 As shown, the specific steps are as follows: Step 3.1: Use equation (7) to obtain the lateral positions of obstacle vehicle i and the vehicle in the two-dimensional coordinate system. Basic physical distance : (7) In equation (7), The width of obstacle vehicle i; Step 3.2: Define the intention-driven correction mechanism using equations (8) and (9), including: the lateral position y of the vehicle in the two-dimensional coordinate system and the lateral velocity. The physical layer velocity correction term The vehicle's lateral position y in a two-dimensional coordinate system and the driver's intention command Cognitive layer intention modification item : (8) (9) In equation (8), the physical layer velocity correction term This characterizes the risk gain at the lateral physical level. When an obstacle has a lateral velocity toward the vehicle (i.e., lateral approach), the field width passively increases to reserve more avoidance space; Let be the lateral velocity of the vehicle relative to obstacle vehicle i in a two-dimensional coordinate system. As a gain factor, this item ensures the physical avoidance capability against vehicles that "cut in" or "cross".

[0022] In equation (9), the cognitive layer intention modification term Characterize the risk bias driven by intent. Let the driving intent command output by the intent module be... Where -1 represents the vehicle's intention to change lanes to the left, 0 represents the vehicle's intention to maintain its lane, and 1 represents the vehicle's intention to change lanes to the right. The cognitive inflation coefficient, Let be the orientation symbol of the vehicle relative to the center of obstacle vehicle i. If the vehicle is to the left of obstacle vehicle i, then let . =+1, if the vehicle is to the right of obstacle vehicle i, then let =-1, This indicates the lateral position of the vehicle in the Cartesian coordinate system. Let represent the position of the obstacle vehicle in the Cartesian coordinate system. If the obstacle vehicle is to the left of the vehicle, then let . =-1, otherwise, let =+1, Let be the cognitive shearing indicator function, and we have: like and ,but =1; like and ,but =1; If neither of the above two conditions is met, then =0.

[0023] Specifically, this manifests in two typical forms of heterogeneity: Scenario 1: Target-side spatial compression. This is activated when an obstacle is located in the target lane (such as the left-hand vehicle during a left lane change) and the vehicle is inside it. This causes the obstacle's risk field to expand asymmetrically towards the vehicle, geometrically compressing the physical gap of the lane change target location.

[0024] Scenario 2: Obstruction of the path. When an obstacle is located in the current or non-target lane (such as the vehicle in front when changing lanes to the left) and the vehicle attempts to cut out from its left, Activation. This causes the risk field to extend towards the target lane, covering potential cut-out paths and creating a "geometric blockade" against blind lane changes.

[0025] This anisotropic "cognitive bulge" is mathematically equivalent to virtually compressing the feasible solution space of the autonomous vehicle. It forces the decision-making level to face more stringent risk constraints than the objective physical environment when evaluating the entry conditions for state transitions (such as from following to lane changing). This design ensures that autonomous vehicles will only trigger lane changing when there is a significantly ample safety buffer in the target area, thus perfectly replicating the defensive driving intuition of human drivers at the algorithm level: "look before moving and be cautious when changing lanes."

[0026] Step 3.3: Use equation (11) to obtain the lateral coordinate y and lateral velocity of the vehicle in the two-dimensional coordinate system. Dynamic lateral shape parameters : (10) In equation (10), Based on the breadth of basic risk.

[0027] Step 3.3: Obtain the driving intention command using equation (11) The lateral coordinate y and lateral velocity of the vehicle in the two-dimensional coordinate system. Horizontal normalized distance : (11) Step 4: Calculate the risk score of the candidate trajectories. After completing the dynamic modeling of the horizontal and vertical risk fields, this embodiment uses the candidate trajectory set generated by the decision layer. The safety of each candidate trajectory is quantitatively evaluated using the "spatiotemporal worst-case scenario principle." Specifically, any candidate trajectory within the planning time domain T is selected. The maximum combined risk field strength experienced by all path points is used as the final risk score for that trajectory. The calculation formula is as follows: (12) In equation (12), This indicates that the vehicle is on the candidate trajectory at time t. The position above; This represents the risk field of the generalized enclosing box. Let t represent the set of obstacle vehicles surrounding the vehicle at time t, where t∈[0,T].

[0028] It should be emphasized that the shape parameters of the risk field are controlled in real time by the output state of the intent module. This means that the decision risk score of the same physical trajectory is different under different intents, which reflects a more perfect evaluation logic.

[0029] Step 5: Establish a finite state machine based on dual risk thresholds and set the expected risk threshold. As a condition for state admission, a maximum acceptable risk threshold is set. As a condition for maintaining the state, the setting of the dual risk threshold is as follows: Figure 5 As shown, the risk score is input into the finite state machine to determine the vehicle's following state. Lane change execution status Or lane change interruption status State transitions. The state transition logic is as follows: Figure 6 As shown, the specific steps are as follows: Step 5.1: Define the vehicle's driving status, including: following status. Lane change execution status Lane change stopped ; Step 5.2: Set the expected risk threshold As a condition for state admission, a maximum acceptable risk threshold is set. As a condition for maintaining the state.

[0030] Step 5.3: Based on the current environment, perform risk simulation on the candidate trajectory set for the target lane. < Then the jump instruction is output, making the planning time domain... The car is in a following state Jump to lane change execution state ;like > Then the termination instruction is output, causing the planning time domain to be terminated. The internal bicycle is in lane change execution state Jump to lane change termination status ; During the lane change process, the safety constraint level is adjusted from the expected risk threshold. Switch to the maximum acceptable risk threshold If the jump is to the lane change termination state During the process < If the output is positive, a return command will be sent, causing the vehicle to exit the lane-changing abort state. Jump back to the lane change execution state ; If the lateral position deviation of the vehicle converges to the allowable range of the target lane during the lane change process, it indicates that the lane change is complete, and a follow command is output, causing the vehicle's driving state to switch to follow mode. ; If the vehicle completely returns to the center line of the original lane during the lane change abort, it indicates that the hazard avoidance is complete, and a reset command is output, resetting the vehicle's driving status to follow-the-car mode. .

[0031] In this embodiment, an electronic device includes a memory and a processor. The memory stores a program that supports the processor in executing the above-described method, and the processor is configured to execute the program stored in the memory.

[0032] In this embodiment, a computer-readable storage medium stores a computer program, which is executed by a processor to perform the steps of the above method.

Claims

1. An autonomous driving behavior decision-making method based on an intent-driven risk field, characterized in that, Includes the following steps: Step 1: Establish a local Cartesian coordinate system with the vehicle's dynamic center as the origin and a two-dimensional coordinate system with the obstacle vehicle's dynamic center as the origin; In a two-dimensional coordinate system, the generalized bounding box distance of the obstacle vehicle relative to the vehicle is calculated, thereby constructing an intent-driven generalized bounding box risk field; Step 2: Introduce a momentum correction factor to construct a longitudinal risk field based on the Sigmoid function, thereby obtaining the longitudinal normalized distance of the vehicle in the two-dimensional coordinate system. Step 3: Introduce an intent-driven correction mechanism and adjust it according to the vehicle's lateral driving intent commands. We construct an intention-driven lateral risk field to obtain the lateral normalized distance in a two-dimensional coordinate system. Step 4: Based on the candidate trajectory set generated by the decision layer The worst-case scenario principle in space and time is used to calculate the planning time domain. Any candidate trajectory Risk score; Step 5: Set the expected risk threshold As a condition for state admission, a maximum acceptable risk threshold is set. As a condition for maintaining the state, the vehicle's position in the planning time domain is determined based on the risk score and using a dual risk threshold. The driving decision-making instructions issued by the vehicle cause the vehicle to switch between different driving states.

2. The autonomous driving behavior decision-making method based on an intent-driven risk field according to claim 1, characterized in that, In step 1, the generalized bounding box risk field is constructed using equation (1): (1) In equation (1), This represents the position of obstacle vehicle i in the two-dimensional coordinate system at time t. The potential energy of risk generated at the location; This represents the position of vehicle i in the obstacle system at time t in a two-dimensional coordinate system. The generalized bounding box distance at the location, and we have: (2) In equation (2), This represents the longitudinal coordinate of vehicle i (the obstacle) at time t in a two-dimensional coordinate system. longitudinal velocity The generalized bounding box distance below, This represents the longitudinal coordinate of vehicle i (the obstacle) at time t in a two-dimensional coordinate system. lateral velocity The generalized bounding box distance.

3. The autonomous driving behavior decision-making method based on an intent-driven risk field according to claim 2, characterized in that, Step 2 includes: Step 2.1: Use equation (3) to obtain the longitudinal position of obstacle vehicle i and the vehicle at time t in the two-dimensional coordinate system. Basic physical distance : (3) In equation (3), Let i be the length of the vehicle with the obstacle. Let t be the longitudinal coordinate of the vehicle in the two-dimensional coordinate system at time t; Step 2.2: Define the momentum correction factor, including: the longitudinal velocity of the vehicle as shown in equation (4). forward momentum parameters and the backward buffer parameter shown in equation (5) : (4) In equation (4), This is the momentum gain coefficient. This is the minimum safe distance factor; Step 2.3: Use equation (5) to obtain the longitudinal coordinate of the vehicle at time t. longitudinal velocity Dynamic longitudinal shape parameters : (5) In equation (5), Vertical axis The standard Sigmoid function, It is a smoothing factor; Step 2.4: Use equation (6) to obtain the longitudinal coordinates of the vehicle in the two-dimensional coordinate system at time t. longitudinal velocity The vertical normalized distance below : (6)。 4. The autonomous driving behavior decision-making method based on an intent-driven risk field according to claim 3, characterized in that, Step 3 includes: Step 3.1: Use equation (7) to obtain the lateral positions of obstacle vehicle i and the vehicle in the two-dimensional coordinate system. Basic physical distance : (7) In equation (7), The width of obstacle vehicle i; Step 3.2: Define the intention-driven correction mechanism using equations (8) and (9), including: the lateral position y of the vehicle in the two-dimensional coordinate system and the lateral velocity. The physical layer velocity correction term The vehicle's lateral position y in a two-dimensional coordinate system and the driver's intention command Cognitive layer intention modification item : (8) (9) In equations (8) and (9), Let be the lateral velocity of the vehicle relative to obstacle vehicle i in a two-dimensional coordinate system. This is the gain coefficient; The cognitive inflation coefficient, Let be the orientation symbol of the vehicle relative to the center of obstacle vehicle i. If the vehicle is to the left of obstacle vehicle i, then let . =+1, if the vehicle is to the right of obstacle vehicle i, then let =-1, This indicates the lateral position of the vehicle in the Cartesian coordinate system. Let represent the position of the obstacle vehicle in the Cartesian coordinate system. If the obstacle vehicle is to the left of the vehicle, then let . =-1, otherwise, let =+1, Let be the cognitive shearing indicator function, and we have: like and Then let =1; like and Then let =1; Otherwise, let =0; Step 3.3: Use equation (11) to obtain the lateral coordinate y and lateral velocity of the vehicle in the two-dimensional coordinate system. Dynamic lateral shape parameters : (10) In equation (10), Basic risk width; Step 3.4: Obtain the driving intention command using equation (11) The lateral coordinate y and lateral velocity of the vehicle in the two-dimensional coordinate system. Horizontal normalized distance : (11)。 5. The autonomous driving behavior decision-making method based on an intent-driven risk field according to claim 1, characterized in that, In step 4, candidate trajectories are obtained using equation (12). Final risk score : (12) In equation (13), This indicates that the vehicle is on the candidate trajectory at time t. The position above; The generalized bounding box risk field represents the obstacle vehicle i. Let t represent the set of obstacle vehicles surrounding the vehicle at time t, where t∈[0,T].

6. The autonomous driving behavior decision-making method based on an intent-driven risk field according to claim 5, characterized in that, Step 5 includes: Step 5.1: Define the vehicle's driving status, including: following status. Lane change execution status Lane change stopped ; Step 5.2: Set the expected risk threshold as follows. Set the maximum acceptable risk threshold as ; Step 5.3, if < Then the jump instruction is output, making the planning time domain... The car is in a following state Jump to lane change execution state ;like > Then the termination instruction is output, causing the planning time domain to be terminated. The internal bicycle is in lane change execution state Jump to lane change termination status ; If the jump reaches the lane change termination state During the process < If the output is positive, a return command will be sent, causing the vehicle to exit the lane-changing abort state. Jump back to the lane change execution state ; If the lateral position deviation of the vehicle converges to the allowable range of the target lane during the lane change process, it indicates that the lane change is complete, and a follow command is output, causing the vehicle's driving state to switch to follow mode. ; If the vehicle completely returns to the center line of the original lane during the lane change abort, it indicates that the hazard avoidance is complete, and a reset command is output, resetting the vehicle's driving status to follow-the-car mode. .

7. An electronic device, comprising a memory and a processor, characterized in that, The memory is used to store a program that supports a processor in executing the method of any one of claims 1-6, the processor being configured to execute the program stored in the memory.

8. A computer-readable storage medium storing a computer program thereon, characterized in that, The computer program is executed by a processor to perform the steps of the method according to any one of claims 1-6.

Images