CN-122009244-A - Intelligent vehicle lane change decision and track planning method and system integrating subjective risk perception and personalized driving style

CN122009244ACN 122009244 ACN122009244 ACN 122009244ACN-122009244-A

Abstract

The invention provides an intelligent vehicle lane change decision and track planning method integrating subjective risk perception and personalized driving style, and belongs to the technical field of intelligent driving. The method realizes the deep fusion of subjective risk perception and personalized style through multi-mode data acquisition and preprocessing, subjective risk perception quantitative modeling based on physiological signals, personalized driving style modeling based on Gaussian mixture models and LSTM, game theory fusion decision, track correction under the restraint of a safety field and integrated output. The invention solves the technical problems of unquantified subjective risk, rough style modeling and separation of decision and planning in the prior art, and remarkably improves humanization, safety and individuation adaptability of lane changing behavior.

Inventors

NI JIE
JIANG DUO
WANG YU
GUO YIWEN
DONG FEI

Assignees

江苏大学

Dates

Publication Date: 20260512
Application Date: 20260330

Claims (9)

1. An intelligent vehicle lane change decision and track planning method integrating subjective risk perception and personalized driving style is characterized by comprising the following steps: Collecting physiological signals, eye movement data, vehicle kinematic data and surrounding traffic environment data of a driver, calibrating the start and stop time of lane change, intercepting time sequence fragments before and after lane change, preprocessing multi-source data, and constructing a standardized lane change sample library; the subjective risk perception quantitative modeling of the driver comprises the steps of constructing a collision risk function based on a fault tree analysis principle, and calculating real-time collision risk values of the target vehicle and all surrounding interactive vehicles Taking the maximum value as a subjective risk perception quantization index Constructing a double-layer convolutional neural network-long-short-term memory network model, taking physiological signals and eye movement data as input, and subjective risk perception quantization indexes Training the output to obtain a mapping model from physiological reaction to subjective risk perception, and calibrating the risk threshold of the individual driver offline by using the mapping model ; Modeling individual driving styles, namely clustering by adopting a Gaussian mixture model based on a lane change transverse speed curve to obtain various typical driving styles, fitting transverse speed characteristics of individual drivers in a weighted combination mode, and determining weight coefficients tau c of each style; The game theory fusion decision process is used for modeling a lane change decision process as a dynamic game between an intelligent vehicle and a rear vehicle of a target lane, setting strategy sets of the two parties, and constructing a game gain function comprising safety, efficiency and comfort, wherein a safety index is introduced into a real-time subjective risk perception quantization index The obtained style weight coefficient tau c is utilized to carry out personalized adjustment on the weight of the profit function, and the game balance is solved to obtain a channel change decision; track modification under the constraint of a safety field, namely constructing a driving safety field comprising a road potential energy field, a kinetic energy field and a behavior field, wherein the behavior field is based on an individual risk threshold value Real-time subjective risk perception quantification index The generated individualized lane change track is placed in a safety field to calculate a risk accumulation value, if the risk exceeds a preset safety threshold, the track is optimized and corrected under the kinematic constraint with the aim of minimizing track deviation; And outputting the final lane change decision and the corrected track to a vehicle control system for execution.
2. The method for intelligent vehicle lane change decision and trajectory planning integrating subjective risk perception and personalized driving style according to claim 1, wherein the subjective risk perception quantization index of the target vehicle The definition is as follows: , Wherein: representing a risk of collision between the target vehicle and the i-th interaction vehicle, ; Wherein, TTC is collision time, tau is time standardization constant, D is actual vehicle distance, sigma D is distance standardization parameter, DDS is minimum safety distance, expressed as: , Wherein t r is the reaction time of the driver, v f 、v i is the speed of the front car and the rear car respectively, mu f 、μ i is the adhesion coefficient of the road surface where the front car and the rear car are located respectively, and g is the gravity acceleration.
3. The method for intelligent vehicle lane change decision and trajectory planning integrating subjective risk perception and personalized driving style according to claim 1, wherein risk threshold values of individual drivers are calibrated offline The calibration method of (1) comprises the following steps: physically calculating average risk of each lane change sample Comparing the high risk or low risk sample with a preset threshold value, and marking the high risk or low risk sample as a real label; average risk for model predictions Traversing candidate thresholds Will be The sample of the driver is judged to be high risk, the classification accuracy is calculated by comparing with the real label, and the threshold with the highest accuracy is taken as the driver 。
4. The intelligent vehicle lane change decision and trajectory planning method integrating subjective risk perception and personalized driving style according to claim 1, wherein the personalized driving style modeling specifically comprises: Carrying out Gaussian fitting on a transverse speed curve of each lane change sample to obtain fitting parameters [ mu, sigma ], [ mu, sigma ] of all samples by adopting a K-means algorithm to obtain a Gaussian model G c (t,p c ,q c of a c-th typical driving style, wherein p c is a mean value and q c is a standard deviation; The lateral speed profile of an individual driver is expressed as: , wherein C is the total number of typical driving styles, τ c is the weight coefficient of the C-th typical driving style, satisfies And solving by minimizing the square Euclidean distance after the channel change track dynamic time warping.
5. The intelligent vehicle lane change decision and trajectory planning method integrating subjective risk perception and personalized driving style according to claim 1, wherein the game benefit function is: , Wherein Deltav is the speed gain obtained after channel change, deltav ref is the reference speed difference, jerk is the longitudinal impact, jerk ref is the reference impact, and the weight coefficient w j is determined according to the following formula: , Wherein the method comprises the steps of The method comprises the steps of setting preset weights corresponding to the c type styles, wherein j=1 corresponds to the safety weight, j=2 corresponds to the efficiency weight, and j=3 corresponds to the comfort weight; The vehicle policy set comprises { execute lane change, give up lane change, delay lane change }, wherein the implementation mechanism of the delay lane change policy is that the initial delay time is set as T d , the online evaluation is re-executed after the delay is finished, if the delay decision is obtained again, the delay time is accumulated, and when the accumulated delay exceeds the preset upper limit or the target lane, the vehicle collision risk is increased Exceeding individual threshold And when the channel is changed, the channel is abandoned.
6. The intelligent vehicle lane change decision and trajectory planning method integrating subjective risk perception and personalized driving style according to claim 1, wherein the driving safety field is defined as: , Wherein E R is a road potential energy field, E V is a kinetic energy field, E D is a behavior field, and the kinetic energy field is obtained by summing contributions of all surrounding vehicles: , , Wherein Deltax j 、Δy j is the longitudinal and transverse distance between the vehicle and the jth vehicle respectively, sigma x is the longitudinal influence range, sigma y is the transverse influence range; the behavior field is modified as follows: , In the formula, Is the individual risk threshold, dimensionless, β is the amplification factor, γ is the sensitivity factor, and the tanh function maps the input to the (0, 1) interval.
7. The intelligent vehicle lane change decision and trajectory planning method integrating subjective risk perception and personalized driving style according to claim 1, wherein the risk accumulation value I e is calculated as follows: The personalized track is discretized into time sequence points (x k ,y k ),k=1,…,N e , each point corresponds to a discrete moment and is represented by t k ; Risk field total value of own vehicle and all surrounding vehicles at time t k Wherein: a kinetic energy field generated at time t k for a jth surrounding vehicle; is the behavioral field at time t k ; calculating risk cumulative value ; Normalized to obtain the safety cost Where T e is the lane change duration and Deltat k is the sampling interval.
8. The method for intelligent vehicle lane change decision and trajectory planning integrating subjective risk perception and personalized driving style according to claim 1, wherein if the risk exceeds a preset safety threshold, the trajectory is optimized and corrected under kinematic constraint with the aim of minimizing the trajectory deviation, specifically as follows: Defining a security likelihood: wherein alpha is a safety sensitivity coefficient; If L e is smaller than the safety likelihood threshold L min , judging that the risk exceeds a preset safety threshold, and starting track correction; the optimization problem of the track correction is expressed as: , the constraint conditions include: Safety restraint ; Curvature constraint ; Acceleration constraint ; Impact degree constraint ; Wherein the method comprises the steps of As the point of the original trajectory, For the track point to be optimized, w k is the time weight of the kth track point, and the time weight of the kth track point satisfies the following conditions 。
9. A planning system, comprising: a memory storing computer executable instructions; a processor configured to execute the computer-executable instructions, which when executed by the processor, implement the intelligent vehicle lane change decision and trajectory planning method of any one of claims 1-8 that fuses subjective risk awareness and personalized driving styles.

Description

Intelligent vehicle lane change decision and track planning method and system integrating subjective risk perception and personalized driving style Technical Field The invention relates to the field of automatic driving, in particular to an intelligent vehicle lane change decision and track planning method and system integrating subjective risk perception and personalized driving style. Background With the rapid development of automatic driving technology, lane changing behavior of intelligent vehicles in complex traffic environments has become a research hotspot in academia and industry. The lane change decision and the track planning are used as core functional modules of the automatic driving system, and are directly related to driving safety, traffic efficiency and riding comfort. The current mainstream lane change decision algorithm is mostly based on objective physical quantity for risk assessment, for example, adopting indexes such as collision time, minimum safety distance or deterministic safety margin. The method can ensure driving safety to a certain extent through accurate mathematical modeling and sensor measurement, and obtains better application effect in a structured road scene. However, there are still several technical drawbacks that are difficult to overcome in the prior art. First, the objective physical quantity evaluation method ignores subjective perception differences of the driver on risks in the actual driving process. A large number of researches show that the risk feelings of different drivers on the same traffic scene often have obvious differences, and the differences are caused by the comprehensive influence of multiple factors such as physiological characteristics, driving experience, psychological states and the like of individuals. A completely safe lane change behavior in physical measurement is perceived as dangerous or uncomfortable by a driver in principle, so that the trust degree and the use willingness of the intelligent driving system are reduced. In other words, the prior art lacks effective quantification and modeling means for subjective risk perception of the driver, resulting in a significant deviation between lane change behavior of the intelligent vehicle and psychological expectations of the driver. Secondly, although the existing lane change trajectory planning method has been focused on driving habit differences, most of the lane change trajectory planning methods still remain on the level of style classification or imitation learning. It is common practice to simply divide the driver into several discrete types of aggressive, general, conservative, etc. by cluster analysis, and then generate a standardized lane change trajectory based on the classification result. The processing method has two fundamental problems that firstly, discrete classification cannot capture dynamic continuous changes of individual styles under different traffic scenes, and secondly, the track generated by the fixed template is difficult to adapt to real-time interaction requirements under complex and mixed driving environments. More importantly, there is an inherent correlation between driving style and risk perception, but the existing methods fail to incorporate both into a unified modeling framework, resulting in a generated trajectory that is either redundant in safety and insufficiently anthropomorphic, or is outstanding in individuality but difficult to guarantee in safety. Again, existing intelligent driving systems typically handle lane change decisions and trajectory planning as two separate modules. The decision module only outputs the lane change intention (yes or no), and the planning module independently generates an execution track. This split architecture results in a lack of collaborative optimization of the two, the decision process does not take into account the availability of subsequent trajectories, nor does the planning process have feedback correction of the decision. In a dynamic interactive traffic environment, the processing mode of the splitting is easy to generate contradiction situations of conservation of decisions and track excitation or infeasibility of the tracks, and the humanization and consistency of the overall behaviors are difficult to meet the actual requirements of man-machine co-driving. In addition, most of existing risk assessment models are judged based on a single physical quantity threshold value, and consideration of a risk accumulation effect in a multi-vehicle interaction scene is lacking. In the course of lane changing, the target vehicle needs to pay attention to multiple interactive objects such as the front vehicle of the original lane, the front vehicle of the target lane, the rear vehicle of the target lane and the like, and the risk contributions of different objects can be mutually overlapped or offset. The existing method only considers the most dangerous single car risks or aggregates multiple car risks in a simple weigh