An autonomous vehicle merging method based on candidate gap and lane potential function

Abstract

The application discloses a kind of based on candidate gap and lane potential function's automatic driving vehicle lane-changing method, comprising: constructing lane-changing scene unified state representation quantity;Generation and prediction candidate lane-changing gap, and then form candidate gap set;Construct candidate gap lane-changing potential function, obtain the lane-changing potential function value of each candidate gap;Feasible candidate gap set is constructed;Select target gap and generate longitudinal and lateral coupling lane-changing trajectory, select the gap of candidate gap lane-changing potential function value minimum as target gap, and generate longitudinal and lateral coupling lane-changing trajectory;Perform closed-loop control, reselect gap and complete determination, output vehicle control instruction, and when target gap is invalid, risk increases, is not accessible or target gap potential function increases more than preset relaxation, re-evaluate candidate gap and target gap selection.The application can realize the vehicle lane-changing of considering safety, efficiency, main road vehicle comfort and complex condition perception.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and more specifically to a lane-merging method for autonomous vehicles based on a candidate gap lane-merging potential function. Background Technology

[0002] Lane merging on highway ramps is a typical high-risk, highly interactive scenario in autonomous driving. The vehicle must adjust its speed and change lanes laterally within a limited acceleration lane, identify available gaps in the target lane, and simultaneously engage in safe and reasonable interactions with the vehicle in front, behind, and adjacent to it. This process is influenced by multiple constraints, including road boundaries, remaining acceleration distance, vehicle dynamics, traffic rules, ride comfort, traffic efficiency, and perception confidence.

[0003] Existing lane-merging methods mainly include rule-based thresholding, artificial potential fields, and model predictive control. Rule-based thresholding is simple to implement but struggles to cover dense traffic and complex yielding behaviors; artificial potential fields can uniformly describe target attraction and obstacle repulsion, but are susceptible to local configurations and struggle to explicitly select specific merging gaps; model predictive control can explicitly handle dynamic and compliance constraints, but is highly sensitive to the prediction model, cost weights, and real-time performance of online solutions.

[0004] Some methods describe the completion of lane merging tasks using merging progress indicators such as remaining distance, lateral deviation, or speed error. However, a single progress indicator usually only reflects the state of the vehicle relative to the road geometry or reference speed, making it difficult to distinguish multiple candidate gaps in the target lane, and also unable to comprehensively evaluate the safety margins of vehicles ahead and behind, braking disturbances of following vehicles, gap movement trends, and dynamic accessibility corresponding to different gaps. Therefore, even if the vehicle is close to the center line of the main lane or the end of the merge, if the speed is not matched with the traffic flow, or if merging will force following vehicles to decelerate sharply, safety risks and unreasonable interaction problems may still exist.

[0005] Furthermore, there are uncertainties in the perception results of surrounding vehicle status, lane lines, and road boundaries. If the lane-changing decision does not take into account perception confidence and prediction error, the vehicle may be overly aggressive in scenarios such as obstruction, rain, fog, low adhesion, or sensor vibration; if a globally conservative threshold is always used, it is easy to cause low-speed hesitation, miss the lane-changing window, or even create new risks at the end of the acceleration lane. Summary of the Invention

[0006] In view of this, the present invention provides an autonomous vehicle lane merging method based on the candidate gap lane merging potential function to solve the safety risks and unreasonable interaction problems existing in the prior art, and avoid overly aggressive behavior under adverse conditions such as obstruction, rain and fog, while also avoiding missing lane merging windows due to a globally conservative threshold when perception is good, so as to achieve vehicle lane merging that takes into account safety, efficiency, main road vehicle comfort and complex perception conditions.

[0007] A lane-merging method for autonomous vehicles based on the candidate gap lane-merging potential function includes: Step S1: Integrate the vehicle's motion state, road topology information, target lane vehicle state, and perception uncertainty information to construct a unified state representation quantity for the lane merging scenario. Step S2: Generate and predict candidate lane merging gaps based on the unified state representation of the lane merging scenario, and form a candidate gap set according to the longitudinal arrangement of vehicles in the target lane and in combination with the backward virtual boundary and the forward virtual boundary. Step S3: Construct the candidate gap merging potential function based on the candidate gap set. For each candidate gap, calculate the merging geometric progress evaluation item, target speed matching evaluation item, lateral completion evaluation item, safety margin evaluation item, interactive disturbance evaluation item, dynamic reachability evaluation item, and perceived uncertainty evaluation item. After normalizing the above evaluation items, weighted combination is performed to obtain the merging potential function value of each candidate gap. Step S4: Based on the candidate gap set and the merging potential function value, construct a feasible candidate gap set, and perform joint constraint screening on the candidate gaps based on safety margin, interactive disturbance, dynamic reachability, continuous availability time and perceived confidence. Step S5: Based on the set of feasible candidate gaps, select the target gap and generate a longitudinal and transverse coupling track merging trajectory. Select the gap with the smallest potential function value of the candidate gap merging trajectory from the set of feasible candidate gaps as the target gap, and generate a longitudinal and transverse coupling track merging trajectory based on the center trajectory of the target gap. Step S6: Perform closed-loop control, re-select gaps and complete the judgment based on the longitudinal and lateral coupling lane merging trajectory, output vehicle control commands, and return to steps S4 and S5 to re-evaluate candidate gaps and select target gaps when the target gap fails, the risk increases, it becomes unreachable, or the target gap potential function increases beyond the preset relaxation amount.

[0008] The lane-merging method for autonomous vehicles based on the candidate gap lane-merging potential function provided by the present invention has the following beneficial effects: (1) This invention constructs a candidate gap merging potential function composed of seven evaluation items: geometric progress, speed matching, lateral completion, safety margin, interactive disturbance, dynamic reachability and perceived uncertainty. This expands the merging task from the absolute progress evaluation of the vehicle to a conditional evaluation for each candidate gap, so that the merging decision can output "whether it is close to completion" and "which gap should be merged into" at the same time, which improves the clarity and interpretability of the decision and solves the problem that a single progress indicator is difficult to distinguish multiple candidate gaps.

[0009] (2) In view of the problem that fixed target speed is difficult to adapt to the dynamic changes of traffic flow on the main road, the present invention generates conditional merging targets based on the candidate gap center position and target speed, so that the target merging position and speed of the vehicle follow the dynamic changes of the target lane traffic flow, without the need to pre-set reference speed curve, thereby improving the scene adaptability in congested, accelerating and decelerating traffic flow scenarios.

[0010] (3) In view of the risk that the vehicle may encroach on the safe distance of the vehicle in front or behind the target lane after merging, the present invention simultaneously calculates the forward effective safety margin and the backward effective safety margin, and uses the smaller of the two as the basis for safety evaluation. Candidate gaps that do not meet the safety margin threshold are excluded from the feasible set, thereby reducing the risk of encroaching on the safe distance of the vehicle in front or behind the target lane after merging from the source of decision-making.

[0011] (4) In response to the problem that the lane merging decision does not take into account the disturbance to vehicles on the main road, this invention introduces an interactive disturbance evaluation term. By quantifying the equivalent longitudinal deceleration required by the rear boundary vehicle to maintain a safe distance, it suppresses the forced insertion behavior that forces the following vehicle to decelerate sharply, so that the lane merging trajectory takes into account both the safety of the vehicle and the comfort of the vehicles on the main road. A dynamic reachability evaluation term is introduced to eliminate gaps that cannot be reached due to insufficient remaining distance or insufficient control capability in advance during the target gap selection stage, so as to avoid the vehicle being passively abandoned when it is close to the end of the acceleration lane.

[0012] (5) To address the problem of overly aggressive or conservative decision-making due to perceived uncertainty, this invention introduces a perceived uncertainty evaluation term and an expanded safety distance, which enables the safety boundary to be adaptively adjusted according to the prediction uncertainty and confidence level. It automatically increases the level of conservatism when there is obstruction, low visibility, low adhesion, or a decrease in sensor confidence, and maintains lane merging efficiency when the perceived confidence level is high. In the closed-loop control, the change of the target gap potential function is monitored, and the lane merging potential function is constrained to decrease over time. When the value of the candidate gap lane merging potential function increases beyond the preset relaxation amount or the target gap fails, the re-selection of gap or abandonment of lane merging strategy is triggered, so that lane merging decision-making, trajectory planning and control execution form a unified closed loop, improving safety, efficiency and robustness in complex traffic scenarios. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating the lane-merging method for autonomous vehicles based on the candidate gap lane-merging potential function. Figure 2 This is a schematic diagram of a candidate gap lane merging scenario in the verification case of this invention; Figure 3 This is a comparison chart of the lane merging success rates of the method of this invention, the DQN method, and the DDPG method under the same Highway-env lane merging simulation environment and a total simulation round count of 5000. Figure 4 This is a schematic diagram of the candidate gap potential function evaluation and target gap selection of the present invention. Detailed Implementation

[0014] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain embodiments of the present invention, and should not be construed as limiting the present invention.

[0015] Please see Figure 1 The present invention provides a lane-merging method for autonomous vehicles based on a candidate gap lane-merging potential function, the method comprising steps S1 to S6: Step S1: Integrate the vehicle's motion state, road topology information, target lane vehicle state, and perception uncertainty information to construct a unified state representation quantity for the lane merging scenario.

[0016] It should be noted that this embodiment uses discrete control cycles. Based on this, the vehicle, the target lane vehicle, the candidate gap, and the closed-loop control process are described in the Frenet coordinate system.

[0017] The unified state representation of the lane merging scenario constructed in step S1 This information is used for subsequent candidate gap generation, potential function calculation, and closed-loop control. Within each control cycle, the vehicle's motion state, road topology, target lane vehicle state, and uncertainty information are unified into the same Frenet coordinate system.

[0018] Specifically, the unified state representation quantity for parallel lane scenarios The observation is based on the vehicle itself, including the vehicle's motion state, vehicle size, vehicle dynamics constraints, road topology state, target lane vehicle state, and uncertainties.

[0019] The vehicle's motion state includes the vehicle's longitudinal position. Horizontal position lateral velocity Heading angle Longitudinal velocity and longitudinal acceleration .

[0020] The vehicle dimensions include the vehicle length. and vehicle width .

[0021] The vehicle dynamics constraints include a lower limit for longitudinal acceleration. Upper limit of longitudinal acceleration Maximum steering angle and maximum steering angular velocity .

[0022] The road topology status includes the vehicle's current lane. Target lane Permitted lane merging area start and end points and Remaining lane merging distance Target lane width and target lane curvature .

[0023] The target lane vehicle status includes the longitudinal distance between the target lane vehicle and the vehicle itself. Horizontal distance Longitudinal velocity Longitudinal acceleration Vehicle length Vehicle width Perception or communication confidence and prediction uncertainty .

[0024] Step S2: Generate and predict candidate lane merging gaps based on the unified state representation of the lane merging scenario. Form a candidate gap set according to the longitudinal arrangement of vehicles in the target lane and by combining the backward virtual boundary and the forward virtual boundary.

[0025] In this process, an extended vehicle sequence is first constructed based on the order of vehicles in the target lane from back to front. To cover scenarios where there are no vehicles in the target lane, only one vehicle, and the first or last open gap, backward virtual boundary vehicles and forward virtual boundary vehicles are set at both ends of the effective vehicle sequence. The backward virtual boundary vehicle is used to represent the boundary collected behind the vehicle, the start of the allowed lane merging area, or the rear boundary of the reserved gap provided by the roadside coordination; the forward virtual boundary vehicle is used to represent the boundary collected in front of the vehicle, the end of the allowed lane merging area, or the front boundary of the reserved gap.

[0026] Specifically, step S2 includes: The vehicles in the target lane are sorted from back to front according to their relative longitudinal position, and a backward virtual boundary and a forward virtual boundary are set at both ends of the sorted target lane vehicle sequence; the backward virtual boundary, the target lane vehicles, and the forward virtual boundary constitute an extended vehicle sequence. :

[0027] in, This represents the actual vehicles in the target lane, sorted by their longitudinal position. and These represent vehicles at the rear virtual boundary and vehicles at the front virtual boundary, respectively.

[0028] In the extended vehicle sequence, any two adjacent boundaries are used as the rear and front boundaries of the candidate gap, respectively. The two adjacent boundaries constitute the candidate gap:

[0029] in, Indicates the first The rear boundary of each candidate gap, Indicates the first The front boundary of each candidate gap, and They are respectively The first in The and the first One element, , This indicates the current number of valid vehicles in the target lane. When there are no valid vehicles in the target lane... An open candidate gap is still formed by the backward virtual boundary and the forward virtual boundary; when there is only one valid vehicle in the target lane, two candidate gaps are generated behind and in front of that vehicle.

[0030] The rear boundary of the candidate gap is determined based on the predicted longitudinal position of the rear and front boundaries. and forward boundary This leads to the formation of a candidate gap set. .

[0031] Specifically, a simplified constant acceleration prediction is used for vehicles at real boundaries, while for vehicles at virtual boundaries, their predicted positions are determined based on the boundaries of permitted lane-changing areas, the sensing and data acquisition boundaries, the target lane reference speed, or vehicle-road cooperative gap information. This is based on the backward boundary. and forward boundary Predict the location of the void center for each candidate void. Target speed Gap length Duration of continuous availability Confidence level and prediction uncertainty ; According to the backward boundary and forward boundary Calculate candidate gap length , location of the center of the gap and gap target velocity :

[0032]

[0033] This transforms the target lane merging of the vehicle from a fixed road position or fixed speed into a conditional target that dynamically changes with the movement of traffic in the target lane. In other words, the target merging position and speed of the vehicle no longer take fixed values, but instead follow the center position of the candidate gap. and target speed Dynamic changes. It should be noted that when both sides are real vehicles, the candidate gap target speed can be the average of the predicted speeds of the vehicles at the front and rear boundaries; when one side is a virtual boundary, the candidate gap target speed is determined by fusing the predicted speed of the real vehicles on the other side with the reference speed of the target lane; when both sides are virtual boundaries, the candidate gap target speed is the reference speed of the target lane.

[0034] In this embodiment, the duration of availability of candidate gaps The longest time that the geometric availability conditions can be continuously met from the start of the current control cycle, i.e., the candidate gap length. Not less than the length of the vehicle With twice the base buffer distance The condition for the sum is over the entire time interval. The maximum duration during which it remains valid; Confidence of candidate gaps It is synthesized from the smaller of the vehicle perception confidence scores at the front boundary and the vehicle perception confidence scores at the rear boundary. , and The first Perception confidence of vehicles at the rear and front boundaries of each candidate gap; Prediction uncertainty of candidate gaps It is obtained by combining the larger of the vehicle prediction uncertainty at the front boundary and the vehicle prediction uncertainty at the rear boundary, i.e. , and They represent the first The prediction uncertainty of vehicles at the rear and front boundaries of each candidate gap.

[0035] When vehicle-to-vehicle communication or vehicle-to-infrastructure (V2I) conditions are available, the center position of the candidate gap is corrected by using the predicted trajectory, yielding intention, cooperative gap number, or reserved gap information broadcast by vehicles in the target lane or roadside units. Target speed Confidence level and prediction uncertainty .

[0036] Then, the candidate gap set satisfy:

[0037]

[0038] in, Indicates the first The attribute vector of each candidate gap. Indicates the first The rear boundary of each candidate gap, Indicates the first The front boundary of each candidate gap, yes hour The value, yes hour The value, yes hour The value of .

[0039] Step S3: Construct the candidate gap merging potential function based on the candidate gap set. For each candidate gap, calculate the merging geometric progress evaluation item, target speed matching evaluation item, lateral completion evaluation item, safety margin evaluation item, interactive disturbance evaluation item, dynamic reachability evaluation item, and perceived uncertainty evaluation item. After normalizing the above evaluation items, weighted combination is performed to obtain the merging potential function value of each candidate gap.

[0040] Among them, the candidate gap channel potential function value Evaluation item based on parallel track geometry Target speed matching evaluation item Horizontal completion evaluation items Safety margin evaluation items Interactive perturbation evaluation items Dynamic reachability evaluation item And perceived uncertainty evaluation items The weighted combination yields:

[0041] in, to These are non-negative weighting coefficients. Candidate gap parallel channel potential function value The smaller the value, the more suitable the corresponding candidate gap is to be merged; the weight of each evaluation item is set or adaptively adjusted according to traffic density, remaining merging distance, road attachment status, visibility and perception confidence.

[0042] The following section provides a detailed explanation of each evaluation item, starting with the definition of the projection operator. Used to limit any scalar z to Interval: .

[0043] Parallel track geometric progress evaluation items Defined as:

[0044] in, Indicates the first The longitudinal position of the candidate gap center at the current moment, in meters; This indicates the vehicle's longitudinal position at the current moment, in meters. This indicates the remaining lane merging distance at the current moment, in meters. Represents extremely small positive numbers, used to prevent division by zero; unit is meters. The smaller the value, the closer the vehicle is to the candidate gap center.

[0045] Target speed matching evaluation item Defined as:

[0046] in, Indicates the first The target velocity of each candidate gap at the current moment, in meters per second; This indicates the longitudinal velocity of the vehicle at the current moment, expressed in meters per second. The smaller the value, the better the match between the vehicle speed and the candidate gap target speed.

[0047] Horizontal completion evaluation items Defined as:

[0048] in, This indicates the vehicle's lateral position at the current moment, in meters. This indicates the lateral coordinates of the target lane's centerline, in meters. Indicates the target lane width, in meters. The smaller the value, the closer the vehicle is to the center line of the target lane.

[0049] The safety margin evaluation item Simultaneously, the effective forward safety margin between the vehicle and the vehicle at the front boundary after the vehicle merges into the candidate gap is also considered. and the effective rearward safety margin with respect to vehicles at the rear boundary. and with the aforementioned forward effective safety margin and backward effective safety margin The smaller of the two values ​​is used as the basis for candidate gap security evaluation, namely the forward effective security margin. and backward effective safety margin Based on the position of the rear of the vehicle at the front boundary, the position of the front of the vehicle at the rear boundary, and the half-length of the vehicle. The vehicle length is calculated by taking into account the uncertainty expansion and without repeatedly deducting the vehicle length of the front or rear boundary vehicle.

[0050] Specifically, The expression is:

[0051]

[0052] in, This represents the minimum safety margin calculated based on the center of the gap at the current moment, in meters; Indicates the first The longitudinal position of each candidate gap on the forward boundary at the current moment, in meters; Indicates the first The longitudinal position of each candidate gap on the backward boundary at the current moment, in meters; This indicates the length of the vehicle, in meters. This indicates the basic safety distance, expressed in meters. The smaller the value, the greater the safety margin.

[0053] The interaction disturbance evaluation item The equivalent longitudinal deceleration required by vehicles at the rear boundary of the target lane to maintain a safe distance It is determined that the equivalent longitudinal deceleration According to the planned merging time of vehicles at the rear boundary speed Candidate gap target velocity and backward effective safety margin Calculations show that when the equivalent longitudinal deceleration... Increasing the value of the interaction perturbation evaluation of the corresponding candidate gap improves the overall value of the perturbation evaluation. .

[0054] Specifically, The expression is:

[0055]

[0056] in, Indicates the first Each candidate gap corresponds to the estimated equivalent deceleration of the vehicle at the rear boundary at the current moment, in meters per second. 2 ; Indicates the first The speed of the vehicle at the boundary of each candidate gap at the current moment, in meters per second; Indicates the first The target velocity of each candidate gap at the current moment, in meters per second; This indicates the comfort deceleration threshold for the following vehicle, in meters per second. 2 Take a positive number. The smaller the value, the smaller the interaction disturbance to the following vehicle.

[0057] The dynamic reachability evaluation item The time for the bicycle to be incorporated into the plan The equivalent longitudinal acceleration required to reach the center position of the candidate gap during the planning time. Determined jointly with the vehicle's longitudinal acceleration capability boundary, and combined with the remaining lane-changing distance. and the duration of availability of candidate gaps Determine whether the candidate gap is reachable.

[0058] Specifically, The expression is:

[0059]

[0060] in, Indicates the planned integration time for the vehicle. The equivalent longitudinal acceleration required to reach the center position of the candidate gap at the planning time, in meters per second. 2 ; Indicates the first The candidate gaps are planned for inclusion in the timeline. The longitudinal position of the center after, in meters; Indicates that the plan is incorporated into the first The time required for each gap is in seconds; This represents the normalized upper bound of the vehicle's longitudinal acceleration capability, expressed in meters per second. 2 . The smaller the value, the better the vehicle's dynamic reachability.

[0061] The perceived uncertainty evaluation item Based on the prediction uncertainty of candidate gaps and confidence level Confirmed, the expression is:

[0062] in, This represents the normalized reference value for uncertainty, in meters. The smaller the value, the less perceived uncertainty.

[0063] When prediction uncertainty Increase confidence When the road adhesion coefficient is reduced, visibility is reduced, or vehicle-road cooperative information is unavailable, improve the evaluation item for perception uncertainty. Safety margin evaluation items Interactive perturbation evaluation items Or dynamic reachability evaluation item The weights are determined, and the candidate gaps and potential function values ​​are recalculated. .

[0064] Step S4: Based on the candidate gap set and the potential function value, construct a feasible candidate gap set, and perform joint constraint screening on the candidate gaps based on safety margin, interactive disturbance, dynamic reachability, continuous availability time and perceived confidence.

[0065] First, based on the prediction uncertainty of the candidate gaps and confidence level The expanded safety distance is obtained by performing uncertainty expansion on the basic safety distance. Represented as:

[0066] in, Based on the safe distance, and These are the safety distance inflation coefficients corresponding to prediction uncertainty and confidence level missing, respectively.

[0067] Then based on the backward boundary of the candidate gap and forward boundary Planning and integration time Candidate gaps are located at the center of the planning and incorporation time. Bicycle length and expansion safety distance Calculate the forward effective safety margin and backward effective safety margin ; Then, based on the speed of the vehicles at the rear boundary during the planned merging time... The target speed of candidate gaps in the planning and incorporation time. and backward effective safety margin Calculate the equivalent longitudinal deceleration required for the vehicle at the rear boundary. :

[0068] in It is a very small positive number; Determine whether each candidate gap simultaneously satisfies all of the following conditions: forward effective safety margin Backward effective safety margin The equivalent longitudinal deceleration required by the vehicle at the rear boundary Not greater than the comfort deceleration threshold of the following vehicle Planning and integration time Not greater than the corresponding candidate gap duration Candidate gap confidence Not lower than the preset reliability threshold and the requirement that the vehicle's dynamics reachability must be met (i.e. ); Candidate gaps that simultaneously meet all the above conditions are determined to be feasible candidate gaps and added to the feasible candidate gap set. Candidate gaps that do not simultaneously meet all of the above conditions will be excluded from the set of feasible candidate gaps or their selection priority will be reduced. Based on traffic density and remaining merging distance Adaptive adjustment of candidate gap and lane merging potential function values ​​based on road adhesion coefficient and visibility Weighting coefficients in to After adjustment, the candidate gap and potential function values ​​are recalculated. .

[0069] Step S5: Based on the set of feasible candidate gaps, select the target gap and generate a longitudinal and transverse coupling track trajectories. Select the gap with the smallest potential function value in the set of feasible candidate gaps as the target gap, and generate a longitudinal and transverse coupling track trajectories based on the center trajectory of the target gap.

[0070] Among them, when the set of feasible candidate gaps When not empty, select candidate gaps from the feasible candidate gap set and combine the potential function values. The smallest candidate gap is selected as the target gap. Target gap The corresponding target candidate gap number is ; The center position of the target gap and target speed As the terminal conditional target of the longitudinal and lateral coupling lane-merging trajectory of the vehicle, that is, let , ,in Indicates the target gap in the prediction time The center longitudinal position after Indicates the target gap in the prediction time The target speed is then set, and the corresponding planning and integration time for the target gap is determined. , for The corresponding planning and integration time; Constructing a longitudinal fifth-order polynomial locus in the Frenet coordinate system and the lateral fifth-order polynomial trajectory :

[0071]

[0072] in, and The fifth degree polynomials in the vertical and horizontal directions are respectively the first... Order coefficient; Longitudinal polynomial trajectory The following boundary conditions must be met:

[0073] in, To plan the longitudinal trajectory in The longitudinal position at that time This represents the vehicle's actual longitudinal position at the current moment. for The first derivative, for The second derivative, Let x be the vehicle's actual longitudinal speed at the current moment. Let x be the actual longitudinal acceleration of the vehicle at the current moment. To plan the longitudinal trajectory at the planning incorporation time The subsequent vertical position, For the target gap in the planning and incorporation time The center longitudinal position (i.e., the terminal target position). To plan the longitudinal trajectory at the planning incorporation time The subsequent longitudinal velocity, For the target gap in the planning and incorporation time The target speed after (i.e., the terminal target speed). To plan the longitudinal trajectory at the planning incorporation time The longitudinal acceleration afterwards.

[0074] Lateral polynomial trajectory The following boundary conditions must be met:

[0075] in, To plan the lateral trajectory in Horizontal position at time for The first derivative, for The second derivative, This represents the vehicle's actual lateral position at the current moment. The vehicle's actual lateral speed at the current moment To plan the lateral trajectory during the planning and incorporation time The horizontal position after that, This indicates the lateral coordinates of the target lane centerline (i.e., the lateral position of the terminal target). To plan the lateral trajectory during the planning and incorporation time Lateral velocity, To plan the lateral trajectory during the planning and incorporation time Lateral acceleration.

[0076] Based on the longitudinal polynomial trajectory and transverse polynomial trajectory Generate longitudinal and lateral coupled parallel trajectories :

[0077] in, and These represent the planned longitudinal velocity and the planned longitudinal acceleration, respectively. Indicates the planned lateral speed. Indicates the planned heading angle.

[0078] Step S6: Perform closed-loop control, re-select gaps and complete the judgment based on the longitudinal and lateral coupling lane merging trajectory, output vehicle control commands, and return to steps S4 and S5 to re-evaluate candidate gaps and select target gaps when the target gap fails, the risk increases, it becomes unreachable, or the target gap potential function increases beyond the preset relaxation amount.

[0079] Specifically, during the process of executing closed-loop control, re-selecting gaps, and determining completion based on the longitudinal and lateral coupling lane merging trajectory, and outputting vehicle control commands, a pre-aiming time is selected. Calculate the lateral error based on the planned trajectory. Heading error and speed error :

[0080]

[0081]

[0082] in, Indicates the preview time for the controller to retrieve the planned trajectory, in seconds; This represents the lateral error, which is the difference between the planned lateral position and the actual lateral position of the vehicle, expressed in meters. This indicates the lateral position of the planned trajectory at the pre-aiming time, in meters; This indicates the vehicle's actual lateral position at the current moment, in meters. This indicates the heading error, which is the difference between the planned heading angle and the actual heading angle of the vehicle, expressed in radians. The heading angle of the planned trajectory at the pre-aiming moment is expressed in radians. This represents the vehicle's actual heading angle at the current moment, in radians. This represents the speed error, which is the difference between the planned longitudinal speed and the actual longitudinal speed of the vehicle, expressed in meters per second. This represents the longitudinal velocity of the planned trajectory at the preview time, expressed in meters per second. This indicates the vehicle's actual longitudinal speed at the current moment, expressed in meters per second.

[0083] Longitudinal acceleration control command and steering angle control commands for:

[0084]

[0085]

[0086] in, This indicates that the input range will be limited to a certain interval. saturation function; This represents the longitudinal acceleration of the planned trajectory at the preview moment, in meters per second. 2 ; Indicates the speed error control gain; This indicates a steering angle control command, expressed in radians. This indicates that the input range will be limited to a certain interval. saturation function; Indicates the lateral error control gain; Indicates the heading error control gain; The steering angle control command for the current control cycle is expressed in radians. This indicates the steering angle control command from the previous control cycle, expressed in radians. Indicates the control cycle, in seconds; This indicates the maximum steering angular velocity, expressed in radians per second.

[0087] If the same target gap is selected in adjacent control cycles, the constraint potential function will not increase significantly.

[0088] Furthermore, in this embodiment, the vehicle status of the target lane and the candidate gap set are continuously updated during the closed-loop execution process. and candidate gap potential function value ; When the same target gap is selected in adjacent control cycles, the candidate gap potential function value of that target gap is monitored. The change relative to the previous control cycle; When the target gap no longer belongs to the set of feasible candidate gaps Forward effective safety margin or backward effective safety margin The equivalent longitudinal deceleration required by the rear boundary vehicle is less than 0. Exceeding the comfort deceleration threshold of the vehicle behind Planning and integration time Exceeding the corresponding candidate gap duration Candidate gap confidence Below the preset confidence threshold Or the candidate gap potential function value increases beyond a preset relaxation amount. If necessary, return to steps S4 and S5 to reselect the target gap or replan the trajectory.

[0089] In this embodiment, if no feasible candidate gap that meets the constraints is found after reselecting the target gap, the vehicle is controlled to switch to deceleration and waiting, maintain the current lane, or abandon the lane-merging strategy; when the vehicle fully enters the target lane and the lateral deviation is less than a preset threshold... The heading angle is consistent with the direction of the target lane, and the lateral speed is close to zero. Matching the vehicle speed with the target gap speed Furthermore, the safety margin with vehicles in front of and behind the target lane meets the requirements and continues to meet the preset time. During each control cycle, the parallel processing is determined to be complete.

[0090] Therefore, step S6 forms a closed loop of target gap selection, trajectory planning and vehicle control, and triggers a re-selection of gap, deceleration and waiting or abandonment of lane merging strategy when the target gap fails.

[0091] Validation Case In one verification embodiment, an autonomous vehicle ramp merging scenario was constructed using the open-source Highway-env simulation environment, and in accordance with... Figure 1 The process shown sequentially executes the unified state representation of the lane merging scenario, candidate gap generation and prediction, candidate gap potential function construction, safety margin, interactive disturbance, joint constraint evaluation of dynamic reachability and perception uncertainty, target gap selection, longitudinal and lateral coupled trajectory generation, and closed-loop control. In the simulation, the vehicle enters the merging area from a single-lane ramp and selects a merging gap in the target lane of the main road within the finite remaining merging distance. Vehicles in the target lane are randomly generated according to a preset speed range, headway, and initial position, forming a pattern as shown below. Figure 2 The scenario shown is a multi-candidate gap strong interaction lane merging scenario. Figure 2It demonstrates the interaction between a vehicle entering the target lane of the main road from the waiting area of ​​the ramp, and the vehicles in front and behind the target lane, the candidate gap, the target gap, the safety margin, and the expansion boundary of perceived uncertainty.

[0092] Under the same simulation scenario, number of simulation rounds, and evaluation rules, the method of this invention is compared with the traditional DQN and DDPG methods. Each method uses whether lane merging is completed, whether a collision occurs, whether the vehicle leaves the drivable area, and whether the maximum decision time is exceeded as round termination criteria, recording the lane merging success rate and the curve stability during the evaluation process. To ensure consistency in the comparison, the DQN and DDPG methods use the same road scenario, initial vehicle distribution, state input, and action constraints. The method of this invention explicitly introduces evaluation terms such as geometric progress, speed matching, lateral completion, safety margin, rear vehicle interaction disturbance, dynamic reachability, and perception uncertainty at the candidate gap level, and selects the target gap based on the set of feasible candidate gaps.

[0093] like Figure 3 As shown, the differences among the three methods in terms of lane merging success rate, curve convergence trend, and later-stage fluctuation stability are illustrated. Over a total of 5000 simulation evaluation rounds, the lane merging success rate of the method described in this invention increases faster in the early stages of evaluation, entering a higher success rate range after approximately 2000 simulation rounds, and maintaining relatively small fluctuations in subsequent evaluations. The DQN method, affected by discrete action search and a highly interactive environment, has a slower convergence speed and more significant fluctuations in the later stages. The DDPG method can learn continuous control strategies, but still exhibits periodic oscillations when candidate gap safety screening and subsequent vehicle interaction constraints are insufficient. The lane merging success rate of the method described in this invention after stabilization is higher than that of the two comparative methods, indicating that candidate gap potential functions and joint constraint screening can improve the clarity, evaluation efficiency, and stability of lane merging decisions.

[0094] like Figure 4 The diagram illustrates the process of creating a candidate gap set, calculating the potential function, screening feasible set constraints, comparing the potential function values ​​of each candidate gap, and selecting the feasible gap with the smallest potential function value as the target gap. It can be seen that the method of this invention does not only evaluate the lane centerline, instantaneous speed error, or single lane merging progress, but first forms a candidate gap set, then calculates the lane merging potential function value for each candidate gap, and then filters it through safety margin, interactive disturbance, dynamic reachability, and perceived uncertainty constraints to form a feasible candidate gap set. Finally, it selects the candidate gap with the smallest potential function value from the feasible candidate gap set as the target gap. This process allows for interpretable intermediate variables in target gap selection and trajectory generation, and enables re-selection of gaps when the target gap fails, the risk increases, or it becomes unreachable, returning to the candidate gap evaluation process.

[0095] comprehensive Figures 2 to 4The method of this invention can be used to uniformly evaluate the geometric relationship, speed relationship, safety margin, interaction effect, dynamic reachability and perception uncertainty of the target lane candidate gap in the complex ramp merging scenario constructed by Highway-env, and demonstrates advantages over the DQN method and DDPG method in terms of merging success rate and decision stability.

[0096] In summary, the lane-merging method for autonomous vehicles based on the candidate gap lane-merging potential function according to the above embodiments has the following beneficial effects: (1) This invention constructs a candidate gap merging potential function composed of seven evaluation items: geometric progress, speed matching, lateral completion, safety margin, interactive disturbance, dynamic reachability and perceived uncertainty. This expands the merging task from the absolute progress evaluation of the vehicle to a conditional evaluation for each candidate gap, so that the merging decision can output "whether it is close to completion" and "which gap should be merged into" at the same time, which improves the clarity and interpretability of the decision and solves the problem that a single progress indicator is difficult to distinguish multiple candidate gaps.

[0097] (2) In view of the problem that fixed target speed is difficult to adapt to the dynamic changes of traffic flow on the main road, the present invention generates conditional merging targets based on the candidate gap center position and target speed, so that the target merging position and speed of the vehicle follow the dynamic changes of the target lane traffic flow, without the need to pre-set reference speed curve, thereby improving the scene adaptability in congested, accelerating and decelerating traffic flow scenarios.

[0098] (3) In view of the risk that the vehicle may encroach on the safe distance of the vehicle in front or behind the target lane after merging, the present invention simultaneously calculates the forward effective safety margin and the backward effective safety margin, and uses the smaller of the two as the basis for safety evaluation. Candidate gaps that do not meet the safety margin threshold are excluded from the feasible set, thereby reducing the risk of encroaching on the safe distance of the vehicle in front or behind the target lane after merging from the source of decision-making.

[0099] (4) In response to the problem that the lane merging decision does not take into account the disturbance to vehicles on the main road, this invention introduces an interactive disturbance evaluation term. By quantifying the equivalent longitudinal deceleration required by the rear boundary vehicle to maintain a safe distance, it suppresses the forced insertion behavior that forces the following vehicle to decelerate sharply, so that the lane merging trajectory takes into account both the safety of the vehicle and the comfort of the vehicles on the main road. A dynamic reachability evaluation term is introduced to eliminate gaps that cannot be reached due to insufficient remaining distance or insufficient control capability in advance during the target gap selection stage, so as to avoid the vehicle being passively abandoned when it is close to the end of the acceleration lane.

[0100] (5) To address the problem of overly aggressive or conservative decision-making due to perceived uncertainty, this invention introduces a perceived uncertainty evaluation term and an expanded safety distance, which enables the safety boundary to be adaptively adjusted according to the prediction uncertainty and confidence level. It automatically increases the level of conservatism when there is obstruction, low visibility, low adhesion, or a decrease in sensor confidence, and maintains lane merging efficiency when the perceived confidence level is high. In the closed-loop control, the change of the target gap potential function is monitored, and the lane merging potential function is constrained to decrease over time. When the value of the candidate gap lane merging potential function increases beyond the preset relaxation amount or the target gap fails, the re-selection of gap or abandonment of lane merging strategy is triggered, so that lane merging decision-making, trajectory planning and control execution form a unified closed loop, improving safety, efficiency and robustness in complex traffic scenarios.

[0101] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A lane-merging method for autonomous vehicles based on the candidate gap lane-merging potential function, characterized in that, include: Step S1: Integrate the vehicle's motion state, road topology information, target lane vehicle state, and perception uncertainty information to construct a unified state representation quantity for the lane merging scenario. Step S2: Generate and predict candidate lane merging gaps based on the unified state representation of the lane merging scenario, and form a candidate gap set according to the longitudinal arrangement of vehicles in the target lane and in combination with the backward virtual boundary and the forward virtual boundary. Step S3: Construct the candidate gap merging potential function based on the candidate gap set. For each candidate gap, calculate the merging geometric progress evaluation item, target speed matching evaluation item, lateral completion evaluation item, safety margin evaluation item, interactive disturbance evaluation item, dynamic reachability evaluation item, and perceived uncertainty evaluation item. After normalizing the above evaluation items, weighted combination is performed to obtain the merging potential function value of each candidate gap. Step S4: Based on the candidate gap set and the merging potential function value, construct a feasible candidate gap set, and perform joint constraint screening on the candidate gaps based on safety margin, interactive disturbance, dynamic reachability, continuous availability time and perceived confidence. Step S5: Based on the set of feasible candidate gaps, select the target gap and generate a longitudinal and transverse coupling track merging trajectory. Select the gap with the smallest potential function value of the candidate gap merging trajectory from the set of feasible candidate gaps as the target gap, and generate a longitudinal and transverse coupling track merging trajectory based on the center trajectory of the target gap. Step S6: Perform closed-loop control, re-select gaps and complete the judgment based on the longitudinal and lateral coupling lane merging trajectory, output vehicle control commands, and return to steps S4 and S5 to re-evaluate candidate gaps and select target gaps when the target gap fails, the risk increases, it becomes unreachable, or the target gap potential function increases beyond the preset relaxation amount.

2. The lane merging method for autonomous vehicles based on the candidate gap lane merging potential function according to claim 1, characterized in that, In step S1, the unified state representation quantity of the lane merging scenario The observation is based on the vehicle itself, including the vehicle's motion state, vehicle size, vehicle dynamics constraints, road topology, target lane vehicle state, and uncertainties. The vehicle's motion state includes the vehicle's longitudinal position. Horizontal position lateral velocity Heading angle Longitudinal velocity and longitudinal acceleration ; The vehicle dimensions include the vehicle length. and vehicle width ; The vehicle dynamics constraints include a lower limit for longitudinal acceleration. Upper limit of longitudinal acceleration Maximum steering angle and maximum steering angular velocity ; The road topology status includes the vehicle's current lane. Target lane Permitted lane merging area start and end points and Remaining lane merging distance Target lane width and target lane curvature ; The target lane vehicle status includes the longitudinal distance between the target lane vehicle and the vehicle itself. Horizontal distance Longitudinal velocity Longitudinal acceleration Vehicle length Vehicle width Perception or communication confidence and prediction uncertainty .

3. The lane merging method for autonomous vehicles based on the candidate gap lane merging potential function according to claim 2, characterized in that, Step S2 specifically includes: The vehicles in the target lane are sorted from back to front according to their relative longitudinal position, and a backward virtual boundary and a forward virtual boundary are set at both ends of the sorted target lane vehicle sequence; the backward virtual boundary, the target lane vehicles, and the forward virtual boundary constitute an extended vehicle sequence. The rear boundary of the candidate gap is determined by taking any two adjacent boundaries in the extended vehicle sequence as the rear and front boundaries, respectively, and then determining the rearward boundary of the candidate gap based on the predicted longitudinal positions of the rear and front boundaries. and forward boundary This leads to the formation of a candidate gap set. ; According to the backward boundary and forward boundary Predict the location of the void center for each candidate void. Target speed Gap length Duration of continuous availability Confidence level and prediction uncertainty ; According to the backward boundary and forward boundary Calculate candidate gap length , location of the gap center and gap target velocity This transforms the target lane merging of the vehicle from a fixed road position or fixed speed into a conditional target that dynamically changes with the movement of traffic in the target lane. In other words, the target merging position and target merging speed of the vehicle no longer take fixed values, but follow the center position of the candidate gap. and target speed Dynamic changes.

4. The autonomous vehicle lane merging method based on the candidate gap lane merging potential function according to claim 3, characterized in that, Duration of availability of candidate gaps The longest time during which the geometric availability conditions are continuously met from the start of the current control cycle; Confidence of candidate gaps It is synthesized from the smaller of the vehicle perception confidence scores at the front boundary and the vehicle perception confidence scores at the rear boundary. , and The first Perception confidence of vehicles at the rear and front boundaries of each candidate gap; Prediction uncertainty of candidate gaps It is obtained by combining the larger of the vehicle prediction uncertainty at the front boundary and the vehicle prediction uncertainty at the rear boundary, i.e. , and They represent the first The prediction uncertainty of vehicles at the rear and front boundaries of each candidate gap. When vehicle-to-vehicle communication or vehicle-to-infrastructure (V2I) conditions are available, the center position of the candidate gap is corrected by using the predicted trajectory, yielding intention, cooperative gap number, or reserved gap information broadcast by vehicles in the target lane or roadside units. Target speed Confidence level and prediction uncertainty .

5. The autonomous vehicle lane merging method based on the candidate gap lane merging potential function according to claim 4, characterized in that, In step S3, the candidate gap potential function value Evaluation item based on parallel track geometry Target speed matching evaluation item Horizontal completion evaluation items Safety margin evaluation items Interactive disturbance evaluation items Dynamic reachability evaluation item And perceived uncertainty evaluation items The weighted combination yields: in, to These are non-negative weighting coefficients. .

6. The autonomous vehicle lane merging method based on the candidate gap lane merging potential function according to claim 5, characterized in that, The safety margin evaluation item Simultaneously, the effective forward safety margin between the vehicle and the vehicle at the front boundary after the vehicle merges into the candidate gap is also considered. and the effective rearward safety margin with respect to vehicles at the rear boundary. and with the aforementioned forward effective safety margin and backward effective safety margin The smaller of the two values ​​is used as the basis for candidate gap security evaluation, namely the forward effective security margin. and backward effective safety margin Based on the position of the rear of the vehicle at the front boundary, the position of the front of the vehicle at the rear boundary, and the half-length of the vehicle. The vehicle length is calculated by taking into account the uncertainty expansion and without deducting the vehicle length of the front or rear boundary vehicle repeatedly. The interaction disturbance evaluation item The equivalent longitudinal deceleration required by vehicles at the rear boundary of the target lane to maintain a safe distance It is determined that the equivalent longitudinal deceleration According to the planned merging time of vehicles at the rear boundary speed Candidate gap target velocity and backward effective safety margin Calculations show that when the equivalent longitudinal deceleration... Increasing the value of the interaction perturbation evaluation of the corresponding candidate gap improves the overall value of the interaction perturbation evaluation. ; The dynamic reachability evaluation item The time for the bicycle to be incorporated into the planning The equivalent longitudinal acceleration required to reach the center position of the candidate gap during the planning time. Determined jointly with the vehicle's longitudinal acceleration capability boundary, and combined with the remaining lane-changing distance. and the duration of availability of candidate gaps Determine if the candidate gap is reachable; The perceived uncertainty evaluation item Based on the prediction uncertainty of candidate gaps and confidence level Sure.

7. The autonomous vehicle lane merging method based on the candidate gap lane merging potential function according to claim 6, characterized in that, Step S4 specifically includes: Based on the backward boundary of the candidate gap and forward boundary Planning and integration time Candidate gaps are located at the center of the planning and incorporation time. Bicycle length and expansion safety distance Calculate the forward effective safety margin and backward effective safety margin Among them, the expansion safety distance Based on the prediction uncertainty of candidate gaps and confidence level The uncertainty dilation of the basic safety distance is obtained; Based on the speed of vehicles at the rear boundary during the planned merging time. The target speed of candidate gaps in the planning and incorporation time. and backward effective safety margin Calculate the equivalent longitudinal deceleration required for the vehicle at the rear boundary. : in It is a very small positive number; Determine whether each candidate gap simultaneously satisfies all of the following conditions: forward effective safety margin Backward effective safety margin The equivalent longitudinal deceleration required by the vehicle at the rear boundary Not greater than the comfort deceleration threshold of the following vehicle Planning and integration time Not greater than the duration of availability of the candidate gap Candidate gap confidence Not lower than the preset reliability threshold And the vehicle's dynamic reachability meets the requirements; Candidate gaps that simultaneously meet all the above conditions are determined to be feasible candidate gaps and added to the feasible candidate gap set. Candidate gaps that do not simultaneously meet all of the above conditions will be excluded from the set of feasible candidate gaps or their selection priority will be reduced. Based on traffic density and remaining merging distance Adaptive adjustment of candidate gap and lane potential function values ​​based on road adhesion coefficient and visibility Weighting coefficients in to After adjustment, the candidate gap and potential function values ​​are recalculated. .

8. The lane merging method for autonomous vehicles based on the candidate gap lane merging potential function according to claim 7, characterized in that, In step S5, when the set of feasible candidate gaps... When not empty, select candidate gaps from the feasible candidate gap set and combine the potential function values. The smallest candidate gap is selected as the target gap. Target gap The corresponding target candidate gap number is ; The center position of the target gap and target speed As the terminal conditional target of the longitudinal and lateral coupling lane-merging trajectory of the vehicle, that is, let , ,in Indicates the target gap in the prediction time The center longitudinal position after Indicates the target gap in the prediction time The target speed is then set, and the corresponding planning and integration time for the target gap is determined. , for The corresponding planning and integration time; Constructing a longitudinal fifth-order polynomial locus in the Frenet coordinate system and the lateral fifth-order polynomial trajectory : in, and The fifth degree polynomials in the vertical and horizontal directions are respectively the first... Order coefficient; Based on the longitudinal polynomial trajectory and transverse polynomial trajectory Generate longitudinal and lateral coupled parallel trajectories : in, and These represent the planned longitudinal velocity and the planned longitudinal acceleration, respectively. Indicates the planned lateral speed. Indicates the planned heading angle.

9. The autonomous vehicle lane merging method based on the candidate gap lane merging potential function according to claim 8, characterized in that, In step S6, the vehicle status of the target lane and the candidate gap set are continuously updated during the closed-loop execution process. and candidate gap potential function value ; When the same target gap is selected in adjacent control cycles, the candidate gap potential function value of that target gap is monitored. The change relative to the previous control cycle; When the target gap no longer belongs to the set of feasible candidate gaps Forward effective safety margin or backward effective safety margin The equivalent longitudinal deceleration required by the rear boundary vehicle is less than 0. Exceeding the comfort deceleration threshold of the vehicle behind Planning and integration time Exceeding the corresponding candidate gap duration Candidate gap confidence Below the preset confidence threshold Or the candidate gap potential function value increases beyond a preset relaxation amount. If necessary, return to steps S4 and S5 to reselect the target gap or replan the trajectory.

10. The autonomous vehicle lane merging method based on the candidate gap lane merging potential function according to claim 9, characterized in that, In step S6, if no feasible candidate gap that meets the constraints is found after reselecting the target gap, the vehicle is controlled to switch to deceleration and waiting, maintain the current lane, or abandon the lane-merging strategy. When the vehicle has fully entered the target lane and the lateral deviation is less than the preset threshold The heading angle is consistent with the direction of the target lane, and the lateral speed is close to zero. Matching the vehicle speed with the target gap speed Furthermore, the safety margin with vehicles in front of and behind the target lane meets the requirements and continues to meet the preset time. During each control cycle, the parallel processing is determined to be complete.

Images