An active gap navigation method based on three-dimensional search

Abstract

The application relates to an active gap-creating detour method based on three-dimensional search, and belongs to the technical field of intelligent driving, wherein the method comprises judging whether a traffic environment around a self vehicle meets a detour triggering condition, the detour triggering condition comprising that a current lane where the self vehicle is located has a first obstacle needing to be detoured, and a target lane has no safe merging gap due to a second obstacle; in the case that the detour triggering condition is met, the actual longitudinal gap with the second obstacle is adjusted by regulating the longitudinal driving speed of the self vehicle in the current lane; and in the case that the actual longitudinal gap is greater than or equal to the safe merging gap, a detour route from the current lane to the target lane for detouring the first obstacle is planned. Therefore, the application can improve the traffic efficiency in high-density and continuous traffic flow by actively creating a detour condition, compared with the current passive response type obstacle detour scheme.

Description

Technical Field

[0001] This invention relates to the field of intelligent driving technology, and in particular to an active bypass method based on three-dimensional search. Background Technology

[0002] With the continuous development of intelligent driving technologies such as autonomous driving and advanced driver assistance systems, the autonomous navigation and decision-making capabilities of vehicles, especially their ability to navigate around obstacles on structured roads, have become one of the key indicators for measuring their level of intelligence.

[0003] Currently, the industry has numerous technical solutions for obstacle avoidance. For example, patent CN116729384B discloses a "detour planning method, device, and vehicle in lane-keeping mode." The core of this method is that when the vehicle detects an obstacle ahead that needs to be avoided, it first determines whether there is sufficient lateral safety space on the side of its own lane (i.e., the target lane). If this condition is met, a lateral deflection detour trajectory is planned and executed within the current lane, thereby completing the obstacle avoidance. This patent CN116729384B represents the current mainstream technical architecture for obstacle avoidance, which is a passive response obstacle avoidance scheme of "detecting an obstacle ahead -> checking lateral space -> detouring when space exists."

[0004] However, existing passive obstacle avoidance solutions have a significant drawback: their decision-making process is reactive. Before bypassing an obstacle, they require the detection of lateral space—that is, sufficient lateral safety space in the target lane. In high-density, continuous traffic flow, this lateral safety space may be lacking for extended periods. In such situations, existing obstacle avoidance solutions cannot bypass the obstacle and are forced to follow or even stop, leading to a severe decrease in traffic efficiency. Therefore, there is an urgent need for a more intelligent and adaptive obstacle avoidance technology that can proactively address scenarios with no safe clearance. Summary of the Invention

[0005] In view of this, it is necessary to provide an active obstacle avoidance method based on three-dimensional search to solve the problem of low traffic efficiency caused by the passive response obstacle avoidance schemes in the existing technology.

[0006] To address the above problems, this invention provides an active gap-filling bypass method based on three-dimensional search, comprising: Determine whether the traffic environment around the vehicle meets the detour triggering conditions. The detour triggering conditions include the presence of a first obstacle that needs to be detoured in the current lane where the vehicle is located, and the absence of a safe merging gap in the target lane due to the presence of a second obstacle. If the detour triggering conditions are met, the actual longitudinal clearance with the second obstacle is adjusted by regulating the longitudinal speed of the vehicle in the current lane. If the actual longitudinal clearance is greater than or equal to the safe merging clearance, a detour route is planned from the current lane to the target lane to bypass the first obstacle.

[0007] In one possible implementation, adjusting the actual longitudinal clearance with the second obstacle by regulating the longitudinal speed of the vehicle in the current lane specifically includes: A longitudinal velocity curve is planned on a time-longitudinal displacement map using a dynamic programming algorithm, wherein the time-longitudinal displacement map reflects the dynamic longitudinal positional relationship between the vehicle and the second obstacle on the target lane over time. The vehicle is controlled to travel according to the longitudinal speed curve in order to adjust the actual longitudinal clearance with the second obstacle.

[0008] In one possible implementation, a longitudinal velocity curve is planned on the time-longitudinal displacement graph using a dynamic programming algorithm, specifically including... Search forward for multiple velocity adjustment strategies on the time-longitudinal displacement graph; The cumulative cost of each speed adjustment strategy over future time steps is evaluated using the dynamic programming algorithm Cost = α × comfort + β × clearance + γ × progress. Cost is the cumulative cost; comfort is the comfort cost term used to penalize acceleration and jerk; clearance is the safety cost term used to reward maintaining a safe distance from vehicles in front and behind; and progress is the traffic efficiency cost term used to encourage driving speed. α, β, and γ are weighting coefficients. The longitudinal speed curve corresponding to the speed adjustment strategy with the lowest cumulative cost is selected as the planned longitudinal speed curve.

[0009] In one possible implementation, the method further includes selecting values ​​for α, β, and γ based on the vehicle's driving style, wherein the driving style includes conservative and aggressive driving styles.

[0010] In one possible implementation, determining whether the traffic environment surrounding the vehicle meets the detour triggering conditions specifically includes: Upon receiving a detour command triggered by a user, determine whether the traffic environment surrounding the vehicle meets the detour triggering conditions; and / or, The system continuously monitors the traffic environment around the vehicle and determines whether the traffic environment around the vehicle meets the detour triggering conditions.

[0011] In one possible implementation, planning a detour route from the current lane to the target lane to bypass the first obstacle specifically includes: in a three-dimensional spatiotemporal state grid composed of longitudinal displacement, lateral displacement and time, with spatiotemporal collision-free conditions as constraints, planning a detour route from the current state of the vehicle to the target state after bypassing the first obstacle in the target lane.

[0012] In one possible implementation, within a three-dimensional spatiotemporal state grid composed of longitudinal displacement, lateral displacement, and time, and under the constraint of spatiotemporal collision-free operation, a detour route is planned from the vehicle's current state to the target state after successfully circumventing the first obstacle in the target lane. Specifically, this includes: In a three-dimensional spatiotemporal state grid composed of longitudinal displacement, lateral displacement and time, with spatiotemporal collision-free conditions as constraints, candidate detour routes are searched. Using the Frenet coordinate system, the proposed detour route is decomposed into longitudinal motion components and lateral motion components; The lateral motion components are fitted using a polynomial curve to generate a lateral trajectory; The lateral trajectory is combined with the longitudinal motion component to obtain the detour route.

[0013] In one possible implementation, after obtaining the detour route, the method further includes: The comfort of the detour route is verified, which includes detecting whether the curvature, acceleration, and jerk of the detour route are within a preset comfort threshold throughout the entire route. If the detour route passes the comfort check, it will be used as the final detour route. If the detour route fails the comfort check, a new detour route from the current lane to the target lane to bypass the first obstacle is planned.

[0014] In one possible implementation, the method further includes: controlling the vehicle to travel along the detour route, changing from the current lane to the target lane, in order to detour around the first obstacle.

[0015] In one possible implementation, the method further includes: Continuously monitor the traffic environment around the vehicle to check for any unexpected safety incidents. If an unexpected safety event is detected, the current detour operation is suspended, and the process of determining whether the traffic environment around the vehicle meets the detour triggering conditions is re-executed.

[0016] The beneficial effects of this invention are: Compared with existing technologies, the active obstacle avoidance method based on three-dimensional search provided in this embodiment includes determining whether the traffic environment around the vehicle meets the obstacle avoidance triggering conditions. These triggering conditions include the presence of a first obstacle requiring detour in the vehicle's current lane, and the absence of a safe merging gap in the target lane due to a second obstacle. If the obstacle avoidance triggering conditions are met, the vehicle's longitudinal speed in the current lane is adjusted to regulate the actual longitudinal gap with the second obstacle. If the actual longitudinal gap is greater than or equal to the safe merging gap, a detour route from the current lane to the target lane to bypass the first obstacle is planned. Therefore, the method provided in this embodiment can actively create obstacle avoidance conditions even if the traffic environment around the vehicle does not meet the obstacle avoidance triggering conditions, by adjusting the vehicle's longitudinal speed in the current lane and thereby regulating the actual longitudinal gap between the vehicle and the second obstacle. Compared with current passive obstacle avoidance schemes, this method can improve traffic efficiency in high-density, continuous traffic flow by actively creating obstacle avoidance conditions, thus solving the problems of existing technologies. Attached Figure Description

[0017] Figure 1 A flowchart of an embodiment of the active insertion and bypass method based on three-dimensional search provided by the present invention; Figure 2 for Figure 1 In step S102, the flowchart of the method for constructing the time-longitudinal displacement map is shown. Figure 3 for Figure 1 A flowchart of a method according to an embodiment of step S103; Figure 4 for Figure 1 A schematic diagram of generating the horizontal trajectory in step S103; Figure 5 for Figure 1 The provided method includes a diagram illustrating the detour route. Figure 6 A schematic diagram of the active insertion and bypass device based on three-dimensional search provided by the present invention; Figure 7 This is a schematic diagram of the vehicle structure provided by the present invention; Figure 8 This is a schematic diagram of the structure of an autonomous driving system in a vehicle provided by the present invention. Detailed Implementation

[0018] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0019] As mentioned earlier, existing obstacle detour solutions are passive response solutions, which have a significant drawback: the decision-making process is reactive, requiring the detection of lateral space—that is, sufficient lateral safety space in the target lane—before detours can proceed. However, in high-density, continuous traffic flow, the target lane may lack this lateral safety space for extended periods. In such cases, existing obstacle detour solutions cannot perform detours and are forced to follow or even stop, leading to a severe decrease in traffic efficiency.

[0020] In view of this, embodiments of this application provide a method, apparatus, vehicle, and storage medium for active gap-filling detours based on three-dimensional search, which can be used to solve the problems in the prior art. For example... Figure 1 The diagram shows a detailed flowchart of the active insertion bypass method based on 3D search. The method includes the following steps: Step S101: Determine whether the traffic environment around the vehicle meets the detour triggering conditions. The detour triggering conditions include the presence of a first obstacle that needs to be detoured in the current lane where the vehicle is located, and the absence of a safe merging gap in the target lane due to the presence of a second obstacle.

[0021] In this application embodiment, a self-driving vehicle refers to a vehicle that needs to detour around an obstacle (referring to the first obstacle). Specifically, the self-driving vehicle in this application can detour around the first obstacle by executing the method provided in this application embodiment. In practical applications, a self-driving vehicle can refer to an intelligent vehicle equipped with a computing processor and various environmental perception modules (such as cameras, lidar, millimeter-wave radar, etc.). The type of self-driving vehicle is not specifically limited here.

[0022] The lane in which the vehicle is currently located is the current lane. When bypassing the first obstacle, the lane that needs to be merged into is the target lane. Generally speaking, the target lane can be the lane to the side of the current lane (either the left or right side). Therefore, in this embodiment, the vehicle bypasses the first obstacle by merging from the current lane into the target lane.

[0023] The first obstacle can refer to an obstacle located in the current lane and in front of the vehicle. The presence of this first obstacle prevents the vehicle from continuing along its original trajectory, thus requiring it to detour. In practical applications, the first obstacle can be a stationary obstacle, such as a stationary construction barrier or a disabled vehicle. It can also be a moving obstacle, such as a slow-moving vehicle traveling at a speed significantly lower than the vehicle's desired speed. For example, if the vehicle is traveling at 60 km / h in the current lane and a truck traveling at only 20 km / h appears 50 meters ahead, the truck is identified as the first obstacle.

[0024] The second obstacle can refer to an obstacle located within the target lane that prevents the vehicle from safely merging directly into the target lane to avoid the first obstacle. This second obstacle is typically a vehicle traveling behind or to the side of the target lane. In this embodiment, the presence of this second obstacle in the target lane temporarily prevents the vehicle from having a safe merging gap, making it difficult to merge directly from the current lane into the target lane. This temporary lack of a safe merging gap can mean that, at the current moment or within a preset short time window, the longitudinal distance or relative speed relationship between the vehicle and the second obstacle does not meet a preset safe lane-changing threshold, resulting in a collision risk if the vehicle attempts to change lanes directly.

[0025] It should be further explained that there are multiple ways to implement step S101 in practical applications. The first way is to execute step S101 when a detour command triggered by the user is received, in order to determine whether the traffic environment around the vehicle meets the detour triggering conditions.

[0026] The detour command triggered by the user can refer to a lane change or overtaking request actively issued by the driver or passenger through the in-vehicle human-machine interface. The source of the detour command can be signals input from various forms of input devices, such as physical levers on the steering wheel, virtual touch buttons on the central control screen, specific voice commands recognized by the voice control module (such as overtaking the vehicle in front), or specific hand gestures captured by the gesture recognition system.

[0027] When the vehicle's (referring to the vehicle's) intelligent driving system detects any of the aforementioned signals, it determines that it has received a detour command triggered by the user. At this time, the intelligent driving system can execute step S101, such as immediately activating the environmental perception module of the intelligent driving system (including LiDAR, cameras, and various sensors) to collect and analyze real-time traffic data around the vehicle at high frequency to determine whether the traffic environment around the vehicle meets the detour triggering conditions. Specifically, the environmental perception module of the intelligent driving system will quickly scan whether there is a first obstacle (such as a slow-moving vehicle, stationary construction cones, etc.) obstructing the vehicle's current lane, and at the same time detect whether there is a second obstacle (such as a vehicle rapidly approaching from behind) in the adjacent target lane, and calculate whether there is a safe merging gap in the target lane at the current moment. Here, the intelligent driving system can be the intelligent driving system installed on the vehicle, which can be an L1, L2, L3, L4 or other levels of intelligent driving system. The level of the intelligent driving system is not limited here.

[0028] For example, when a driver on a highway spots a truck traveling at a speed limit ahead and believes there is space to overtake in the right lane, they can pull down the right turn signal stalk and hold it for more than a preset time (e.g., 1.5 seconds). This action is interpreted as a clear lane change instruction. Upon receiving this instruction, the intelligent driving system immediately retrieves data from millimeter-wave radar and cameras to confirm the distance and relative speed of the vehicle to the right rear. If the vehicle to the right rear is too close and does not meet the safe merging clearance, the intelligent driving system will not immediately execute a lane change but will proceed to the subsequent steps in this application; if the safe merging clearance is met, it will directly plan a route. This response mechanism ensures that, in scenarios requiring manual intervention, the system respects the driver's intentions while strictly adhering to safety standards, avoiding blindly executing dangerous maneuvers.

[0029] The second implementation of step S101 can be to acquire the traffic environment around the vehicle continuously monitored by the intelligent driving system and determine whether the traffic environment around the vehicle meets the detour triggering conditions. In this second implementation, the intelligent driving system continuously monitors the traffic environment around the vehicle. For example, the intelligent driving system continuously collects and analyzes real-time traffic data around the vehicle at high frequency through the environmental perception module to determine whether the traffic environment around the vehicle meets the detour triggering conditions. In this implementation, the intelligent driving system can automatically identify whether a detour needs to be initiated.

[0030] For example, the intelligent driving system monitors the traffic environment around the vehicle every 100 milliseconds. If at a certain moment, the intelligent driving system detects that a slow vehicle ahead will force the vehicle's speed to drop below a threshold, it needs to detour to the target lane. However, if the target lane does not have a safe merging gap due to the presence of a second obstacle, the subsequent steps of the method provided in this application embodiment can be executed. Of course, if there is a safe merging gap in the target lane, the vehicle can directly merge into the target lane.

[0031] Of course, in practical applications, the first and second implementation methods mentioned above can be combined. For example, the environmental perception module in the intelligent driving system can continuously monitor the traffic environment around the vehicle to determine whether the traffic environment around the vehicle meets the detour triggering conditions, and during the continuous monitoring process, determine whether a detour command triggered by the user has been received.

[0032] Step S102: If the detour triggering conditions are met, adjust the actual longitudinal clearance with the second obstacle by controlling the longitudinal speed of the vehicle in the current lane.

[0033] Specifically, adjusting the longitudinal speed of the vehicle in the current lane can refer to changing its longitudinal speed by adjusting its own longitudinal acceleration (such as accelerating or decelerating) while keeping the vehicle in the current lane, thereby adjusting the actual longitudinal clearance between the vehicle and the second obstacle. This actual longitudinal clearance can refer to the real-time distance between the vehicle and the second obstacle in the target lane along the road's extension direction.

[0034] In step S102, the actual longitudinal clearance between the vehicle and the second obstacle is adjusted by regulating the vehicle's longitudinal speed in the current lane. This process is a dynamic interplay and creation process, the core of which lies in using the difference in longitudinal speed between the vehicle and the second obstacle to change their relative positional relationship. Specifically, for example, when the second obstacle is located to the side and rear of the vehicle and the distance is too close (less than the safe merging clearance), the intelligent driving system controls the vehicle to appropriately decelerate or accelerate, increasing the actual longitudinal distance to the second obstacle, ultimately creating a safe merging clearance. This process is not a passive response solution of existing technologies, but rather adjusts the actual longitudinal clearance with the second obstacle through the vehicle's active longitudinal speed regulation, thereby transforming a traffic situation that originally lacked lane-changing conditions into one that did, significantly improving the vehicle's maneuverability in dense traffic flow.

[0035] It is important to note that in practical applications, step S102 can be implemented in the following way: First, a longitudinal velocity curve can be planned on the time-longitudinal displacement graph using a dynamic programming algorithm, and then the vehicle can be controlled to travel according to the longitudinal velocity curve, thereby adjusting the actual longitudinal gap between the vehicle and the second obstacle.

[0036] The time-longitudinal displacement diagram (ST diagram) reflects the dynamic longitudinal positional relationship between the vehicle and a second obstacle in the target lane over time. For example, the ST diagram can be a two-dimensional coordinate system, where the horizontal axis represents time and the vertical axis represents the displacement of the vehicle or the target lane in the longitudinal direction of the road. This diagram characterizes the dynamic evolution of the relative positional relationship between the vehicle and the second obstacle (i.e., interfering vehicle) in the target lane over time. In this embodiment, the second obstacle is represented in the ST diagram as a no-passage area or polygonal block that extends over time, and its slope reflects the obstacle's speed.

[0037] In the process of planning the longitudinal velocity curve on the time-longitudinal displacement graph using dynamic programming, multiple speed adjustment strategies can be searched forward on the time-longitudinal displacement graph first. Each speed adjustment strategy corresponds to a path on the ST graph starting from the current state. After obtaining multiple speed adjustment strategies, the cumulative cost of each speed adjustment strategy in the future time step (e.g., 3-5 seconds) can be evaluated using the dynamic programming algorithm Cost = α × comfort + β × clearance + γ × progress. In this formula, Cost is the cumulative cost; comfort is the comfort cost term used to penalize acceleration and jerk; clearance is the safety cost term used to reward maintaining a safe distance from vehicles in front and behind; progress is the traffic efficiency cost term used to encourage driving speed; and α, β, and γ are weighting coefficients.

[0038] The dynamic programming algorithm Cost = α × comfort + β × clearance + γ × progress evaluates the merits of each speed adjustment strategy by calculating its cumulative cost. This cumulative cost comprehensively considers comfort (comfort cost), safety (safety cost), and traffic efficiency (traffic efficiency cost). For example, if a speed adjustment strategy causes excessive acceleration, its comfort cost increases, leading to a higher Cost; if the strategy results in the vehicle's distance from the second obstacle falling below a safe threshold, its safety cost increases significantly. Ultimately, the cumulative cost of each speed adjustment strategy over future time steps can be obtained using the dynamic programming algorithm Cost = α × comfort + β × clearance + γ × progress. The speed adjustment strategy with the lowest cumulative cost is then selected, and the longitudinal speed curve corresponding to this strategy is used as the planned longitudinal speed curve.

[0039] Furthermore, this method can further include selecting corresponding values ​​for α, β, and γ based on the vehicle's driving style. The driving style can refer to the vehicle's autonomous driving system's preset or user-defined behavioral mode configuration, used to characterize the vehicle's preferences for safety, comfort, and traffic efficiency during driving. In practical applications, this driving style specifically includes various types such as conservative and aggressive, and its source can be a factory-fixed default setting or a parameter combination selected and determined by the driver in real time through the in-vehicle human-machine interface. The driving style serves as the mapping basis for the weighting coefficients α, β, and γ. For example, when the selected driving style is conservative, the intelligent driving system will select a larger β value and a smaller γ value, while setting a moderate α value. This allows clearance (safety cost) to dominate the cumulative cost, while the weight of progress (traffic efficiency cost) is weakened. This reduces the algorithm's sensitivity to speed loss penalties, ensuring that the actual longitudinal clearance between the vehicle and a second obstacle remains above a high safety threshold in scenarios with high traffic density, thus minimizing collision risk. Conversely, when the selected driving style is aggressive, the intelligent driving system will select a larger γ value and a relatively smaller β value. In this case, progress dominates the cumulative cost, while the weight of clearance is reduced, which means that the vehicle is allowed to cut into the target lane with a tighter longitudinal gap while meeting relatively small safety distance constraints.

[0040] Of course, after obtaining the longitudinal speed curve, the autonomous vehicle's intelligent driving system can further control the vehicle to travel according to the longitudinal speed curve to adjust the actual longitudinal clearance between the vehicle and the second obstacle. This longitudinal speed curve contains a series of target speed values ​​or acceleration commands with timestamps. Therefore, the autonomous vehicle's intelligent driving system can adjust the vehicle's speed based on this, changing the relative motion between the vehicle and the second obstacle in the target lane, and thus dynamically adjusting the actual longitudinal clearance between them. For example, if the planned longitudinal speed curve includes a moderate deceleration process, the vehicle will pull back a greater distance relative to a faster-moving second obstacle, or slow down its approach rate relative to a slower-moving second obstacle, thereby adjusting the actual longitudinal clearance between the vehicle and the second obstacle.

[0041] It needs to be further explained that, such as Figure 2 As shown, the time-longitudinal displacement diagram can also be constructed in the following manner according to the embodiments of this application, which includes the following steps: Step S1021: Obtain the predicted motion information of the vehicle and the second obstacle within the future time step.

[0042] The predicted motion information may include a set of motion state data of the vehicle and a second obstacle in the target lane that may appear over a future period of time. Specifically, this includes parameters such as time points, position coordinates, longitudinal velocity, longitudinal acceleration, and lateral motion trends. In practical applications, this predicted motion information can be collected by the vehicle's intelligent driving system, which controls environmental perception modules such as cameras and lidar to collect relevant data. This data may include a series of time points and corresponding positions of the second obstacle (or a series of time points and corresponding positions of the vehicle itself). This data can then be input into the prediction module of the intelligent driving system to calculate the predicted motion information.

[0043] The prediction module typically employs an interactive motion prediction model based on Gaussian process regression or recurrent neural networks to infer the behavioral intentions of traffic participants (including the vehicle and the second obstacle), thereby outputting a future trajectory prediction with a probability distribution. For example, the prediction module in an intelligent driving system can predict the expected longitudinal position and its confidence interval of the second obstacle at each moment within a set future time step, with a resolution of 0.1 seconds, based on data collected by the environmental perception module, as the predicted motion information.

[0044] In step S1021, the purpose of acquiring this predicted motion information is to provide forward-looking input data for the subsequent construction of the spatiotemporal environment model. This enables the intelligent driving system to anticipate potential conflict points or available safety gaps, avoiding short-sighted decisions based solely on the current instantaneous state. Furthermore, the simultaneous acquisition of predicted motion information for both the vehicle and the second obstacle ensures that their relative motion analysis is performed on a unified time reference, providing a data foundation for creating safe merging gaps.

[0045] Step S1022: Establish a two-dimensional coordinate system with time as the horizontal axis and the longitudinal displacement of the current lane or target lane as the vertical axis.

[0046] This two-dimensional coordinate system is a mathematical model used to characterize the longitudinal motion of a vehicle and its temporal relationship; it forms the basic framework of a time-longitudinal displacement (ST) graph. The horizontal axis of this coordinate system represents time (T), and the vertical axis represents longitudinal displacement (S). Its origin is typically defined as the longitudinal position of the vehicle at the current decision-making moment. The coordinate system is established based on the road topology. If the focus is on analyzing the interaction between the vehicle and vehicles in the target lane, the vertical axis is mapped to the longitudinal mileage marker of the target lane; if comparing the current lane following situation, it can be mapped to the current lane's longitudinal mileage. This two-dimensional coordinate system provides a unified quantification space, simplifying complex three-dimensional motion into a trajectory evolution problem on a two-dimensional plane. This allows for intuitive geometric comparison and collision detection of the motion states of different traffic participants within the same view.

[0047] Step S1023: In the two-dimensional coordinate system, based on the predicted motion information, the predicted trajectory regions corresponding to the vehicle and the second obstacle within the future time step are generated respectively to obtain the time-longitudinal displacement map.

[0048] After establishing a two-dimensional coordinate system through step S1022, in step S1023, based on the predicted motion information from step S1021, the predicted trajectory regions corresponding to the vehicle and the second obstacle within the future time step are generated in the two-dimensional coordinate system, thereby obtaining the time-longitudinal displacement map.

[0049] The predicted trajectory region refers to the spatiotemporal occupied area bounded by the predicted motion information of the vehicle and the second obstacle in the aforementioned two-dimensional coordinate system. Its generation process involves projecting the predicted positions and uncertainty ranges at discrete time points in the predicted motion information onto the two-dimensional coordinate system. Furthermore, interpolation or envelope fitting can be used to form continuous region blocks. The resulting time-longitudinal displacement map not only includes static road information but also dynamically integrates the future movement trends of surrounding vehicles, forming an environmental model that includes prohibited areas. This ensures that subsequently planned detour routes can proactively avoid the predicted occupied area of ​​the second obstacle, creating a safe entry window in the spatiotemporal dimension.

[0050] Step S103: If the actual longitudinal clearance is greater than or equal to the safe merging clearance, plan a detour route from the current lane to the target lane to bypass the first obstacle.

[0051] The safe merging gap can refer to the minimum permissible longitudinal distance threshold calculated based on the vehicle's dimensions, the speed of the second obstacle, the relative speed, and a preset safe time headway. The planned detour route can be a complete driving trajectory generated after confirming that the longitudinal space meets safety requirements (i.e., the actual longitudinal gap is greater than or equal to the safe merging gap). This trajectory includes longitudinal displacement, lateral displacement, and time dimensions. The starting point of this trajectory is the vehicle's real-time state in the current lane (i.e., the vehicle's current state), and the ending point is the target state after the vehicle has fully entered the target lane and passed the first obstacle. The vehicle's current state includes its current longitudinal position, lateral position, longitudinal speed, lateral speed, and current time. The target state refers to the desired state of the vehicle after completing the detour around the first obstacle in the target lane, typically including a specific longitudinal position within the target lane, zero lateral offset (i.e., returning to the center of the lane), and the desired driving speed.

[0052] The specific implementation of step S103 can be achieved by planning a detour route from the current state of the vehicle to the target state after completing the detour around the first obstacle in the target lane within a three-dimensional spatiotemporal state grid composed of longitudinal displacement, lateral displacement and time, with spatiotemporal collision-free conditions as the constraint.

[0053] The three-dimensional spatiotemporal state grid refers to a data structure that models the vehicle's motion planning problem in a three-dimensional space containing longitudinal displacement (s), lateral displacement (d), and time (t). The longitudinal displacement axis of this three-dimensional spatiotemporal state grid represents the distance the vehicle travels along the road centerline, the lateral displacement axis represents the lateral offset of the vehicle relative to the lane centerline, and the time axis represents the planning time span. In practical applications, the environmental perception module of the intelligent driving system can acquire static obstacles (such as road boundaries and construction zones) and dynamic obstacles (such as other vehicles) around the vehicle, predict their trajectories over a future period, and project the prediction results into a three-dimensional coordinate system to generate the desired result.

[0054] When planning this detour route, the constraint of no collision in time and space is required. Specifically, during the planning process, any point (s, d, t) on the generated detour route cannot fall into the prohibited passage area. This means that not only is it required that the vehicle does not overlap with the first obstacle and the second obstacle in spatial position, but it is also required that it does not overlap with the first obstacle and the second obstacle at the same spatial point at the same time, thus solving the coupling problem of when to change lanes and how to change lanes.

[0055] It needs to be further explained that, such as Figure 3 As shown, in a three-dimensional spatiotemporal state grid composed of longitudinal displacement, lateral displacement, and time, with spatiotemporal collision-free conditions as the constraint, a detour route from the current state of the vehicle to the target state after completing the detour around the first obstacle in the target lane is planned. The specific implementation method may include the following steps: Step S1031: In a three-dimensional spatiotemporal state grid composed of longitudinal displacement, lateral displacement and time, the candidate detour route is searched under the constraint of spatiotemporal collision-free.

[0056] The proposed detour route is determined by using a dynamic programming algorithm or a heuristic search algorithm to search for one or more feasible path sequences from the vehicle's current state to the target state, under the premise of satisfying spatiotemporal collision-free constraints. Specifically, the search process evaluates the cumulative cost of each state node from the vehicle's current state to the target state and avoids the spatiotemporal regions occupied by the first and second obstacles, thereby initially determining a coarse path that is geometrically and temporally safe. This step S1031 aims to solve the global reachability problem in complex dynamic environments, providing a basic solution space for subsequent trajectory optimization.

[0057] Step S1032: Using the Frenet coordinate system, decompose the candidate detour route into longitudinal motion components and lateral motion components.

[0058] The Frenet coordinate system is a curved coordinate system established along the centerline of a road, with its longitudinal axis s extending along the reference line and its transverse axis d perpendicular to the reference line.

[0059] In step S1032, the candidate detour route is decomposed into longitudinal and lateral motion components. This may involve transforming the candidate detour route from the global Cartesian coordinate system to the Frenet coordinate system, thereby separating the longitudinal motion component describing the vehicle's distance traveled along the road and its speed changes, and the lateral motion component describing the vehicle's lateral deviation and lane-changing behavior relative to the lane centerline. The longitudinal motion component mainly focuses on the vehicle's speed planning in the time dimension to match the gap creation in the target lane, while the lateral motion component focuses on the vehicle's smooth lane changes in the spatial dimension. This decomposition method decouples the complex three-dimensional spatiotemporal trajectory planning problem into two relatively independent one-dimensional planning problems, significantly reducing computational complexity and making the subsequent planning results more closely match the road's geometry.

[0060] Step S1033: Use a polynomial curve to fit the lateral motion component to generate a lateral trajectory.

[0061] Since the lateral motion components decomposed above may only be a series of discrete planning points, direct execution may result in abrupt vehicle steering. Therefore, in step S1033 of this application, a polynomial curve is further used to fit the lateral motion components to generate a lateral trajectory. The polynomial curve can be a cubic polynomial curve, a quintic polynomial curve, etc., which can be used to smooth the decomposed lateral motion components.

[0062] Specifically, the fifth-order polynomial curve can be represented as d(t) = a0 + a1 × t + a2 × t 2 +a3×t 3 +a4×t 4 +a5×t 5Where d(t) is the lateral displacement at time t, for the coefficients a0~a5 in the fifth-order polynomial curve, several sets of boundary conditions {t=0, d(0)=0} and {t=T, d(T)=D} can be obtained by setting the lateral displacement, lateral velocity, and lateral acceleration (usually all of which are 0) of the current state and the target state, where T is the total time required to complete the lane change and D is the distance from the center line of the current lane to the center line of the target lane. Then, by combining the boundary conditions with the first derivative of d(t) as the lateral velocity and the second derivative as the lateral acceleration, a0~a5 can be calculated, and finally the lateral trajectory d(t) = a0 + a1×t + a2×t is obtained. 2 +a3×t 3 +a4×t 4 +a5×t 5 .

[0063] for example Figure 4 As shown, the fifth-order polynomial curve d(t) = a0 + a1×t + a2×t is used. 2 +a3×t 3 +a4×t 4 +a5×t 5 To fit the lateral motion components, an example of a lateral trajectory is generated. In this example, d(0)=0, d'(0)=0, d''(0)=0, indicating that at t=0, the lateral displacement d(0), lateral velocity d'(0), and lateral acceleration d''(0) are all 0; at t=T=4, the lateral displacement d(4)=3.5 meters, the lateral velocity d'(4)=0, and the lateral acceleration d''(4)=0; and at the midpoint t=2, the lateral displacement d(2)=1.77 meters, the lateral velocity d'(2)=1.64 m / s, and the lateral acceleration d''(2)=0. Through these conditions, a0~a5 can be calculated, where the values ​​of a0, a1, and a2 are all 0, a3=0.547, a4=-0.205, and a5=0.021, thus finally obtaining the lateral trajectory d(t)=0.547×t. 3 -0.205×t 4 +0.021×t 5 .

[0064] Of course, by using cubic polynomial curves or other polynomial curves, it is also possible to fit the lateral motion component and ultimately generate the lateral trajectory. Here, the type of polynomial curve is not specifically limited.

[0065] Step S1034: Combine the lateral trajectory with the longitudinal motion component to obtain the detour route.

[0066] In step S1034, the synthesis of the lateral trajectory and longitudinal motion components can be achieved by aligning and recombining the lateral trajectory generated in step S1033 and the longitudinal motion components obtained in step S1032 on the time axis. For example, at time t, the corresponding longitudinal displacement s(t), longitudinal velocity v_s(t), lateral displacement d(t), and lateral velocity v_d(t) can be combined and restored to the vehicle's global Cartesian coordinate system through coordinate transformation to form a complete pose sequence containing position, attitude, velocity, and acceleration information, thus obtaining the detour route. The resulting detour route not only satisfies the collision-free constraint in the three-dimensional spatiotemporal grid but also possesses excellent smoothness and dynamic feasibility.

[0067] Of course, after obtaining the detour route through step S1034 above, the method can further include controlling the vehicle to travel along the detour route and change from the current lane to the target lane to detour around the first obstacle. Specifically, the vehicle's intelligent driving system can generate a series of control commands that can be recognized by the vehicle's underlying actuators based on the detour route, including underlying control quantities such as steering wheel angle, speed, acceleration, and braking pressure, thereby driving the vehicle chassis system to strictly reproduce the lateral and longitudinal motion components in the planned trajectory, thereby controlling the vehicle to travel along the detour route, and finally controlling the vehicle to change from the current lane to the target lane to detour around the first obstacle.

[0068] like Figure 5 As shown, after obtaining the detour route, the vehicle can travel from the starting point of the current lane (at which point t=0, and the lateral displacement d and longitudinal displacement s are also 0) along the detour route for 4.5s, and then change from the current lane to the end point of the target lane, thereby detouring around the first obstacle. At the end point, t=4.5s, lateral displacement d=3.5 meters, and longitudinal displacement s=80 meters.

[0069] Compared with existing technologies, the active obstacle avoidance method based on three-dimensional search provided in this embodiment includes determining whether the traffic environment around the vehicle meets the obstacle avoidance triggering conditions. These triggering conditions include the presence of a first obstacle requiring detour in the vehicle's current lane, and the absence of a safe merging gap in the target lane due to a second obstacle. If the obstacle avoidance triggering conditions are met, the vehicle's longitudinal speed in the current lane is adjusted to regulate the actual longitudinal gap with the second obstacle. If the actual longitudinal gap is greater than or equal to the safe merging gap, a detour route from the current lane to the target lane to bypass the first obstacle is planned. Therefore, the method provided in this embodiment can actively create obstacle avoidance conditions even if the traffic environment around the vehicle does not meet the obstacle avoidance triggering conditions. This solves the problem of low traffic efficiency in high-density, continuous traffic flow caused by existing passive obstacle avoidance schemes, thus improving traffic efficiency by actively creating obstacle avoidance conditions.

[0070] It should be noted that after obtaining the detour route through step S103 above, and before controlling the vehicle to travel along the detour route, the method may further include performing a comfort check on the detour route. The comfort check includes detecting whether the curvature, acceleration, and jerk of the detour route are within a preset comfort threshold (i.e., less than the comfort threshold) throughout the entire route. If the curvature, acceleration, and jerk of the detour route are within the preset comfort threshold throughout the entire route, it indicates that the detour route has passed the comfort check. This means that the passenger's comfort meets the requirements when traveling along the detour route. Therefore, the detour route can be used as the final detour route and then used by the autonomous driving system to control the vehicle to travel along the detour route.

[0071] Conversely, if the curvature, acceleration, and jerk of the detour route are not within the preset comfort threshold throughout the entire route, it means that the detour route has failed the comfort test, and that the passenger's comfort does not meet the requirements when detouring according to the detour route. In this case, a new detour route from the current lane to the target lane to bypass the first obstacle can be planned, for example, according to the above step S103.

[0072] It should be further explained that, during the process of detouring around the first obstacle, the method provided in this application embodiment is further equipped with a fault tolerance mechanism to improve the safety of detouring. Specifically, it can continuously monitor whether unexpected safety events occur in the traffic environment around the vehicle. These unexpected safety events may include sudden situations that occur in the traffic environment after the detouring trajectory planning is completed and execution begins, which are not covered by the initial prediction model and pose a potential collision risk or serious disruption to comfort of the current detouring operation. For example, these unexpected safety events may include the second obstacle suddenly accelerating, decelerating or changing direction, or pedestrians crossing or entering the construction area.

[0073] If no unexpected safety event is detected, the current detour operation can continue, eventually changing from the current lane to the target lane to detour around the first obstacle. Conversely, if an unexpected safety event is detected, the current detour operation needs to be terminated, and step S101 needs to be executed again to determine whether the traffic environment around the vehicle meets the detour triggering condition. If the detour triggering condition is still met, the current detour operation can continue. Otherwise, if the detour triggering condition is not met, steps S102 and S103 can be executed again.

[0074] Based on the same inventive concept as the active gap-filling bypass method based on three-dimensional search provided in the embodiments of this application, the embodiments of this application can also provide an active gap-filling bypass device based on three-dimensional search. For any unclear aspects regarding this device embodiment, please refer to the relevant content of the method embodiment. Figure 6 As shown, the active insertion and bypass device (hereinafter referred to as device 20) may include: a judgment unit 201, a control unit 202, and a planning unit 203, wherein: The judgment unit 201 is used to determine whether the traffic environment around the vehicle meets the detour triggering conditions. The detour triggering conditions include that there is a first obstacle that needs to be detoured in the current lane where the vehicle is located, and that there is no safe merging gap in the target lane due to the presence of a second obstacle. The control unit 202 is used to adjust the actual longitudinal clearance with the second obstacle by adjusting the longitudinal speed of the vehicle in the current lane when the detour triggering condition is met. Planning unit 203 is used to plan a detour route from the current lane to the target lane to bypass the first obstacle when the actual longitudinal gap is greater than or equal to the safe merging gap.

[0075] Since the device 20 adopts the same inventive concept as the method provided in the embodiments of this application, and the device 20 can also solve the technical problem if the method can solve the prior art problem, this will not be elaborated here.

[0076] like Figure 7 As shown, the present invention also provides a vehicle 300, which includes a processor 301 and a memory 302. Figure 7 Only some components of vehicle 300 are shown, but it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0077] In some embodiments, processor 301 may be a central processing unit (CPU), microprocessor, or other data processing chip, used to run program code stored in memory 302 or process data, such as the active insertion bypass method based on three-dimensional search in this invention. In some embodiments, processor 301 may be a single server or a group of servers. The server group may be centralized or distributed. In some embodiments, processor 301 may be local or remote. In some embodiments, processor 301 may be implemented on a cloud platform. In one embodiment, the cloud platform may include a private cloud, public cloud, hybrid cloud, community cloud, distributed cloud, internal cloud, multi-cloud, etc., or any combination thereof.

[0078] In some embodiments, memory 302 may be an internal storage unit of vehicle 300, such as a hard drive or memory of the autonomous vehicle 300. In other embodiments, memory 302 may be an external storage device of the autonomous vehicle 300, such as a plug-in hard drive, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the autonomous vehicle 300. Furthermore, memory 302 may include both internal storage units and external storage devices of the vehicle 300. Memory 302 is used to store application software and various types of data installed on the vehicle 300.

[0079] The vehicle 300 also includes a computer program stored in a memory 302 and executable on a processor 301. When the computer program is executed by the processor 302, it implements the active insertion bypass method based on three-dimensional search in the embodiments of this application.

[0080] Of course, the vehicle 300 may also include a display 303, which in some embodiments may be an LED display, a liquid crystal display, a touch-screen liquid crystal display, or an OLED (Organic Light) display. Emitting Diode (Organic Light Emitting Diode) touchscreen, etc. Display 303 is used to display information about vehicle 300 and to display a visual user interface. Component 301 of vehicle 300 303 communicates with each other via the system bus.

[0081] like Figure 8 As shown, this application embodiment also provides an autonomous driving system 40 for a vehicle, which may include the following core modules: Perception module 401: It includes sensors such as cameras, LiDAR, and millimeter-wave radar, and can fuse the data collected by these sensors to identify and track static obstacles (such as road boundaries and construction zones) and dynamic obstacles (such as other vehicles and pedestrians) in the surrounding environment, and output their precise position, speed, acceleration, size and other status information. Prediction Module 402: Based on the output of the perception module, this module establishes an interactive, probabilistic motion prediction model (e.g., using Gaussian process regression or a model based on a recurrent neural network). This module predicts the possible trajectories and probability distributions of other traffic participants within the next 3-5 seconds, providing crucial forward-looking information for the planning module.

[0082] The decision-making and planning module 403 includes a behavior decision layer and a trajectory planning layer. The behavior decision layer is used to execute step S12 of this application, and the trajectory planning layer is used to execute step S13 of this application.

[0083] The control interface module 404 is used to convert the detour route output by the trajectory planning layer into control commands that the vehicle's underlying controller can understand and send them for execution.

[0084] This application also provides a computer-readable storage medium for storing a computer-readable program or instruction. When the program or instruction is executed by a processor, it can implement the steps or functions provided in the methods described above.

[0085] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0086] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for active gap-filling bypassing based on three-dimensional search, characterized in that, include: Determine whether the traffic environment around the vehicle meets the detour triggering conditions. The detour triggering conditions include the presence of a first obstacle that needs to be detoured in the current lane where the vehicle is located, and the absence of a safe merging gap in the target lane due to the presence of a second obstacle. If the detour triggering conditions are met, the actual longitudinal clearance with the second obstacle is adjusted by regulating the longitudinal speed of the vehicle in the current lane. If the actual longitudinal clearance is greater than or equal to the safe merging clearance, a detour route is planned from the current lane to the target lane to bypass the first obstacle.

2. The method according to claim 1, characterized in that, Adjusting the actual longitudinal clearance with the second obstacle by regulating the vehicle's longitudinal speed in the current lane includes: A longitudinal velocity curve is planned on a time-longitudinal displacement map using a dynamic programming algorithm, wherein the time-longitudinal displacement map reflects the dynamic longitudinal positional relationship between the vehicle and the second obstacle on the target lane over time. The vehicle is controlled to travel according to the longitudinal speed curve in order to adjust the actual longitudinal clearance with the second obstacle.

3. The method according to claim 2, characterized in that, The longitudinal velocity curve is planned on the time-longitudinal displacement graph using a dynamic programming algorithm, specifically including: Search forward for multiple velocity adjustment strategies on the time-longitudinal displacement graph; The cumulative cost of each speed adjustment strategy over future time steps is evaluated using the dynamic programming algorithm Cost = α × comfort + β × clearance + γ × progress. Cost is the cumulative cost; comfort is the comfort cost term used to penalize acceleration and jerk; clearance is the safety cost term used to reward maintaining a safe distance from vehicles in front and behind; and progress is the traffic efficiency cost term used to encourage driving speed. α, β, and γ are weighting coefficients. The longitudinal speed curve corresponding to the speed adjustment strategy with the lowest cumulative cost is selected as the planned longitudinal speed curve.

4. The method according to claim 3, characterized in that, The method further includes: selecting corresponding values ​​for α, β, and γ based on the vehicle's driving style, wherein the driving style includes conservative and aggressive driving styles.

5. The method according to claim 1, characterized in that, Determine whether the traffic environment around the vehicle meets the detour trigger conditions, specifically including: Upon receiving a detour command triggered by a user, determine whether the traffic environment surrounding the vehicle meets the detour triggering conditions; and / or, The system continuously monitors the traffic environment around the vehicle and determines whether the traffic environment around the vehicle meets the detour triggering conditions.

6. The method according to claim 1, characterized in that, Planning a detour route from the current lane to the target lane to bypass the first obstacle specifically includes: in a three-dimensional spatiotemporal state grid composed of longitudinal displacement, lateral displacement and time, with spatiotemporal collision-free conditions as constraints, planning a detour route from the current state of the vehicle to the target state after bypassing the first obstacle in the target lane.

7. The method according to claim 6, characterized in that, In a three-dimensional spatiotemporal state grid composed of longitudinal displacement, lateral displacement, and time, and with spatiotemporal collision-free conditions as constraints, a detour route is planned from the vehicle's current state to the target state after completing the detour around the first obstacle in the target lane. Specifically, this includes: In a three-dimensional spatiotemporal state grid composed of longitudinal displacement, lateral displacement and time, with spatiotemporal collision-free conditions as constraints, candidate detour routes are searched. Using the Frenet coordinate system, the proposed detour route is decomposed into longitudinal motion components and lateral motion components; The lateral motion components are fitted using a polynomial curve to generate a lateral trajectory; The lateral trajectory is combined with the longitudinal motion component to obtain the detour route.

8. The method according to claim 7, characterized in that, After obtaining the detour route, the method further includes: The comfort of the detour route is verified, which includes detecting whether the curvature, acceleration, and jerk of the detour route are within a preset comfort threshold throughout the entire route. If the detour route passes the comfort check, it will be used as the final detour route. If the detour route fails the comfort check, a new detour route from the current lane to the target lane to bypass the first obstacle is planned.

9. The method according to claim 1, characterized in that, The method further includes: controlling the vehicle to travel along the detour route and changing from the current lane to the target lane to detour around the first obstacle.

10. The method according to claim 9, characterized in that, The method further includes: Continuously monitor the traffic environment around the vehicle to check for any unexpected safety incidents. If an unexpected safety event is detected, the current detour operation is suspended, and the process of determining whether the traffic environment around the vehicle meets the detour triggering conditions is re-executed.

Images