A motion planning method considering visibility costs in urban multi-blind-spot scenarios

Abstract

This invention relates to a motion planning method considering visibility costs in urban multi-blind-spot scenarios. The method first extracts all blind spots in the scene, generates a blind spot boundary set based on road topology, and then filters out key blind spots according to the vehicle's driving intention. Subsequently, it initializes, grows, updates, and predicts a subset of states of phantom traffic participants within the key blind spots, then generates candidate paths. Based on this subset of states, it calculates the visibility cost of each path, combines it with other costs to obtain the total cost, and selects the path with the minimum cost as the optimal path. Finally, using the optimal path as a reference, it searches for discrete acceleration actions to generate a velocity curve that satisfies the constraints; these two together constitute the vehicle's motion plan. Compared with existing technologies, this invention has the advantages of simultaneously considering safety, traffic efficiency, and comfort.

Description

Technical Field

[0001] This invention relates to the field of motion planning technology, and in particular to a motion planning method that considers visibility costs in urban multi-blind zone scenarios. Background Technology

[0002] Due to limitations in the sensor detection capabilities of autonomous vehicles and the uncertainties in environmental perception, vehicles encounter blind spots created by static and dynamic obstacles (such as roadside buildings and slow-moving trucks). These blind spots may contain phantom traffic participants whose movement cannot be accurately captured, posing a significant challenge to the vehicle's motion planning and making it difficult to balance traffic efficiency, comfort, and safety. Therefore, properly addressing blind spots and designing appropriate planning methods will offer the possibility of achieving a balance between safety, comfort, and efficiency in urban scenarios with multiple blind spots, improving traffic efficiency without sacrificing safety and comfort.

[0003] In existing technologies, one type of invention incorporates vehicle-to-everything (V2X) technology to obtain information about phantom traffic participants in blind spots through information exchange between vehicles or between vehicles and infrastructure. For example, patent CN114932902B designs a pedestrian "ghost peek" warning and avoidance system based on V2X technology to detect blind spot risks; patent CN115052266B transmits the status information of phantom traffic participants in blind spots to the vehicle through communication between the vehicle and other vehicles; and patent CN120348231A utilizes communication between vehicles and infrastructure to effectively monitor blind spots around large vehicles. These patents utilize V2X communication to share blind spot perception results and leverage roadside sensors to enhance vehicle perception capabilities. However, these methods have high requirements for network bandwidth, communication latency, and infrastructure deployment, resulting in high costs and making large-scale adoption difficult in the short term.

[0004] Another type of invention estimates the states of phantom traffic participants that may exist in blind spots, ensuring the safe driving of the vehicle while the environment is partially observable. The reachability set method is a common strategy, representing all possible states of phantom traffic participants as a set. References 1, 2, and 3 all employ the most conservative approach, considering the vehicle safe only if its trajectory has no intersection with the reachability sets of phantom traffic participants. Most of these methods assume the vehicle travels along a predetermined reference path, focusing only on longitudinal speed planning. However, active lateral offset by the vehicle has a significant effect on increasing the field of view and reducing blind spot occlusion; only a few studies consider active offset by the vehicle to improve visibility. Reference 4 uses the vehicle's field of view angle as a visibility cost to guide dynamic overtaking decisions, Reference 5 proposes using a particle method to measure the visibility cost with collision probability, and Reference 6 estimates the visible area of ​​the intersection and converts it into a visibility cost. Although these methods consider visibility metrics, they cannot effectively handle the conflict of visibility costs between blind spots in urban multi-blind-spot scenarios, potentially leading to unstable vehicle paths and phenomena such as random offsets or no offset at all.

[0005] Existing literature references:

[0006] Document 1: "R. Poncelet, A. Verroust-Blondet and F. Nashashibi. SafeGeometric Speed ​​Planning Approach for Autonomous Driving through OccludedIntersections[C]. IEEE International Conference on Control, Automation, Robotics and Vision (ICARCV), 2020, pp. 393-399";

[0007] Document 2: "JMG Sánchez, T. Nyberg, C. Pek, J. Tumova and M. Törngren. Foresee the Unseen: Sequential Reasoning about Hidden Obstacles for Safe Driving[C]. IEEE Intelligent Vehicles Symposium (IV), 2022, pp. 255-264";

[0008] Document 3: "H. Park, J. Choi, H. Chin, SH Lee and D. Baek. Occlusion-Aware Risk Assessment and Driving Strategy for Autonomous Vehicles Using Simplified Reachability Quantification[J]. IEEE Robotics and Automation Letters, 2023";

[0009] Document 4: "H. Andersen et al. Trajectory optimization for autonomous overtaking with visibility maximization[C]. IEEE International Conference on Intelligent Transportation Systems (ITSC), 2017";

[0010] Document 5: "L. Wang, C. Burger and C. Stiller. Reasoning about PotentialHidden Traffic Participants by Tracking Occluded Areas[C]. IEEE International Intelligent Transportation Systems Conference (ITSC), 2021";

[0011] Document 6: "P. Narksri, H. Darweesh, E. Takeuchi, Y. Ninomiya and K. Takeda. Occlusion-Aware Motion Planning With Visibility Maximization viaActive Lateral Position Adjustment[J]. IEEE Access, 2022";

[0012] In summary, existing motion planning methods cannot simultaneously address safety, traffic efficiency, and comfort in urban scenarios with multiple blind spots, particularly in resolving conflicts arising from visibility costs. Summary of the Invention

[0013] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a motion planning method that considers the cost of visibility in urban multi-blind zone scenarios.

[0014] The objective of this invention can be achieved through the following technical solutions:

[0015] According to one aspect of the present invention, an active offset motion planning method for urban multi-blind-spot scenarios is provided, the method steps including:

[0016] S1. Extract all blind spots in the scene, and generate a set of blind spot boundaries by combining the extracted blind spots with the road topology in the scene;

[0017] S2. Based on the vehicle's driving intention, key blind spots are selected from the set of blind spot boundaries;

[0018] S3. Initialize, grow, update and predict the state subset of all phantom traffic participants in the critical blind zone;

[0019] S4. Generate candidate paths. Based on the state subset of phantom traffic participants, calculate the visibility cost of each candidate path, and combine the visibility cost with other costs to calculate the total cost of each candidate path; select the path with the minimum total cost as the optimal path.

[0020] S5. Using the optimal path as the reference path for velocity planning, search for discrete acceleration actions, generate velocity curves that satisfy the constraints, and the optimal path and velocity curves together constitute the motion planning of the vehicle.

[0021] As a preferred technical solution, S1 extracts all blind spots in the scene using the geometric ray method. First, dynamic and static occlusions are treated as polygons. Then, with the vehicle center as the origin, rays are emitted towards the dynamic and static occlusions respectively and connected to their vertices. The two rays with the largest included angle are selected and extended to the maximum sensing range of 100m, and the ends of the rays are connected to form the corresponding blind spot areas.

[0022] The specific process of generating the blind spot boundary set by combining the road topology in the scene is as follows: the intersection operation of the blind spot area with the driving corridors of all phantom traffic participants is performed to obtain the blind spot boundary set.

[0023] As a preferred technical solution, the specific process of S2 is as follows: combining the vehicle's intention, constructing the vehicle's driving corridor, and sequentially analyzing the forward driving corridors corresponding to each boundary segment in the blind spot boundary set, and screening out the key blind spots that affect the future driving of the vehicle.

[0024] The specific criteria for screening critical blind spots that affect future vehicle travel include: when the phantom traffic participant is a vehicle, if the forward travel corridor corresponding to the boundary segment is completely contained within the vehicle's travel corridor, then the boundary segment is considered to be in the same direction as the vehicle's movement and does not pose a risk, and the blind spot boundary is removed; if the forward travel corridor corresponding to the boundary segment intersects with the vehicle's travel corridor, then the phantom traffic participant in the blind spot is considered to have a risk of future conflict with the vehicle, and the blind spot boundary is retained; when the phantom traffic participant is a pedestrian, the boundary segment is considered a critical blind spot boundary; the critical blind spot is determined by the critical blind spot boundary.

[0025] As a preferred technical solution, the specific process of S3 includes:

[0026] Initialize the state set. For vehicles, construct the state set in the longitudinal distance-velocity space and discretize it into several subsets. For pedestrians, construct the state set in the longitudinal distance-lateral offset space and discretize it into several subsets.

[0027] As time goes by, the state subsets grow. The vehicle state subset grows according to the acceleration range, while the pedestrian state subset is pre-set to grow at a constant speed.

[0028] By combining the vehicle's perception information, the subset of states that have been exposed in the vehicle's field of vision is removed, and the remaining subset of states is obtained by updating in real time;

[0029] The future occupied area of ​​each state subset is estimated based on the set prediction method.

[0030] As a preferred technical solution, state subset growth includes state subset growth for phantom vehicle type and state subset growth for phantom pedestrian type;

[0031] For the state subset of the phantom vehicle type, its lateral position is set to 0m, and the state subset is within the acceleration range. Internal growth, among which and These are the preset minimum and maximum accelerations, respectively;

[0032] For the state subset of phantom pedestrians, set them to move at a preset constant speed along the zebra crossing direction;

[0033] The update formulas for the longitudinal position and velocity boundaries of the state subset of the phantom vehicle type, and the update formulas for the longitudinal position and lateral offset boundaries of the state subset of the phantom pedestrian type are as follows:

[0034] ;

[0035] ;

[0036] ;

[0037] in, The updated longitudinal position minimum boundary for the state subset of the phantom vehicle type; The updated longitudinal position maximum boundary for the state subset of the Phantom vehicle type; The minimum boundary of the current longitudinal position of the state subset of the Phantom vehicle type; The current minimum speed of a subset of the states of the Phantom vehicle type; This is the lower bound of acceleration; For time; The maximum boundary of the current longitudinal position of the state subset of the Phantom vehicle type; This is the speed limit value; The current maximum speed for a subset of the Phantom vehicle's state; This is the upper bound of acceleration; The updated minimum speed for a subset of the states of the Phantom vehicle type; The updated maximum speed for a subset of the Phantom vehicle type's state; The updated minimum boundary of the vertical position for the state subset of the phantom pedestrian type; The minimum boundary of the current vertical position of the state subset of the phantom pedestrian type; Preset constant speed; The updated vertical position maximum bound for the state subset of the phantom pedestrian type; The maximum boundary of the current vertical position of the state subset of the phantom pedestrian type; The minimum updated lateral offset for the subset of states of the Phantom Pedestrian type; The minimum current lateral offset of the state subset of the Phantom Pedestrian type; The maximum updated lateral offset for the subset of states of the Phantom Pedestrian type; The maximum current lateral offset for a subset of states of the Phantom Pedestrian type.

[0038] As a preferred technical solution, the specific process for calculating the visibility cost of each candidate path in S4 is as follows: each candidate path is discretized into multiple sampling points; the ratio of the remaining state subsets at each sampling point is calculated based on the number of remaining state subsets and the total number of initial state subsets at the current position of the vehicle; finally, the visibility cost of the entire candidate path is calculated based on the ratio of the remaining state subsets at each sampling point. r The specific formula is:

[0039] ;

[0040] ;

[0041] in, The visibility cost for the entire candidate path; The number of samples; s Here, s is the sampling point number. When s=1, it represents the sampling point closest to the current vehicle position. Discount factor; The ratio of the remaining state subsets at each sampling point; This represents the number of the remaining state subsets. This represents the total number of initial state subsets of the vehicle's current position.

[0042] As a preferred technical solution, the total cost in S4 is composed of a weighted average of center deviation cost, path switching cost, collision cost, and visibility cost.

[0043] As a preferred technical solution, when searching for discrete acceleration actions and generating velocity curves that satisfy the constraints in S5, a hybrid A* algorithm is used, the specific process of which includes:

[0044] A three-dimensional state vector is constructed, consisting of the length, speed, and time traveled along the reference path, to describe the motion state of the vehicle on the reference path;

[0045] Set the current node, expand the search nodes based on the vehicle kinematics model using the hybrid A* algorithm, generate child nodes at preset time intervals according to the sampled acceleration within a preset range, and retain child nodes that satisfy physical and collision constraints;

[0046] During the search process, the cost of each node is calculated based on a preset cost function. The node with the lowest cost is selected as the next node in each search. The search ends when the termination condition is met.

[0047] The termination condition is that the current node belongs to the preset target node set.

[0048] As a preferred technical solution, the specific rules for generating child nodes are as follows:

[0049] ;

[0050] ;

[0051] in, For the current node, where, , and These represent the length, speed, and time traveled along the reference path at the current node; , and These represent the length, speed, and time traveled along the reference path in the child node, respectively. This is a preset time interval; The acceleration of the child node; This represents the maximum speed of the vehicle. and Preset minimum and maximum acceleration; It is the projection of the obstacle onto the ST diagram.

[0052] As a preferred technical solution, the preset cost function consists of heuristic cost and cumulative cost, and its specific formula is as follows:

[0053] ;

[0054] ;

[0055] ;

[0056] in, As a pre-set cost; The cost of inspiration; For cumulative costs; Preset target node; For nodes n Length along the reference path; The parent node of the current node; The weighting coefficient for the speed deviation cost; For nodes n speed; For reference speed; The weighting coefficients for acceleration costs; For nodes n The acceleration; The weighting coefficients for the cost of acceleration; For nodes n acceleration.

[0057] According to another aspect of the present invention, a motion planning system considering visibility costs in urban multi-blind-spot scenarios is provided. The system includes a blind-spot perception module, a key blind-spot screening module, a phantom state management module, a path optimization module, and a speed planning module.

[0058] The blind spot perception module is used to extract all blind spots in the scene, and generate a set of blind spot boundaries by combining the extracted blind spots with the road topology in the scene;

[0059] The critical blind spot filtering module is used to filter out critical blind spots from the set of blind spot boundaries based on the vehicle's driving intention;

[0060] The phantom state management module is used to initialize, grow, update, and predict a subset of the states of all phantom traffic participants within critical blind spots;

[0061] The path optimization module is used to generate candidate paths. It calculates the visibility cost and total cost based on the state subset of phantom traffic participants and selects the path with the minimum total cost as the optimal path.

[0062] The velocity planning module is used to search for discrete acceleration actions using the optimal path as a reference path and generate velocity curves that meet the constraints.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] 1. In this invention, motion planning is completed by extracting blind spots, screening key blind spots, predicting the status of phantom traffic participants, calculating visibility costs, selecting the optimal path, and generating speed curves. The core is to maximize the vehicle's field of vision and reduce the impact of blind spots through active offset. This method covers multiple urban scenarios and improves vehicle traffic efficiency while ensuring safety and comfort. By maximizing the vehicle's field of vision and reducing the impact of blind spots, it can improve vehicle traffic efficiency in various typical urban scenarios while ensuring safety and comfort, thus simultaneously taking into account safety, traffic efficiency, and comfort.

[0065] 2. In this invention, the blind spots formed by dynamic and static obstructions are accurately defined by the geometric ray method, and the boundaries are clarified by the intersection operation of the road topology, providing an accurate spatial basis for subsequent risk assessment.

[0066] 3. In this invention, a vehicle driving corridor is constructed based on the vehicle's intent, and the forward driving corridors corresponding to each boundary segment in the blind spot boundary set are analyzed in sequence to screen key blind spots that affect future vehicle driving. The screening criteria include different boundary segment judgment rules corresponding to vehicles and pedestrians, and key blind spots are confirmed by the key blind spot boundaries. Furthermore, the blind spot boundary set is generated based on the scene road topology, making this method adaptable to various road topologies and suitable for urban multi-blind spot scenarios.

[0067] 4. In this invention, by initializing, growing, updating and predicting the state subsets of phantom traffic participants in the critical blind spot, and combining the vehicle's perception information to remove the exposed state subsets, the remaining state subsets are updated in real time. This reflects the inter-frame continuity estimation, and the visibility cost is calculated by the ratio of the number of remaining state subsets to the total number of initial state subsets. The cost of the entire path is obtained based on the ratio of each sampling point. Considering the inter-frame continuity estimation of the future state subsets of phantom traffic participants and calculating the visibility cost based on their quantity changes, this can alleviate the overestimation problem of blind spot risk in the traditional reachability set method.

[0068] 5. In this invention, the total cost is composed of the center deviation cost, path switching cost, collision cost and visibility cost weighted together. By integrating multi-dimensional cost factors, the rationality of the path, driving stability, safety and visibility are taken into account, avoiding the one-sidedness of planning caused by a single indicator, and making the optimal path more in line with actual driving needs.

[0069] 6. In this invention, a hybrid A* algorithm is used for speed planning. Based on a three-dimensional state vector and a vehicle kinematics model, acceleration actions are searched to ensure that the generated speed curve meets physical and collision constraints, achieving a balance between safety and efficiency. By generating sub-nodes through explicit rules, the range of key parameters such as speed and acceleration is limited, ensuring the feasibility and smoothness of the speed curve and improving driving comfort. Furthermore, a preset cost function combines heuristic and cumulative costs, taking into account both goal orientation and process optimization. Simultaneously, by adjusting indicators such as speed and acceleration through multiple weighted coefficients, speed planning becomes more flexible and adaptable to different driving scenarios. Attached Figure Description

[0070] Figure 1 This is a schematic diagram of the method steps of the present invention;

[0071] Figure 2 This is a schematic diagram of the motion planning algorithm architecture in the embodiment;

[0072] Figure 3a This is a diagram showing the results of filtering for critical blind spots in a no-signal intersection scenario in the embodiment.

[0073] Figure 3b This is a diagram showing the key blind spot filtering results in an unsignalized intersection scenario when considering the straight-ahead intention of vehicles in the embodiment.

[0074] Figure 3c This is a diagram showing the key blind spot filtering results in an unsignalized intersection scenario when considering the left-turn intention of a vehicle in the example.

[0075] Figure 3d This is a diagram showing the key blind spot filtering results in an unsignalized intersection scenario when considering the right-turn intention of a vehicle in the example.

[0076] Figure 4a This is a result image of the pedestrian ghost peek-out scenario in the embodiment, without considering the key blind spot screening.

[0077] Figure 4b This is a key blind spot filtering result diagram for a pedestrian "ghost peek" scenario, considering the straight-going intention of a vehicle in the example.

[0078] Figure 5a This is a result image of the two-lane overtaking scenario in the embodiment, without considering the key blind spot screening.

[0079] Figure 5bThis is a key blind spot filtering result diagram for a two-lane overtaking scenario when considering the overtaking intention of the vehicle in the embodiment.

[0080] Figure 6a This is a comparison chart of acceleration-time curves for a no-signal intersection scenario in the embodiment.

[0081] Figure 6b This is a comparison chart of speed-time curves for a no-signal intersection scenario in the examples;

[0082] Figure 6c For the example t Simulation results of motion planning for a no-signal intersection scenario at time =0.5s;

[0083] Figure 6d For the example t Simulation results of motion planning for a no-signal intersection scenario at time = 5.0s;

[0084] Figure 6e For the example t Simulation results of motion planning for a no-signal intersection scenario at 7.0s;

[0085] Figure 6f For the example t Simulation results of motion planning for a no-signal intersection scenario at time =12.0s;

[0086] Figure 7a This is a comparison chart of acceleration-time curves for a pedestrian ghost-protruding scene in the embodiment;

[0087] Figure 7b This is a comparison chart of speed-time curves for a pedestrian ghost-appearing scene in the example;

[0088] Figure 7c For the example t Simulation results of motion planning for a pedestrian ghost-protruding scene at 0.5s;

[0089] Figure 7d For the example t Simulation results of motion planning for a pedestrian ghost-protruding scene at 2.5s;

[0090] Figure 7e For the example t Simulation results of motion planning for a pedestrian ghost-protruding scene at 5.0s;

[0091] Figure 7f For the example t Simulation results of motion planning for a no-signal intersection scenario at 9.5s;

[0092] Figure 8aThis is a comparison chart of acceleration-time curves for a two-lane overtaking scenario in the example.

[0093] Figure 8b This is a comparison chart of speed-time curves for a two-lane overtaking scenario in the example;

[0094] Figure 8c For the example t Simulation results of motion planning for a two-lane overtaking scenario at 1.0s;

[0095] Figure 8d For the example t Simulation results of motion planning for a two-lane overtaking scenario at 4.5s;

[0096] Figure 8e For the example t Simulation results of motion planning for a two-lane overtaking scenario at 6.5s. Detailed Implementation

[0097] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0098] This solution implements a motion planning method for urban scenarios with multiple blind spots, aiming to improve vehicle traffic efficiency while ensuring safety and comfort. The method first extracts key blind spots based on road topology and uses ensemble estimation to determine the state subsets of phantom traffic participants. Then, a visibility cost is constructed based on its dynamic changes and incorporated into sampling-based path planning to guide the vehicle to lateral offsets, thereby improving its own field of vision. The optimal path is then fed into a hybrid A* algorithm for speed planning. This method is neither overly conservative nor overly aggressive, improving vehicle traffic efficiency while ensuring safety and comfort.

[0099] Example 1

[0100] In this embodiment, an active offset motion planning method for urban multi-blind-spot scenarios is adopted, and the method steps are as follows: Figure 1 As shown, it specifically includes:

[0101] S1. Extract all blind spots in the scene, and generate a set of blind spot boundaries by combining the extracted blind spots with the road topology in the scene;

[0102] S2. Based on the vehicle's driving intention, key blind spots are selected from the set of blind spot boundaries;

[0103] S3. Initialize, grow, update and predict the state subset of all phantom traffic participants in the critical blind zone;

[0104] S4. Generate candidate paths. Based on the state subset of phantom traffic participants, calculate the visibility cost of each candidate path, and combine the visibility cost with other costs to calculate the total cost of each candidate path; select the path with the minimum total cost as the optimal path.

[0105] S5. Using the optimal path as the reference path for velocity planning, search for discrete acceleration actions, generate velocity curves that satisfy the constraints, and the optimal path and velocity curves together constitute the motion planning of the vehicle.

[0106] The specific implementation of this method is as follows:

[0107] Figure 2 This is a schematic diagram of the motion planning algorithm architecture in this method. Figure 2 In the diagram, light blue areas represent non-passable lanes for the vehicle; light beige areas represent critical blind spots; light gray areas represent enclosed or obstructed areas of the vehicle's view; dark blue areas represent static obstacles; dark blue solid lines represent critical blind spot boundaries for vehicle types; dark blue dashed lines represent non-critical blind spot boundaries for vehicle types; light blue solid lines represent critical blind spot boundaries for pedestrian types; light blue dashed lines represent non-critical blind spot boundaries for pedestrian types; green vehicles represent other vehicles; blue vehicles represent the vehicle itself; yellow vehicles represent slow-moving trucks; blue grid areas represent the state set for vehicle types; green grid areas represent the state set for pedestrian types; blue areas represent future occupancy for vehicle types; green areas represent future occupancy for pedestrian types.

[0108] 1) Use the geometric ray method to extract all dynamic and static blind zones and generate a set of blind zone boundaries.

[0109] Rays are emitted from the center of the vehicle to the moving and static obstructions (considered as polygons) and connected to their vertices; the two rays with the largest included angle are selected and extended to the maximum sensing range of 100 m, and the ends of the rays are connected to form the corresponding blind zone area.

[0110] 2) Key blind spots are identified based on road topology and vehicle driving intentions.

[0111] 201) Perform an intersection operation between the blind spot area and the driving corridors of all phantom traffic participants to obtain the set of blind spot boundary segments. ;

[0112] 202) Construct the vehicle's driving corridor based on the vehicle's intention. And sequentially for each boundary segment Corresponding forward driving corridor We will conduct analysis to identify key blind spots that could affect future vehicle driving.

[0113] If the traffic participants in the phantom are vehicles and If the direction of movement is consistent with the vehicle's direction and does not pose a risk, the blind spot should be eliminated; if This would pose a risk of future conflicts with private vehicles, thus remaining a critical blind spot. If the Phantom's traffic participants are pedestrians, then the boundary segment... All are considered critical blind zone boundaries. Furthermore, within the same lane or of the same type of candidate boundary segment, only the blind zone boundary closest to the potential conflict area is retained.

[0114] The results of key blind spot screening in the unsignalized intersection scenario are shown in Figure 3, which are consistent with the driving habits of human drivers. Figure 3a This indicates that when the vehicle's intentions are not considered, there are 4 blind spot areas that need to be taken into account, numbered 1-4. Figure 3b This indicates that when considering the vehicle's intention to go straight, the blind spot area numbered 1 is considered the most critical. Figure 3c This indicates that when considering the vehicle's intention to turn left, the blind spot area numbered 1 is considered the most critical. Figure 3d This means that when considering the vehicle's intention to turn right, there are no critical blind spots in the scene, and they can all be ignored.

[0115] Figure 4 shows the results of key blind spot filtering for pedestrian ghost peek-out scenarios, which are consistent with the driving habits of human drivers; Figure 4a This indicates that when the vehicle's intention is not considered, there are 3 blind spot areas that need to be considered, numbered 1-3; Figure 4b This indicates that when considering the vehicle's intention to go straight, the blind spot area numbered 2 is considered the most critical.

[0116] The key blind spot screening results for the two-lane overtaking scenario are shown in Figure 5, which are consistent with the driving habits of human drivers. Figure 5a This indicates that when the vehicle's intention is not considered, there are 3 blind spot areas that need to be considered, numbered 1-3; Figure 5b This indicates that when considering the vehicle's intention to go straight, the blind spot area numbered 1 is considered the most critical.

[0117] 3) Initialize, grow, update, and predict the state subset of phantom traffic participants within the blind zone.

[0118] 301) Initialize the state set, where the vehicle's longitudinal distance minus its speed is... Construct a set of states in space and discretize it into several subsets ( ), pedestrians' longitudinal distance minus lateral offset Spatial construction and discretization ( );

[0119] 302) As time goes by, the state subset grows. The vehicle state subset grows according to the acceleration interval, and the pedestrian state subset grows at a constant speed.

[0120] For the state subset of the phantom vehicle type, assuming the lateral position... In the acceleration range Internal growth, among which and For minimum and maximum acceleration. Longitudinal position boundaries of the state subset. and velocity boundary The update formula is as follows:

[0121]

[0122]

[0123] For a subset of states of the phantom pedestrian type, assume it is based on speed If moving at a constant speed along the zebra crossing, then the longitudinal position boundary... and lateral offset boundary The update formula is as follows:

[0124]

[0125] 303) Combine the vehicle's perception information to remove the subset of states that have been exposed in the vehicle's field of vision;

[0126] 304) Estimate the future occupied area of ​​each state subset based on the set prediction method.

[0127] 4) Sample multiple candidate paths, calculate the visibility cost based on the proportion of the remaining state subset of phantom traffic participants, and select the path with the minimum cost by weighting it with the center deviation, path switching and collision costs.

[0128] 401) Generate along its parallel direction based on the global reference path. There are 10 candidate paths, with a horizontal spacing of 1 / 2000 between each path. This constitutes a uniformly distributed set of paths. ;

[0129] 402) Discrete path sampling points, for each candidate path Discretized sampling points ,in For the sampling point closest to the current vehicle position, the ratio of the remaining state subset at each sampling point is obtained according to the following formula: ,in This represents the total number of initial state subsets for the vehicle's current position. This represents the number of remaining state subsets. A discount factor is introduced. To represent uncertainty, the visibility cost for the entire path is calculated using the following formula: ,Pick .

[0130] 403) Calculate the total cost of each candidate path. This is the deviation from the center. Path switching ,collision and visibility cost Weighted combination:

[0131]

[0132] Output the optimal path that minimizes the total cost.

[0133] 5) Using the optimal path as the reference path for velocity planning, use hybrid A* search to search for discrete acceleration actions and generate velocity curves that satisfy the constraints.

[0134] 501) Construct the length of travel along the reference path ,speed With time The three-dimensional state vector , used to describe the motion state of the vehicle on the reference path;

[0135] 502) Let the current node be Hybrid A* extends the search node search based on the vehicle kinematics model, according to sampled acceleration. exist Generate the current node child nodes : Retain child nodes that satisfy the following physical and collision constraints: and It is the projection of the obstacle onto the ST diagram.

[0136] 503) The cost function for each node is: The cost of inspiration is The cumulative cost is Each search selects the node with the lowest cost as the next node, and the termination condition is that the current node belongs to the target node set. , .

[0137] like Figure 6c , 6d As shown in Figures 6e and 6f, in an unsignalized intersection scenario, the blue vehicle is traveling from south to north. Two gray buildings on the side of the road create blind spots. A phantom vehicle is traveling from east to west. The blue squares represent the blue vehicle, and the purple squares represent the other vehicle. The light gray area represents the obstruction area, the light yellow area represents the critical blind spot area, and the light green area represents the estimated future occupancy area of ​​the phantom traffic participant. The motion planning performance in this scenario is compared to... Figure 6a and 6b As shown in Table 1, the quantitative indicators are compared. The results demonstrate that the method of this invention can achieve higher driving speeds while maintaining stable acceleration. Compared with conservative planners and active offset planners, the proposed method improves traffic efficiency by 18.21% and 13.33%, respectively; safety by 21.74% and 18.18%, respectively; and comfort by 4.17% and 4.17%, respectively.

[0138] Table 1 Comparison of Planning Indicators for Unsignalized Intersection Scenarios

[0139]

[0140] like Figure 7c , 7d As shown in Figures 7e and 7f, in the "phantom pedestrian peeking out" scenario, a blue car is traveling from west to east, three green cars are parked on the side of the road, and a phantom pedestrian is traveling from south to north; the light yellow area represents the critical blind spot, and the light green area represents the estimated future occupancy area of ​​the phantom pedestrian. The motion planning effect of this scenario is compared to... Figure 7a and 7b As shown in Table 2, the quantitative indicators are compared. The results demonstrate that the method of this invention can achieve higher driving speeds while maintaining stable acceleration. Compared with conservative planners and active offset planners, the proposed method improves efficiency by 30.65% and 24.72%, respectively; safety by 23.81% and 20.00%, respectively; and comfort by 15.15% and 14.29%, respectively.

[0141] Table 2 Comparison of Planning Indicators for Pedestrian Ghost Pedestrian Sightseeing Scenarios

[0142]

[0143] like Figure 8c , 8d As shown in Figure 8e, in a two-lane overtaking scenario, the blue car is traveling from west to east. A yellow truck in front causes dynamic occlusion, another vehicle is represented by purple, and a building behind causes static occlusion. A phantom car is traveling from east to west in the oncoming lane. The light yellow area represents the critical blind spot, and the light green area represents the estimated future area occupied by the phantom car. The motion planning effect in this scenario is compared to... Figure 8a and 8bAs shown in Table 3, the quantitative indicators are compared. The conservative planner, lacking visibility cost, does not actively explore the environment and conservatively follows the slow-moving truck. The active offset planner and the planner of this invention exhibit the same speed change trend, but their blind spots are not filtered, leading to conflicting visibility cost changes in the left and right blind spots. The vehicle initially does not offset but gradually offsets later to gain more visibility. Compared to the conservative planner, this invention is superior in decision-making, with only a slight decrease in comfort. Furthermore, using the same overtaking strategy as the active offset planner, efficiency is improved by 3.79%, safety by 33.33%, and comfort by 44.44%.

[0144] Table 3 Comparison of Planning Indicators for Two-Lane Overtaking Scenarios

[0145]

[0146] In summary, this method covers multiple urban scenarios, improving vehicle traffic efficiency while ensuring safety and comfort. By maximizing vehicle visibility and reducing blind spot impact, it can improve vehicle traffic efficiency in various typical urban scenarios while ensuring safety and comfort, thus simultaneously balancing safety, traffic efficiency, and comfort.

[0147] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An active offset motion planning method for urban multi-blind-spot scenarios, characterized in that, The method steps include: S1. Extract all blind spots in the scene, and generate a set of blind spot boundaries by combining the extracted blind spots with the road topology in the scene; S2. Based on the vehicle's driving intention, key blind spots that affect the vehicle's future driving are selected from the blind spot boundary set; S3. Initialize, grow, update and predict the state subset of all phantom traffic participants in the critical blind zone; S4. Generate candidate paths. Based on the state subset of phantom traffic participants, calculate the visibility cost of each candidate path, and combine the visibility cost with other costs to calculate the total cost of each candidate path; select the path with the minimum total cost as the optimal path. S5. Using the optimal path as the reference path for velocity planning, search for discrete acceleration actions, generate velocity curves that satisfy the constraints, and the optimal path and velocity curves together constitute the motion planning of the vehicle. The specific process for calculating the visibility cost of each candidate path in S4 is as follows: each candidate path is discretized into multiple sampling points; the ratio of the remaining state subsets at each sampling point is calculated based on the number of remaining state subsets and the total number of initial state subsets at the current position of the vehicle; finally, the visibility cost of the entire candidate path is calculated based on the ratio of the remaining state subsets at each sampling point. r The specific formula is: ； ； in, The visibility cost for the entire candidate path; The number of samples; s Here, s is the sampling point number. When s=1, it represents the sampling point closest to the current vehicle position. Discount factor; The ratio of the remaining state subsets at each sampling point; This represents the number of the remaining state subsets. This represents the total number of initial state subsets of the vehicle's current position.

2. The active offset motion planning method for urban multi-blind-spot scenarios according to claim 1, characterized in that, The extraction of all blind spots in the scene in S1 is specifically done using the geometric ray method. First, dynamic and static occlusions are treated as polygons. Then, with the center of the vehicle as the origin, rays are emitted towards the dynamic and static occlusions respectively and connected to their vertices. The two rays with the largest included angle are selected and extended to the maximum sensing range of 100m, and the ends of the rays are connected to form the corresponding blind spot areas. The specific process of generating the blind spot boundary set by combining the road topology in the scene is as follows: the intersection operation of the blind spot area with the driving corridors of all phantom traffic participants is performed to obtain the blind spot boundary set.

3. The active offset motion planning method for urban multi-blind-spot scenarios according to claim 1, characterized in that, The specific process of S2 is as follows: combining the vehicle's intention, constructing the vehicle's driving corridor, and sequentially analyzing the forward driving corridors corresponding to each boundary segment in the blind spot boundary set, and screening out the key blind spots that affect the future driving of the vehicle. The specific criteria for screening critical blind spots that affect future vehicle travel include: when the phantom traffic participant is a vehicle, if the forward travel corridor corresponding to the boundary segment is completely contained within the vehicle's travel corridor, then the boundary segment is considered to be in the same direction as the vehicle's movement and does not pose a risk, and the blind spot boundary is removed; if the forward travel corridor corresponding to the boundary segment intersects with the vehicle's travel corridor, then the phantom traffic participant in the blind spot is considered to have a risk of future conflict with the vehicle, and the blind spot boundary is retained; when the phantom traffic participant is a pedestrian, the boundary segment is considered a critical blind spot boundary; the critical blind spot is determined by the critical blind spot boundary.

4. The active offset motion planning method for urban multi-blind-spot scenarios according to claim 1, characterized in that, The specific process of S3 includes: Initialize the state set. For vehicles, construct the state set in the longitudinal distance-velocity space and discretize it into several subsets. For pedestrians, construct the state set in the longitudinal distance-lateral offset space and discretize it into several subsets. As time goes by, the state subsets grow. The vehicle state subset grows according to the acceleration range, while the pedestrian state subset is pre-set to grow at a constant speed. By combining the vehicle's perception information, the subset of states that have been exposed in the vehicle's field of vision is removed, and the remaining subset of states is obtained by updating in real time; The future occupied area of ​​each state subset is estimated based on the set prediction method.

5. The active offset motion planning method for urban multi-blind-spot scenarios according to claim 4, characterized in that, The state subset growth includes state subset growth for phantom vehicle types and state subset growth for phantom pedestrian types; For the state subset of the phantom vehicle type, its lateral position is set to 0m, and the state subset is within the acceleration range. Internal growth, among which and These are the preset minimum and maximum accelerations, respectively; For the state subset of phantom pedestrians, set them to move at a preset constant speed along the zebra crossing direction; The update formulas for the longitudinal position and velocity boundaries of the state subset of the phantom vehicle type, and the update formulas for the longitudinal position and lateral offset boundaries of the state subset of the phantom pedestrian type are as follows: ； ； ； in, The updated longitudinal position minimum boundary for the state subset of the phantom vehicle type; The updated longitudinal position maximum boundary for the state subset of the Phantom vehicle type; The minimum boundary of the current longitudinal position of the state subset of the Phantom vehicle type; The current minimum speed of a subset of the states of the Phantom vehicle type; This is the lower bound of acceleration; For time; The maximum boundary of the current longitudinal position of the state subset of the Phantom vehicle type; This is the speed limit value; The current maximum speed for a subset of the Phantom vehicle's state; This is the upper bound of acceleration; The updated minimum speed for a subset of the states of the Phantom vehicle type; The updated maximum speed for a subset of the Phantom vehicle type's state; The updated minimum boundary of the vertical position for the state subset of the phantom pedestrian type; The minimum boundary of the current vertical position of the state subset of the phantom pedestrian type; Preset constant speed; The updated vertical position maximum bound for the state subset of the phantom pedestrian type; The maximum boundary of the current vertical position of the state subset of the phantom pedestrian type; The minimum updated lateral offset for the subset of states of the Phantom Pedestrian type; The minimum current lateral offset of the state subset of the Phantom Pedestrian type; The maximum updated lateral offset for the subset of states of the Phantom Pedestrian type; The maximum current lateral offset for a subset of states of the Phantom Pedestrian type.

6. The active offset motion planning method for urban multi-blind-spot scenarios according to claim 1, characterized in that, The total cost in S4 is composed of a weighted average of center deviation cost, path switching cost, collision cost, and visibility cost.

7. The active offset motion planning method for urban multi-blind-spot scenarios according to claim 1, characterized in that, In S5, when searching for discrete acceleration actions and generating velocity curves that satisfy constraints, a hybrid A* algorithm is used, the specific process of which includes: A three-dimensional state vector is constructed, consisting of the length, speed, and time traveled along the reference path, to describe the motion state of the vehicle on the reference path; Set the current node, expand the search nodes based on the vehicle kinematics model using the hybrid A* algorithm, generate child nodes at preset time intervals according to the sampled acceleration within a preset range, and retain child nodes that satisfy physical and collision constraints; During the search process, the cost of each node is calculated based on a preset cost function. The node with the lowest cost is selected as the next node in each search. The search ends when the termination condition is met. The termination condition is that the current node belongs to the preset target node set.

8. The active offset motion planning method for urban multi-blind-spot scenarios according to claim 7, characterized in that, The specific rules for generating child nodes are as follows: ； ； in, For the current node, where, , and These represent the length, speed, and time traveled along the reference path at the current node; , and These represent the length, speed, and time traveled along the reference path in the child node, respectively. This is a preset time interval; The acceleration of the child node; This represents the maximum speed of the vehicle. and Preset minimum and maximum acceleration; It is the projection of the obstacle onto the ST diagram.

9. The active offset motion planning method for urban multi-blind-spot scenarios according to claim 7, characterized in that, The preset cost function consists of heuristic cost and cumulative cost, and its specific formula is as follows: ； ； ； in, As a pre-set cost; The cost of inspiration; For cumulative costs; Preset target node; For nodes n Length along the reference path; The parent node of the current node; The weighting coefficient for the speed deviation cost; For nodes n speed; For reference speed; The weighting coefficients for acceleration costs; For nodes n The acceleration; The weighting coefficients for the cost of acceleration; For nodes n acceleration.

Images