Lateral avoidance methods and devices for autonomous vehicles

By acquiring lane and obstacle information, dynamically calculating the avoidance range, and adopting segmented S-curve control, the problem of improper avoidance by autonomous vehicles under different lane widths is solved, achieving precise and safe lateral avoidance, and improving the avoidance reliability and ride comfort of autonomous vehicles.

CN123104636APending Publication Date: 2026-11-17BEIJING TRUNK TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611521628.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-29
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

When encountering stationary/low-speed obstacles, existing autonomous vehicles mostly use fixed offsets for lateral avoidance technology. This can lead to vehicles crossing the lane line or going beyond the lane boundary in narrow lanes, and insufficient avoidance in wide lanes, resulting in poor reliability of lateral avoidance.

Method used

By acquiring lane information and obstacle positions, the system dynamically calculates the target avoidance range, generates a lateral avoidance path within the vehicle's lane, and uses a segmented S-curve to control the change in offset. Combined with perception frame period and safety detection, it ensures that the avoidance range adapts to different road widths and obstacle positions.

Benefits of technology

It achieves precision and safety in obstacle avoidance, avoids vehicle lane departure problems caused by fixed offset, and improves the obstacle avoidance reliability and ride comfort of autonomous vehicles under different road conditions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123104636A_ABST
    Figure CN123104636A_ABST
Patent Text Reader

Abstract

This disclosure relates to a method and apparatus for lateral obstacle avoidance by an autonomous vehicle. The method includes: acquiring lane information of the lane in which the vehicle is located and spatial location information of a target obstacle to avoid, wherein the lane information includes lane width; calculating the degree of lateral intrusion of the target obstacle relative to the lane in which the vehicle is located based on the spatial location information of the target obstacle to avoid; determining a target avoidance range based on the lane width or the degree of lateral intrusion; generating a lateral avoidance path within the lane in which the vehicle is located based on the target avoidance range; and controlling the vehicle to perform lateral avoidance according to the lateral avoidance path.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This disclosure relates to the field of autonomous driving, and in particular to methods and devices for lateral obstacle avoidance by autonomous vehicles. Background Technology

[0002] Currently, autonomous vehicles frequently encounter stationary / low-speed obstacles that have crossed into their own lane or adjacent lanes, such as disabled vehicles, road construction areas, or slow-moving vehicles. In such scenarios, autonomous vehicles need to perform lane-to-lane nipping maneuvers, which involves making a slight lateral shift relative to the lane centerline without leaving the current lane, thereby avoiding obstacles and ensuring driving safety.

[0003] Existing lateral avoidance technologies mostly employ fixed offsets or simple proportional controls. For example, when an obstacle is detected, a preset lateral offset value (such as 0.3 meters) is directly output, causing the vehicle to deviate from the center line. However, when the lane is narrow, a fixed offset may cause the vehicle to cross the line or exceed the lane boundary, while when the lane is wide, a fixed offset may result in insufficient avoidance, thus making the reliability of lateral avoidance poor. Summary of the Invention

[0004] To overcome the problems existing in related technologies, this disclosure provides a method and apparatus for lateral obstacle avoidance by autonomous vehicles. The technical solution is as follows: According to a first aspect of the present disclosure, a method for lateral obstacle avoidance by an autonomous vehicle is provided, the method comprising: The system acquires lane information of the vehicle's lane and spatial location information of the obstacle the target is avoiding. The lane information includes: lane width. Based on the spatial location information of the target obstacle, the degree of lateral intrusion of the target obstacle relative to the lane where the vehicle is located is calculated; The target avoidance range is determined based on the lane width or the degree of lateral intrusion. Based on the target avoidance range, a lateral avoidance path is generated within the lane where the vehicle is located; The vehicle is controlled to perform lateral avoidance according to the lateral avoidance path.

[0005] In one embodiment, the method further includes: Candidate determination is performed on each obstacle detected according to the sensing frame period: For each obstacle, calculate the minimum value of the longitudinal coordinate of the obstacle's various boundary points in the Frenet coordinate system, and calculate the maximum value of the vehicle's longitudinal coordinate in the Frenet coordinate system; the Frenet coordinate system is based on the reference line of the lane where the vehicle is located; the longitudinal distance between the obstacle and the vehicle is obtained based on the minimum value of the obstacle's longitudinal coordinate and the maximum value of the vehicle's longitudinal coordinate. Obstacles that simultaneously meet the following conditions are identified as avoidable candidate obstacles: The obstacle's speed is less than the vehicle's speed; the longitudinal distance between the obstacle and the vehicle is within a preset effective range; and the estimated collision time of the obstacle is less than the collision time threshold. For each dodgeable candidate obstacle, in each perception frame, if the obstacle is determined to be a dodgeable candidate obstacle, the corresponding frame count is increased; if the obstacle is not determined to be a dodgeable candidate obstacle, the corresponding frame count is decreased; dodgeable candidate obstacles with a frame count greater than a preset counting threshold are identified as target dodgeable obstacles.

[0006] In one embodiment, the method further includes: Calculate the maximum and minimum values ​​of the lateral coordinates of each edge point of the target obstacle avoidance in the Frenet coordinate system; If the maximum value of the lateral coordinate of the target obstacle falls within the first preset range, it is determined to be an obstacle that can be avoided on the left. If the minimum value of the lateral coordinate of the obstacle to be avoided by the target falls within the second preset range, it is determined to be an obstacle that can be avoided on the right.

[0007] In one embodiment, the lower limit of the first preset interval is: the position after the left lane boundary of the lane where the vehicle is located is offset away from the reference line of the vehicle by a preset lateral threshold. The upper limit of the first preset interval is: the position after the left lane boundary is offset towards the reference line of the vehicle by a preset avoidable adjustment width; The lower limit of the second preset interval is: the position after the right lane boundary of the lane where the vehicle is located is offset towards the reference line of the vehicle by a preset avoidable adjustment width; The upper limit of the second preset interval is: the position after the right lane boundary is offset away from the vehicle reference line by a preset lateral threshold; Specifically, with the direction the vehicle's front is facing forward, when facing forward, the left lane boundary is the lane boundary located on the left side of the vehicle, and the right lane boundary is the lane boundary located on the right side of the vehicle; the vehicle reference line is the center line of the lane in which the vehicle is located.

[0008] In one embodiment, determining the target avoidance range based on the lane width or the degree of lateral intrusion includes: The intrusion level is determined based on the degree of lateral intrusion of the target obstacle, and the lateral offset amplitude of the target is determined based on the intrusion level. Different intrusion levels correspond to different lateral offset amplitudes. or, The target lateral offset amplitude is determined based on the lane width, with different lane widths corresponding to different lateral offset amplitudes; The total avoidance time is obtained based on the target lateral offset amplitude, whereby the target avoidance amplitude includes both the target lateral offset amplitude and the total avoidance time.

[0009] In one embodiment, generating a lateral avoidance path within the lane where the vehicle is located based on the target avoidance magnitude includes: When the system detects that it has entered the obstacle avoidance operation state, the obstacle avoidance operation process is divided into an acceleration phase and a deceleration phase based on the total obstacle avoidance time. In each perception frame, based on the duration of the current avoidance, it is determined whether the current perception frame is in the avoidance acceleration phase or the avoidance deceleration phase. If the current sensing frame is in the avoidance acceleration phase, the lateral offset of the current sensing frame is calculated based on the first constant acceleration value and the duration of avoidance. The lateral offset of the current sensing frame is positively correlated with the square of the duration of avoidance. If the current sensing frame is in the avoidance deceleration phase, the lateral offset of the current sensing frame is calculated based on the second constant acceleration value and the remaining avoidance time. The lateral offset of the current sensing frame is negatively correlated with the square of the remaining avoidance time. The second constant acceleration value is equal in magnitude and opposite in direction to the first constant acceleration value. When the duration of the current avoidance is the same as the total avoidance time, it is determined that the avoidance deceleration phase has ended and the lateral offset has reached the target lateral offset amplitude. After calculating the lateral offset for each frame, a safety check is performed. If the safety check confirms that there is no obstacle to avoid, the currently calculated lateral offset is used. If there is an obstacle to avoid, the lateral offset obtained from the previous sensing frame is used. The lateral avoidance path is obtained based on the lateral offset used in each sensing frame.

[0010] In one embodiment, the security detection includes: Obtain the lateral coordinate range of each obstacle in the Frenet coordinate system; If the lateral coordinate range of the obstacle overlaps with the lane area in the target offset direction, and the longitudinal distance between the obstacle and the vehicle is less than the safe distance determined based on relative speed, then the current avoidance is deemed unsafe and there is an obstacle to avoidance.

[0011] In one embodiment, the method further includes: When the avoidance exit condition is met, the actual lateral offset used in the current perception frame is obtained as the return starting offset; The total return time is determined based on the return start offset, and the return process is divided into a return acceleration segment and a return deceleration segment; In each sensing frame, the current sensing frame is determined to be in the return acceleration phase or the return deceleration phase based on the duration of the return. If the current sensing frame is in the return acceleration segment, the lateral offset is gradually reduced from the return start offset according to the third constant acceleration value; If the current sensing frame is in the return deceleration phase, the lateral offset continues to decrease gradually according to the fourth constant acceleration value, which is equal in magnitude and opposite in direction to the third constant acceleration value. When the duration of the return is the same as the total return duration, or when the lateral offset decreases to less than a preset threshold, the return process is determined to be over and the avoidance operation state is exited.

[0012] In one embodiment, before controlling the vehicle to perform lateral avoidance based on the lateral avoidance path, the method further includes: Detect whether there are avoidable obstacles on both the left and right sides of the vehicle at the same time; When both avoidable obstacles on the left and right are detected simultaneously, the vehicle is prohibited from performing lateral avoidance according to the lateral avoidance path.

[0013] According to a second aspect of the present disclosure, an autonomous vehicle lateral avoidance device is provided, the device comprising: The planning module is used to obtain lane information of the lane in which the vehicle is located and spatial location information of the target obstacle to avoid. The lane information includes: lane width; The planning module is also used to calculate the degree of lateral intrusion of the target obstacle relative to the lane where the vehicle is located, based on the spatial location information of the target obstacle. The planning module is also used to determine the target avoidance range based on the lane width or the degree of lateral intrusion; The planning module is also used to generate a lateral avoidance path within the lane where the vehicle is located based on the target avoidance range; The control module is used to control the vehicle to perform lateral avoidance according to the lateral avoidance path.

[0014] In this disclosure, the target avoidance range is dynamically calculated based on the lane width and the degree of lateral intrusion of the obstacle, so that the avoidance range can adapt to different road widths and the actual position of the obstacle, making the avoidance range more accurate. This ensures avoidance safety and avoids excessive deviation of the vehicle from the lane due to excessive fixed offset.

[0015] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0017] Figure 1 This is a schematic diagram illustrating a lateral avoidance method for an autonomous vehicle according to an exemplary embodiment.

[0018] Figure 2 This is a schematic diagram of an obstacle avoidance scenario illustrated according to an exemplary embodiment. Figure 1 .

[0019] Figure 3 This is a schematic diagram of an obstacle avoidance scenario illustrated according to an exemplary embodiment. Figure 2 .

[0020] Figure 4 This is a schematic diagram of an obstacle avoidance scenario illustrated according to an exemplary embodiment. Figure 3 .

[0021] Figure 5 This is a schematic diagram of an obstacle avoidance scenario illustrated according to an exemplary embodiment. Figure 4 .

[0022] Figure 6 This is a schematic diagram of the structure of a lateral avoidance device for an autonomous vehicle according to an exemplary embodiment. Detailed Implementation

[0023] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0024] Figure 1 This is a schematic diagram illustrating a lateral avoidance method for an autonomous vehicle according to an exemplary embodiment, such as... Figure 1 As shown, the method includes the following steps: S101. Obtain lane information of the lane where the vehicle is located and spatial location information of the target obstacle to avoid. The lane information includes: lane width.

[0025] The method disclosed herein can be used as follows: Figure 2 The scenario shown is a slow-moving obstacle avoidance vehicle and Figure 3 The diagram shows a stationary obstacle avoidance scenario, illustrating the planning module within an autonomous vehicle.

[0026] The planning module in an autonomous vehicle obtains lane information and spatial location information of surrounding obstacles from the perception module. Lane information includes at least the lane width (lane_width), which serves as a reference for subsequent calculations of target avoidance maneuvers. The spatial location information of obstacles can include their position coordinates, velocity, acceleration, type (e.g., car, truck, stationary object), and contour information (xy_contour).

[0027] The perception module functions similarly to the perception modules in related technologies, providing information such as the vehicle's position, speed, acceleration, lane boundary lines, and the position, speed, type (car, truck, stationary object, etc.) and outline of surrounding obstacles.

[0028] In one embodiment, to avoid frequent switching of avoidance targets due to perceived noise, candidate judgment and inter-frame stability screening are performed on each perceived obstacle before determining the obstacle to be avoided. Specifically, the above method further includes performing candidate judgment on each perceived obstacle according to the perception frame period. The judgment process includes the following sub-steps A1-A2: A1. For each obstacle, calculate the minimum value of the longitudinal coordinate of each edge point of the obstacle in the Frenet coordinate system, and calculate the maximum value of the longitudinal coordinate of the vehicle in the Frenet coordinate system; the Frenet coordinate system is based on the reference line of the lane where the vehicle is located; obtain the longitudinal distance between the obstacle and the vehicle based on the minimum value of the longitudinal coordinate of the obstacle and the maximum value of the longitudinal coordinate of the vehicle.

[0029] The position information of the vehicle and surrounding obstacles is projected uniformly onto the Frenet coordinate system. The Frenet coordinate system uses the reference line of the vehicle's lane as a reference, where the reference line is the centerline of the lane. In the Frenet coordinate system, the s-coordinate represents the longitudinal distance along the reference line, and the l-coordinate represents the lateral offset relative to the reference line.

[0030] (1) Obtain the longitudinal occupied interval of the vehicle [s_start, s_end].

[0031] By combining the projected coordinates 's' of the vehicle on the reference line with the distances from the front to the front overhang (base2front_) and from the rear to the rear overhang (head_base2tail_), the longitudinal interval [s_start, s_end] occupied by the vehicle on the reference line is obtained. Specifically: The lower limit of the longitudinal section occupied by the vehicle is s_start = s_ego - head_base2tail_, where s_ego is the projection s coordinate of the vehicle's reference point (such as the rear axle center) on the reference line, and head_base2tail_ is the distance from the rear of the vehicle to the rear overhang. The upper limit of the longitudinal section occupied by the vehicle is s_end = s_ego + base2front_, where base2front_ is the distance from the front of the vehicle to the front overhang.

[0032] This gives us the longitudinal occupancy range [s_start, s_end] of the vehicle on the reference line, which represents the longitudinal space range occupied by the vehicle along the road direction.

[0033] (2) Obtain the longitudinal and transverse coordinate ranges of the obstacle.

[0034] For each obstacle detected, the bounding points on its outline are projected onto the same reference line to obtain the vertical coordinates (s-coordinates) and horizontal coordinates (l-coordinates) of each bounding point in the Frenet coordinate system.

[0035] The minimum value of the s coordinates of each boundary point of the obstacle is taken as the lower limit obs_s_min of the longitudinal interval of the obstacle, and the maximum value of the s coordinates of each boundary point of the obstacle is taken as the upper limit obs_s_max of the longitudinal interval of the obstacle, thus obtaining the longitudinal interval [obs_s_min, obs_s_max] of the obstacle on the reference line.

[0036] The minimum value of the l coordinates of each edge point of the obstacle is taken as the minimum value min_l of the horizontal coordinate interval of the obstacle, and the maximum value of the l coordinates of each edge point of the obstacle is taken as the maximum value max_l of the horizontal coordinate interval of the obstacle, thus obtaining the horizontal coordinate interval [min_l, max_l] of the obstacle on the reference line.

[0037] A2. Obstacles that simultaneously meet the following conditions are identified as avoidable candidate obstacles: the speed of the obstacle is less than the speed of the vehicle; the longitudinal distance between the obstacle and the vehicle is within a preset effective range; and the expected collision time of the obstacle is less than the collision time threshold.

[0038] The longitudinal distance is the difference between the lower limit of the longitudinal interval of the obstacle (obs_s_min) and the upper limit of the longitudinal interval occupied by the vehicle (s_end), that is: longitudinal distance = obs_s_min - s_end; this difference represents the longitudinal distance between the rear end of the obstacle (the end closest to the vehicle) and the front end of the vehicle (the end closest to the obstacle).

[0039] In this system, the obstacle's speed is less than the vehicle's speed, ensuring the obstacle is stationary or moving at low speed relative to the vehicle, requiring the vehicle to actively avoid it. The obstacle's longitudinal distance is within a preset effective range, determined based on the vehicle's length and a reserved safety distance, used to filter out invalid obstacles that are too far or too close. The obstacle's estimated time to collision (TTC) is less than a collision time threshold, which is obtained by looking up a table based on the obstacle's speed. For static obstacles, a smaller collision time threshold is used, causing the vehicle to trigger avoidance when the relative distance is relatively close; for dynamic obstacles, a relatively larger collision time threshold is used, causing the vehicle to trigger avoidance when the relative distance is relatively far. This collision time threshold mechanism enables differentiated processing of static and dynamic obstacles.

[0040] A3. For each dodgeable candidate obstacle, in each perception frame, if the obstacle is determined to be a dodgeable candidate obstacle, the corresponding cumulative frame count (index_map_) is increased (e.g., the cumulative frame count of the previous frame is multiplied by an entry factor greater than 1); if the obstacle is not determined to be a dodgeable candidate obstacle, the corresponding cumulative frame count is decreased (e.g., the cumulative frame count of the previous frame is multiplied by an exit factor less than 1); when the count value corresponding to the obstacle is greater than the preset count threshold (obstacle_valid_count), the dodgeable candidate obstacle with a cumulative frame count greater than the preset count threshold is determined as the target dodgeable obstacle (tar_obs_).

[0041] In related technologies, the same avoidance strategy is used for both static and dynamic obstacles. However, this approach cannot provide more aggressive avoidance behavior in emergency situations (such as rapidly approaching a stationary vehicle). In this disclosure, a smaller collision time threshold is used for static obstacles, causing the vehicle to trigger avoidance when the relative distance is relatively close; for dynamic obstacles, a relatively larger collision time threshold is used, causing the vehicle to trigger avoidance when the relative distance is relatively far. This collision time threshold mechanism achieves differentiated processing for static and dynamic obstacles.

[0042] This mechanism ensures that an obstacle is selected as the final target obstacle to avoid only when it has been identified as an avoidable candidate in multiple consecutive frames, thereby improving the stability and reliability of target selection.

[0043] In one embodiment, to further determine whether the target obstacle is located on the left or right side of the lane, the method further includes the following sub-steps B1-B3: B1. Calculate the maximum (max_l) and minimum (min_l) values ​​of the lateral coordinates of each edge point of the target obstacle avoidance in the Frenet coordinate system.

[0044] B2. If the maximum value (max_l) of the lateral coordinate of the target obstacle falls within the first preset range, then the target obstacle is determined to be an obstacle that can be avoided on the left.

[0045] The lower limit of the first preset interval is: the position after the left lane boundary of the lane where the vehicle is located is offset away from the vehicle reference line by a preset lateral threshold; the upper limit of the first preset interval is: the position after the left lane boundary is offset towards the vehicle reference line by a preset avoidable adjustment width; wherein, with the vehicle's front facing forward as the forward direction, the left lane boundary is the lane boundary located on the left side of the vehicle, and the right lane boundary is the lane boundary located on the right side of the vehicle; the vehicle reference line is the center line of the lane where the vehicle is located.

[0046] In one embodiment, the lateral position of the obstacle to be avoided is determined using the Frenet coordinate system. The Frenet coordinate system uses the centerline of the vehicle's lane as a reference line, where the s-coordinate represents the longitudinal position along the reference line, and the l-coordinate represents the lateral position relative to the reference line. In this embodiment, the l-coordinate is defined as negative on the left side and positive on the right side of the reference line when viewed from the direction of increasing s-coordinate. Therefore, the l-coordinate of the left lane boundary (i.e., the lane boundary on the left side of the vehicle when facing forward) is -lane_width / 2, and the l-coordinate of the right lane boundary (i.e., the lane boundary on the right side of the vehicle when facing forward) is lane_width / 2, where lane_width is the width of the vehicle's lane.

[0047] Based on the above definition, a first preset interval is set for the determination of an obstacle that can be avoided on the left side. The lower limit of the first preset interval is the position obtained by offsetting the left lane boundary of the lane where the host vehicle is located by a preset lateral threshold check_outer in the direction away from the host vehicle reference line, that is, -lane_width / 2-check_outer; the upper limit of the first preset interval is the position obtained by offsetting the left lane boundary by a preset avoidable adjustment width level_3_inner in the direction close to the host vehicle reference line, that is, -lane_width / 2+level_3_inner. If the maximum value max_l of the lateral coordinates of each frame point of the target obstacle to be avoided relative to the reference line is greater than the lower limit of the first preset interval and less than the upper limit of the first preset interval, that is, -lane_width / 2-check_outer<max_l<-lane_width / 2+level_3_inner is satisfied, the target obstacle to be avoided is determined as an obstacle that can be avoided on the left side. The lateral threshold check_outer is set according to different vehicle models, and is used to reserve a lateral safety distance; the avoidable adjustment width level_3_inner is a fixed value, and is used to define the tolerable lateral range after an obstacle intrudes into the current lane of the host vehicle.

[0048] B3. If the minimum value (min_l) of the lateral coordinates of the target obstacle to be avoided falls into a second preset interval, the target obstacle to be avoided is determined as an obstacle that can be avoided on the right side.

[0049] The lower limit of the second preset interval is the position obtained by offsetting the right lane boundary of the lane where the host vehicle is located by a preset avoidable adjustment width in the direction close to the host vehicle reference line; the upper limit of the second preset interval is the position obtained by offsetting the right lane boundary by a preset lateral threshold in the direction away from the host vehicle reference line; wherein, taking the heading direction of the host vehicle as the front direction, when facing the front direction, the left lane boundary is the lane boundary located on the left hand side of the host vehicle, the right lane boundary is the lane boundary located on the right hand side of the host vehicle; and the host vehicle reference line is the center line of the lane where the host vehicle is located.

[0050] For the determination of an obstacle that can be avoided on the right side, a second preset interval is set. The lower limit of the second preset interval is: the position after the right lane boundary of the lane where the host vehicle is located is offset by the preset avoidable adjustment width level_3_inner in the direction close to the host vehicle reference line, that is, lane_width / 2-level_3_inner; the upper limit of the second preset interval is: the position after the right lane boundary is offset by the preset lateral threshold check_outer in the direction away from the host vehicle reference line, that is, lane_width / 2+check_outer. If the minimum value min_l of the lateral coordinates of each bounding box point of the target obstacle to be avoided relative to the reference line is greater than the lower limit of the second preset interval and less than the upper limit of the second preset interval, that is, satisfying lane_width / 2-level_3_inner<min_l<lane_width / 2+check_outer, the target obstacle to be avoided is determined as an obstacle that can be avoided on the right side.

[0051] Through the above steps B1-B3, as shown Figure 4 , when too many obstacle vehicles intrude, no avoidance is performed.

[0052] S102. Calculate the lateral intrusion degree of the target obstacle to be avoided relative to the lane where the host vehicle is located according to the spatial position information of the target obstacle to be avoided.

[0053] In this embodiment, the lateral intrusion degree of the target obstacle relative to the boundary of the lane where the host vehicle is located is calculated according to the spatial position information of the target obstacle to be avoided acquired in step S101. The lateral intrusion degree is used to characterize the lateral distance or depth by which the obstacle intrudes from the outer side of the lane to the inner side of the lane.

[0054] Specifically, for a left-side obstacle, its lateral intrusion degree may be determined according to the difference between the maximum lateral coordinate (max_l) of the obstacle and the left lane boundary (-lane_width / 2); for a right-side obstacle, its lateral intrusion degree may be determined according to the difference between the right lane boundary (lane_width / 2) and the minimum lateral coordinate (min_l) of the obstacle. A larger lateral intrusion degree indicates that more part of the obstacle has intruded into the host lane, and the required avoidance amplitude is also larger.

[0055] S103. Determine a target avoidance amplitude according to the lane width or the lateral intrusion degree.

[0056] Wherein, the target avoidance amplitude includes a target lateral offset amplitude (bias_) and the total avoidance time (total_nudge_time_) required to reach the amplitude.

[0057] In one embodiment, determining the target avoidance range based on the lane width or the degree of lateral intrusion includes the following sub-steps S1031-S1033: S1031. Determine the intrusion level based on the degree of lateral intrusion of the target to avoid the obstacle, and determine the lateral offset amplitude of the target based on the intrusion level. Different intrusion levels correspond to different lateral offset amplitudes. or, The target lateral offset amplitude is determined based on the lane width, with different lane widths corresponding to different lateral offset amplitudes.

[0058] In one embodiment, determining the target lateral offset amplitude based on the intrusion level can be applied to the I4 architecture, that is, to the highway mode (nudge_mode == 1). In this case, the lateral intrusion depth of the obstacle (the relationship between max_l or min_l and the lane boundary) is divided into three levels (level_1 / 2 / 3), and different intrusion levels correspond to different lateral offset amplitudes (level_1, nudge, etc.). The higher the intrusion level (i.e., the deeper the obstacle intrudes), the larger the corresponding lateral offset amplitude.

[0059] In one embodiment, determining the target lateral offset amplitude based on lane width can be applied to the I2 architecture, that is, to the city mode (nudge_mode == 2). In this case, the target lateral offset amplitude can be determined by looking up a table. The wider the lane, the larger the target lateral offset amplitude. For example, a lane width of 3.5 meters corresponds to a target lateral offset amplitude of 0.275 meters, and a lane width of 3.75 meters corresponds to a target lateral offset amplitude of 0.4 meters.

[0060] S1033. Obtain the total avoidance time based on the target's lateral offset magnitude.

[0061] Total avoidance time = target lateral offset amplitude / (0.4 / config_.nudge_time()), which is the total number of frames required to reach the target lateral offset amplitude starting from 0 (calculated based on a 25Hz control cycle). This total avoidance time is used to control the rate of change of offset during subsequent avoidance processes, ensuring that the avoidance action is neither too fast nor too slow. Here, config_.nudge_time() is the preset maximum lane avoidance execution time.

[0062] In one implementation, this disclosure employs a state machine to manage the lifecycle of the lateral avoidance function. The state machine includes at least the following three states: OFF state: This indicates that the lateral avoidance function is not activated, and the vehicle travels along the original reference line without any lateral offset.

[0063] RUN state (avoidance in progress): This indicates that the lateral avoidance function is in progress, and the vehicle is gradually increasing its lateral offset in the target direction to avoid the obstacle.

[0064] CANCEL status (returning to lane): indicates that the vehicle is exiting the avoidance state and is gradually returning to the lane centerline from its current lateral deviation.

[0065] The state transition logic of the state machine is as follows: (1) Switching conditions from OFF to RUN: When the avoidance trigger condition (ExitTrigger) is met, that is, there is a valid target obstacle to avoid and it is not in a gentle suppression state (i.e. it is not suppressed by the pinch scene detection or safety detection), the state machine switches from OFF to RUN and begins to perform lateral avoidance.

[0066] (2) Switching conditions from RUN to CANCEL: In the RUN state, if any of the following exit trigger conditions are met, the state machine switches from RUN to CANCEL and begins the process of returning to the center of the lane: The target obstacle is lost (i.e., the target obstacle is not perceived in multiple consecutive frames); Triggering low-level alarms (e.g., the longitudinal distance of an obstacle exceeds the safe range, or the lateral position of an obstacle changes abruptly). The avoidance direction is inconsistent (that is, the currently calculated avoidance direction is opposite to the previously recorded avoidance direction, indicating that the position of the obstacle has changed drastically).

[0067] (3) Switching conditions for RUN or CANCEL→OFF: In the RUN or CANCEL state, if a high-level alarm is triggered (e.g., the vehicle enters the adaptive cruise control (ACC) mode, the driver actively takes over the steering, the vehicle performs an automatic lane change operation, or the system detects a higher level of safety risk), the state machine immediately switches back to the OFF state and resets all parameters related to avoidance (including the accumulator counter, current offset, target obstacle ID, etc.) to ensure that the high-level driving task is not disturbed.

[0068] Through the aforementioned state machine management mechanism, refined state control of the avoidance function is achieved, ensuring that the avoidance behavior is triggered, executed, and smoothly exited at the appropriate time. This avoids false triggering and invalid avoidance caused by perceived jitter or changes in system state, thereby improving the robustness and security of the system.

[0069] S104. Based on the target avoidance range, generate a lateral avoidance path within the lane where the vehicle is located.

[0070] Avoidance path generation is performed only when the state machine is in the RUN state, return path generation is performed in the CANCEL state, and no lateral offset is added in the OFF state.

[0071] In this embodiment, according to the target avoidance range (including the target lateral offset amplitude and the total avoidance time), a segmented S-curve is adopted to control the change process of the offset, so as to generate a smooth lateral avoidance path.

[0072] In one embodiment, the generating a lateral avoidance path in the lane where the host vehicle is located based on the target avoidance range includes the following sub-steps S1041 to S1047: S1041. When it is detected that the vehicle enters the avoidance operation state, the avoidance operation process is divided into an acceleration section and a deceleration section based on the total_nudge_time.

[0073] For example, the total avoidance time may be equally divided, with half of the time being the avoidance acceleration section and the other half being the avoidance deceleration section. The avoidance acceleration section is the time interval from 0 to total_nudge_time_ / 2, and the avoidance deceleration section is the time interval from total_nudge_time_ / 2 to total_nudge_time_.

[0074] S1042. In each perception frame, it is determined whether the current perception frame is in the avoidance acceleration section or the avoidance deceleration section according to the duration that the current avoidance has lasted.

[0075] For example, when it is detected that the vehicle enters the avoidance operation state (RUN state), an accumulative counter (slice_count_) is configured, the accumulative counter is incremented in each perception frame, and the accumulative counter is used to represent the duration t that the avoidance has lasted.

[0076] If t < total_nudge_time_ / 2, the vehicle is in the avoidance acceleration section; If t ≥ total_nudge_time_ / 2, the vehicle is in the avoidance deceleration section.

[0077] S1043. If the current perception frame is in the avoidance acceleration section, the lateral offset of the current perception frame is calculated according to the first constant acceleration value and the duration that the avoidance has lasted, and the lateral offset of the current perception frame is positively correlated with the square of the duration that the avoidance has lasted.

[0078] Specifically, in the avoidance acceleration section, the lateral offset is calculated according to the uniform acceleration change rule: B=0.5×acc×t², where acc is the first constant acceleration value, acc=bias_ / ((total_nudge_time_ / 2)²), and bias_ is the target lateral offset amplitude.

[0079] S1044. If the current sensing frame is in the avoidance deceleration phase, calculate the lateral offset of the current sensing frame based on the second constant acceleration value and the remaining avoidance time. The lateral offset of the current sensing frame is negatively correlated with the square of the remaining avoidance time. The second constant acceleration value is equal in magnitude and opposite in direction to the first constant acceleration value.

[0080] The deceleration phase and the acceleration phase are symmetrical, with equal magnitudes and opposite directions of acceleration.

[0081] The second constant acceleration value is -acc, and the formula for the lateral offset during the avoidance deceleration phase can be written as: bias = B - 0.5 acc τ 2 , where τ=total_nudge_time_-t is the remaining time for avoidance.

[0082] S1045. When the duration of the current avoidance is the same as the total avoidance time, it is determined that the avoidance deceleration phase has ended and the lateral offset has reached the target lateral offset amplitude.

[0083] When t = total_nudge_time, τ = 0. Substituting this into the avoidance deceleration formula, bias = B - 0.5 acc 0 2 =B, reaching the target avoidance offset value B.

[0084] By using the above-mentioned segmented uniform acceleration-uniform deceleration S-curve control, the change in lateral offset is made continuous and smooth, without abrupt changes.

[0085] S1046. After calculating the lateral offset in each frame, perform a safety check (IsClearToNudge). If the safety check result confirms that there is no obstacle to avoid, use the currently calculated lateral offset. If there is an obstacle to avoid, use the lateral offset obtained in the previous sensing frame.

[0086] In related technologies, the lateral offset change from normal driving to obstacle avoidance often adopts a step or linear abrupt change, resulting in discontinuous yaw motion of the vehicle and affecting ride comfort. This disclosure adopts a piecewise S-curve (uniform acceleration-uniform deceleration) to control the change process of lateral offset, making the lateral motion of the vehicle smooth and without abrupt changes, improving ride comfort and reducing occupant discomfort.

[0087] Safety detection (IsClearToNudge) is a real-time safety check performed during the lateral avoidance path generation process. After calculating the lateral offset in each perception frame, the offset is not used directly. Instead, a safety detection is performed first to confirm that there is no collision risk in the adjacent lanes in the target avoidance direction before the lateral offset of the current frame is used.

[0088] The security detection includes the following sub-steps C1-C2: C1. Obtain the lateral coordinate range of each obstacle in the Frenet coordinate system.

[0089] First, the position information of the vehicle and surrounding obstacles is projected uniformly onto the Frenet coordinate system. The Frenet coordinate system uses the reference line of the vehicle's lane as a reference, where the reference line is the centerline of the lane. In the Frenet coordinate system, the s-coordinate represents the longitudinal distance along the reference line, and the l-coordinate represents the lateral offset relative to the reference line.

[0090] (1) Obtain the longitudinal occupied interval of the vehicle [s_start, s_end].

[0091] By combining the projected coordinates 's' of the vehicle on the reference line with the distances from the front to the front overhang (base2front_) and from the rear to the rear overhang (head_base2tail_), the longitudinal interval [s_start, s_end] occupied by the vehicle on the reference line is obtained. Specifically: The lower limit of the longitudinal section occupied by the vehicle is s_start = s_ego - head_base2tail_, where s_ego is the projection s coordinate of the vehicle's reference point (such as the rear axle center) on the reference line, and head_base2tail_ is the distance from the rear of the vehicle to the rear overhang. The upper limit of the longitudinal section occupied by the vehicle is s_end = s_ego + base2front_, where base2front_ is the distance from the front of the vehicle to the front overhang.

[0092] This gives us the longitudinal occupancy range [s_start, s_end] of the vehicle on the reference line, which represents the longitudinal space range occupied by the vehicle along the road direction.

[0093] (2) Obtain the longitudinal and transverse coordinate ranges of the obstacle.

[0094] For each obstacle detected, the bounding points on its outline are projected onto the same reference line to obtain the vertical coordinates (s-coordinates) and horizontal coordinates (l-coordinates) of each bounding point in the Frenet coordinate system.

[0095] The minimum value of the s coordinates of each boundary point of the obstacle is taken as the lower limit obs_s_min of the longitudinal interval of the obstacle, and the maximum value of the s coordinates of each boundary point of the obstacle is taken as the upper limit obs_s_max of the longitudinal interval of the obstacle, thus obtaining the longitudinal interval [obs_s_min, obs_s_max] of the obstacle on the reference line.

[0096] The minimum value of the l coordinates of each edge point of the obstacle is taken as the minimum value min_l of the horizontal coordinate interval of the obstacle, and the maximum value of the l coordinates of each edge point of the obstacle is taken as the maximum value max_l of the horizontal coordinate interval of the obstacle, thus obtaining the horizontal coordinate interval [min_l, max_l] of the obstacle on the reference line.

[0097] C2. If the lateral coordinate range of the obstacle overlaps with the lane area in the target offset direction, and the longitudinal distance between the obstacle and the vehicle is less than the safe distance determined based on relative speed, then the current avoidance is deemed unsafe and there is an obstacle to avoidance.

[0098] After obtaining the longitudinal occupancy interval [s_start, s_end] of the vehicle, the longitudinal interval [obs_s_min, obs_s_max] of the obstacle, and the lateral coordinate interval [min_l, max_l], the following judgment is performed on each obstacle.

[0099] (1) Horizontal overlap judgment.

[0100] Determine whether the lateral coordinate interval [min_l, max_l] of the obstacle overlaps with the lane area in the direction the target avoids.

[0101] The lane area in the target avoidance direction refers to: when the vehicle plans to avoid to the left, it is the left adjacent lane area; when the vehicle plans to avoid to the right, it is the right adjacent lane area. In the Frenet coordinate system, this lane area can be represented as [-lane_width / 2, lane_width / 2], the left adjacent lane area can be represented as [-lane_width / 2 -lane_width, -lane_width / 2], and the right adjacent lane area can be represented as [lane_width / 2, lane_width / 2 + lane_width].

[0102] When making the judgment, the influence of the lateral offset (bias) already generated by the vehicle also needs to be considered. Specifically, based on the lateral offset bias calculated in the current perception frame, the actual lateral range occupied by the vehicle in the target avoidance direction is determined, and this lateral range is merged with the lane area in the target avoidance direction to obtain the target avoidance area after considering the influence of bias. Then, it is determined whether the lateral coordinate interval [min_l, max_l] of the obstacle overlaps with the target avoidance area.

[0103] If the lateral coordinate range of the obstacle overlaps with the target avoidance area, the lateral condition is met, indicating that the obstacle is located within the lane area that the vehicle wants to veer into; otherwise, the lateral condition is not met, indicating that the obstacle is not in the target avoidance direction and does not pose a threat.

[0104] (2) Calculation of longitudinal distance.

[0105] If the lateral conditions are met, then the longitudinal distance between the obstacle and the vehicle is further calculated.

[0106] (3) Determining the safe distance.

[0107] The safe distance is calculated based on the relative speed between the vehicle and the obstacle, and is used to avoid a rear-end collision with the obstacle after the vehicle veers laterally. The greater the relative speed, the greater the required safe distance.

[0108] When both lateral and longitudinal conditions are met—that is, the lateral coordinate range of the obstacle overlaps with the target avoidance area, and the longitudinal distance between the obstacle and the vehicle is less than the corresponding safe distance—then the current avoidance is deemed unsafe, and an obstacle to avoidance exists. In this case, further increases in the lateral offset are prohibited (stop_nudge = true). The current perception frame does not use the newly calculated lateral offset but instead uses the lateral offset from the previous perception frame to maintain the vehicle's current lateral position, waiting for the obstacle in the adjacent lane to pass or move away before continuing the avoidance operation.

[0109] This disclosure expands the detection area in the target offset direction to check whether there are longitudinally overlapping obstacles in adjacent lanes, thereby preventing lateral collisions and ensuring the safety of the avoidance path.

[0110] S1047. Obtain the lateral avoidance path based on the lateral offset used in each sensing frame.

[0111] The lateral offsets used in each sensing frame are combined in chronological order to form a continuous lateral avoidance trajectory within the lane where the vehicle is located, thus obtaining the lateral avoidance path. This lateral avoidance path is generated by dynamically superimposing the lateral offset (B) on the original reference line.

[0112] S105. Control the vehicle to perform lateral avoidance according to the lateral avoidance path.

[0113] In this embodiment, the planning module outputs the lateral avoidance path to the control module. The control module controls the vehicle's steering, throttle, and braking according to the lateral avoidance path, so that the vehicle travels along the lateral avoidance path and achieves lateral avoidance.

[0114] In one embodiment, when the avoidance exit conditions are met, the planning module of the autonomous vehicle executes a return process, smoothly returning the vehicle from the current avoidance state to the lane centerline. The avoidance exit conditions include, for example, that the vehicle has passed the target obstacle, the target obstacle has disappeared, the target obstacle no longer meets the avoidable candidate conditions, or a low-level alarm has been received. The return process and the aforementioned avoidance operation process employ symmetrical S-curve control logic to ensure that the change in lateral offset is continuous, smooth, and shock-free. Specifically, the method further includes the following sub-steps D1 to D6: D1. Obtain the actual horizontal offset used in the current sensing frame as the return starting offset.

[0115] When the avoidance exit condition is met, the planning module records the actual lateral offset used in the current sensing frame and uses it as the return starting offset (denoted as bias_ret). The actual lateral offset used in the current sensing frame refers to the lateral offset that is finally determined after safety detection and used to generate the avoidance path. For example, if the lateral offset of the previous sensing frame is used due to the safety detection determining that there is an obstacle to avoidance during the avoidance process, then the return starting offset should be the actual offset that was used, not the offset calculated in the current frame but not used.

[0116] The purpose of recording the initial offset of the return is to provide an initial value for the return process, so that the vehicle can smoothly return from its current actual lateral position and avoid shocks caused by jumping directly to zero or suddenly reversing.

[0117] D2. Based on the return start offset, determine the total return time and divide the return process into a return acceleration segment and a return deceleration segment.

[0118] The total return time (denoted as T_ret) is determined based on the return starting offset. Specifically, the total return time is directly proportional to the return starting offset and can be determined based on the return starting offset and a preset offset rate. For example, the same offset rate calculation method as the avoidance process can be used: T_ret = bias_ret / (0.4 / config_.nudge_time()); Herein, config_.nudge_time() is the preset maximum lane avoidance execution time, which shares the same parameter with the avoidance process. This calculation method indicates that the larger the return starting offset, the longer the total return duration; the smaller the return starting offset, the shorter the total return duration, thereby ensuring that the lateral speed of the return process is equivalent to that of the avoidance process, and realizing symmetric smooth regression.

[0119] After the total return duration is determined, the return process is divided into a return acceleration phase and a return deceleration phase. Specifically, the total return duration may be equally divided into two segments: the return acceleration phase is the time interval from 0 to T_ret / 2, and the return deceleration phase is the time interval from T_ret / 2 to T_ret. The division manner of the return acceleration phase and the return deceleration phase is symmetric to the division manner of the acceleration phase and the deceleration phase in the avoidance process.

[0120] D3. In each perception frame, it is determined whether the current perception frame is in the return acceleration phase or the return deceleration phase according to the elapsed return duration.

[0121] During the return process, a return accumulation counter is configured, and the return accumulation counter is incremented in each perception frame to obtain the elapsed return duration (denoted as t_ret). In each perception frame, according to the relationship between the elapsed return duration t_ret and the total return duration T_ret, it is determined whether the current perception frame is in the return acceleration phase or the return deceleration phase: If t_ret < T_ret / 2, the current perception frame is in the return acceleration phase; If t_ret ≥ T_ret / 2, the current perception frame is in the return deceleration phase.

[0122] By performing frame-by-frame judgment, a corresponding lateral offset calculation method can be selected for the return phase where the current perception frame is located, so as to ensure that the return process is executed step by step according to the predetermined S-shaped curve.

[0123] D4. If the current perception frame is in the return acceleration phase, the lateral offset is gradually decreased starting from the return starting offset according to a third constant acceleration value.

[0124] If the current perception frame is in the return acceleration phase, the lateral offset is gradually decreased starting from the return starting offset bias_ret according to a third constant acceleration value (denoted as a3). Specifically, within the return acceleration phase, the variation rule of the lateral offset with time is: bias(t_ret) = bias_ret - 0.5 × a3 × t_ret²; Herein, a3 is the third constant acceleration value, and its value is determined according to the return starting offset bias_ret and the total return duration T_ret, for example: a3 = bias_ret / ( (T_ret / 2)² ); As can be seen from the above formula, during the return acceleration phase, the lateral offset gradually decreases from bias_ret, and the rate of decrease accelerates (because the decreasing displacement is proportional to the square of time). This phase corresponds to the "accelerated decrease" process of the lateral offset returning to zero from its current value.

[0125] D5. If the current sensing frame is in the return deceleration phase, the lateral offset continues to decrease gradually according to the fourth constant acceleration value, which is equal in magnitude and opposite in direction to the third constant acceleration value.

[0126] If the current sensing frame is in the return deceleration phase, the lateral offset continues to gradually decrease according to the fourth constant acceleration value (denoted as a4). The fourth constant acceleration value a4 is equal in magnitude and opposite in direction to the third constant acceleration value a3, i.e., a4 = -a3. Within the return deceleration phase, the lateral offset changes over time as follows: bias(t_ret) = 0.5 × a3 × (T_ret - t_ret)²; Where (T_ret - t_ret) represents the remaining return time. As can be seen from the formula above, during the deceleration phase of the return, the lateral offset continues to decrease, but at a slower rate (because the square of the remaining time gradually decreases), until the remaining return time decreases to zero, at which point the lateral offset also decreases to zero. This phase corresponds to the "deceleration reduction" process where the lateral offset returns to zero from its value at the end of the acceleration phase.

[0127] By using a symmetrical design for the return acceleration and return deceleration phases, the lateral offset variation curve is S-shaped, smoothly transitioning from bias_ret to zero, thus avoiding lateral oscillations during the return process.

[0128] D6. When the duration of the return is the same as the total duration of the return, or when the lateral offset decreases to less than a preset threshold, the return process is determined to be over and the avoidance operation state is exited.

[0129] In each sensing frame, it is determined whether the return process meets the termination condition. Specifically, the return process is determined to have ended when any of the following conditions are met: The duration of the return t_ret is the same as the total duration of the return T_ret (i.e., t_ret ≥ T_ret). The lateral offset is reduced to less than a preset threshold (e.g., 0.01 meters).

[0130] When the return process is complete, the vehicle exits the avoidance operation state, switches the state machine from the CANCEL state back to the OFF state, and resets parameters such as the return accumulator counter and the return starting offset. The vehicle then travels normally along the original reference line.

[0131] In related technologies, after overtaking an obstacle, the process of a vehicle returning to the lane centerline from a yielding state usually involves a fixed slope change, which can easily lead to lateral oscillations or slow recovery. In this disclosure, an S-shaped curve or a linear gradient method symmetrical to the yielding process is used to reduce the offset, thereby achieving shock-free recovery and avoiding lateral swaying.

[0132] It is worth noting that for embedded platforms with limited computing resources, a linear gradient method can be used instead of an S-curve. After obtaining the target avoidance range, during the avoidance process, the current bias is increased by a fixed step size delta_bias_ (e.g., 0.02 meters) in each frame until the target avoidance range is reached. During the return process, the corresponding step size is decreased in each frame.

[0133] In one embodiment, before controlling the vehicle to perform lateral avoidance based on the lateral avoidance path, the method further includes a pinch scene detection step: like Figure 5 As shown, the system detects whether there are avoidable obstacles on both the left and right sides of the vehicle. When both avoidable obstacles on the left and right sides are detected simultaneously (i.e., nudgable_obstacles_ contains both left and right avoidable obstacles), it is determined that the vehicle is currently in a sandwich scenario. The system prohibits the vehicle from performing lateral avoidance according to the lateral avoidance path and clears all candidate obstacles to prevent the vehicle from blindly avoiding obstacles on both sides and being sandwiched, thus ensuring safety.

[0134] It is worth noting that the method disclosed herein can be applied to: Autonomous commercial vehicles (trucks): Especially suitable for long-bodied vehicles with trailers, which have higher requirements for smooth lateral avoidance.

[0135] Autonomous passenger vehicles: adaptable to urban and highway scenarios.

[0136] Mobile robot obstacle avoidance: When sweeping robots, warehouse AGVs, and other similar devices avoid static obstacles in narrow passages, they can use S-shaped curves for lateral offset.

[0137] Drone swarm flight: Avoiding sudden obstacles and moving smoothly laterally in dense formation.

[0138] Driver assistance system: As a lateral assistance function for the driver, it actively avoids large vehicles that intrude into the adjacent lane while maintaining lane position.

[0139] Autonomous driving domain controller: A software algorithm module that integrates the method of this invention.

[0140] The smallest unit available for sale: a planning algorithm software package containing the method of the present invention, or a functional module pre-installed in an autonomous driving system.

[0141] The core algorithm disclosed herein—dynamic offset control based on S-curves—can be extended to longitudinal avoidance (such as smooth speed changes during lane changes) or, more generally, transition segment generation in path planning.

[0142] Figure 6 This is a schematic diagram of a lateral avoidance device for an autonomous vehicle according to an exemplary embodiment, such as... Figure 6 As shown, it includes: The planning module is used to obtain lane information of the lane in which the vehicle is located and spatial location information of the target obstacle to avoid. The lane information includes: lane width; The planning module is also used to calculate the degree of lateral intrusion of the target obstacle relative to the lane where the vehicle is located, based on the spatial location information of the target obstacle. The planning module is also used to determine the target avoidance range based on the lane width or the degree of lateral intrusion; The planning module is also used to generate a lateral avoidance path within the lane where the vehicle is located based on the target avoidance range; The control module is used to control the vehicle to perform lateral avoidance according to the lateral avoidance path.

[0143] In one embodiment, the planning module is further configured to: Candidate determination is performed on each obstacle detected according to the sensing frame period: For each obstacle, calculate the minimum value of the longitudinal coordinate of the obstacle's various boundary points in the Frenet coordinate system, and calculate the maximum value of the vehicle's longitudinal coordinate in the Frenet coordinate system; the Frenet coordinate system is based on the reference line of the lane where the vehicle is located; the longitudinal distance between the obstacle and the vehicle is obtained based on the minimum value of the obstacle's longitudinal coordinate and the maximum value of the vehicle's longitudinal coordinate. Obstacles that simultaneously meet the following conditions are identified as avoidable candidate obstacles: The obstacle's speed is less than the vehicle's speed; the longitudinal distance between the obstacle and the vehicle is within a preset effective range; and the estimated collision time of the obstacle is less than the collision time threshold. For each dodgeable candidate obstacle, in each perception frame, if the obstacle is determined to be a dodgeable candidate obstacle, the corresponding frame count is increased; if the obstacle is not determined to be a dodgeable candidate obstacle, the corresponding frame count is decreased; dodgeable candidate obstacles with a frame count greater than a preset counting threshold are identified as target dodgeable obstacles.

[0144] In one embodiment, the planning module is further configured to: Calculate the maximum and minimum values ​​of the lateral coordinates of each edge point of the target obstacle avoidance in the Frenet coordinate system; If the maximum value of the lateral coordinate of the target obstacle falls within the first preset range, it is determined to be an obstacle that can be avoided on the left. If the minimum value of the lateral coordinate of the obstacle to be avoided by the target falls within the second preset range, it is determined to be an obstacle that can be avoided on the right.

[0145] In one embodiment, the lower limit of the first preset interval is: the position after the left lane boundary of the lane where the vehicle is located is offset away from the reference line of the vehicle by a preset lateral threshold. The upper limit of the first preset interval is: the position after the left lane boundary is offset towards the reference line of the vehicle by a preset avoidable adjustment width; The lower limit of the second preset interval is: the position after the right lane boundary of the lane where the vehicle is located is offset towards the reference line of the vehicle by a preset avoidable adjustment width; The upper limit of the second preset interval is: the position after the right lane boundary is offset away from the vehicle reference line by a preset lateral threshold; Specifically, with the direction the vehicle's front is facing forward, when facing forward, the left lane boundary is the lane boundary located on the left side of the vehicle, and the right lane boundary is the lane boundary located on the right side of the vehicle; the vehicle reference line is the center line of the lane in which the vehicle is located.

[0146] In one embodiment, the planning module is further configured to: The intrusion level is determined based on the degree of lateral intrusion of the target obstacle, and the lateral offset amplitude of the target is determined based on the intrusion level. Different intrusion levels correspond to different lateral offset amplitudes. or, The target lateral offset amplitude is determined based on the lane width, with different lane widths corresponding to different lateral offset amplitudes; The total avoidance time is obtained based on the target lateral offset amplitude, whereby the target avoidance amplitude includes both the target lateral offset amplitude and the total avoidance time.

[0147] In one embodiment, the planning module is further configured to: When the system detects that it has entered the obstacle avoidance operation state, the obstacle avoidance operation process is divided into an acceleration phase and a deceleration phase based on the total obstacle avoidance time. In each perception frame, based on the duration of the current avoidance, it is determined whether the current perception frame is in the avoidance acceleration phase or the avoidance deceleration phase. If the current sensing frame is in the avoidance acceleration phase, the lateral offset of the current sensing frame is calculated based on the first constant acceleration value and the duration of avoidance. The lateral offset of the current sensing frame is positively correlated with the square of the duration of avoidance. If the current sensing frame is in the avoidance deceleration phase, the lateral offset of the current sensing frame is calculated based on the second constant acceleration value and the remaining avoidance time. The lateral offset of the current sensing frame is negatively correlated with the square of the remaining avoidance time. The second constant acceleration value is equal in magnitude and opposite in direction to the first constant acceleration value. When the duration of the current avoidance is the same as the total avoidance time, it is determined that the avoidance deceleration phase has ended and the lateral offset has reached the target lateral offset amplitude. After calculating the lateral offset for each frame, a safety check is performed. If the safety check confirms that there is no obstacle to avoid, the currently calculated lateral offset is used. If there is an obstacle to avoid, the lateral offset obtained from the previous sensing frame is used. The lateral avoidance path is obtained based on the lateral offset used in each sensing frame.

[0148] In one embodiment, the security detection includes: Obtain the lateral coordinate range of each obstacle in the Frenet coordinate system; If the lateral coordinate range of the obstacle overlaps with the lane area in the target offset direction, and the longitudinal distance between the obstacle and the vehicle is less than the safe distance determined based on relative speed, then the current avoidance is deemed unsafe and there is an obstacle to avoidance.

[0149] In one embodiment, the planning module is further configured to: When the avoidance exit condition is met, the actual lateral offset used in the current perception frame is obtained as the return starting offset; The total return time is determined based on the return start offset, and the return process is divided into a return acceleration segment and a return deceleration segment; In each sensing frame, the current sensing frame is determined to be in the return acceleration phase or the return deceleration phase based on the duration of the return. If the current sensing frame is in the return acceleration segment, the lateral offset is gradually reduced from the return start offset according to the third constant acceleration value; If the current sensing frame is in the return deceleration phase, the lateral offset continues to decrease gradually according to the fourth constant acceleration value, which is equal in magnitude and opposite in direction to the third constant acceleration value. When the duration of the return is the same as the total return duration, or when the lateral offset decreases to less than a preset threshold, the return process is determined to be over and the avoidance operation state is exited.

[0150] In one embodiment, the planning module is further configured to: Detect whether there are avoidable obstacles on both the left and right sides of the vehicle at the same time; When both avoidable obstacles on the left and right are detected simultaneously, the vehicle is prohibited from performing lateral avoidance according to the lateral avoidance path.

[0151] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0152] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A method for lateral obstacle avoidance by an autonomous vehicle, characterized in that, The method includes: The system acquires lane information of the vehicle's lane and spatial location information of the obstacle the target is avoiding. The lane information includes: lane width. Based on the spatial location information of the target obstacle, the degree of lateral intrusion of the target obstacle relative to the lane where the vehicle is located is calculated; The target avoidance range is determined based on the lane width or the degree of lateral intrusion. Based on the target avoidance range, a lateral avoidance path is generated within the lane where the vehicle is located; The vehicle is controlled to perform lateral avoidance according to the lateral avoidance path.

2. The method according to claim 1, characterized in that, The method further includes: Candidate determination is performed on each obstacle detected according to the sensing frame period: For each obstacle, calculate the minimum value of the longitudinal coordinate of the obstacle's various boundary points in the Frenet coordinate system, and calculate the maximum value of the vehicle's longitudinal coordinate in the Frenet coordinate system; the Frenet coordinate system is based on the reference line of the lane where the vehicle is located; the longitudinal distance between the obstacle and the vehicle is obtained based on the minimum value of the obstacle's longitudinal coordinate and the maximum value of the vehicle's longitudinal coordinate. Obstacles that simultaneously meet the following conditions are identified as avoidable candidate obstacles: The obstacle's speed is less than the vehicle's speed; the longitudinal distance between the obstacle and the vehicle is within a preset effective range; and the estimated collision time of the obstacle is less than the collision time threshold. For each dodgeable candidate obstacle, in each perception frame, if the obstacle is determined to be a dodgeable candidate obstacle, the corresponding frame count is increased; if the obstacle is not determined to be a dodgeable candidate obstacle, the corresponding frame count is decreased; dodgeable candidate obstacles with a frame count greater than a preset counting threshold are identified as target dodgeable obstacles.

3. The method according to claim 2, characterized in that, The method further includes: Calculate the maximum and minimum values ​​of the lateral coordinates of each edge point of the target obstacle avoidance in the Frenet coordinate system; If the maximum value of the lateral coordinate of the target obstacle falls within the first preset range, it is determined to be an obstacle that can be avoided on the left. If the minimum value of the lateral coordinate of the obstacle to be avoided by the target falls within the second preset range, it is determined to be an obstacle that can be avoided on the right.

4. The method according to claim 3, characterized in that, The lower limit of the first preset interval is: the position after the left lane boundary of the lane where the vehicle is located is offset away from the reference line of the vehicle by a preset lateral threshold. The upper limit of the first preset interval is: the position after the left lane boundary is offset towards the reference line of the vehicle by a preset avoidable adjustment width; The lower limit of the second preset interval is: the position after the right lane boundary of the lane where the vehicle is located is offset towards the reference line of the vehicle by a preset avoidable adjustment width; The upper limit of the second preset interval is: the position after the right lane boundary is offset away from the vehicle reference line by a preset lateral threshold; Specifically, with the direction the vehicle's front is facing forward, when facing forward, the left lane boundary is the lane boundary located on the left side of the vehicle, and the right lane boundary is the lane boundary located on the right side of the vehicle; the vehicle reference line is the center line of the lane in which the vehicle is located.

5. The method according to claim 2, characterized in that, Determining the target avoidance range based on the lane width or the degree of lateral intrusion includes: The intrusion level is determined based on the degree of lateral intrusion of the target obstacle, and the lateral offset amplitude of the target is determined based on the intrusion level. Different intrusion levels correspond to different lateral offset amplitudes. or, The target lateral offset amplitude is determined based on the lane width, with different lane widths corresponding to different lateral offset amplitudes; The total avoidance time is obtained based on the target lateral offset amplitude, whereby the target avoidance amplitude includes both the target lateral offset amplitude and the total avoidance time.

6. The method according to claim 5, characterized in that, The step of generating a lateral avoidance path within the lane where the vehicle is located based on the target avoidance range includes: When the system detects that it has entered the obstacle avoidance operation state, the obstacle avoidance operation process is divided into an acceleration phase and a deceleration phase based on the total obstacle avoidance time. In each perception frame, based on the duration of the current avoidance, it is determined whether the current perception frame is in the avoidance acceleration phase or the avoidance deceleration phase. If the current sensing frame is in the avoidance acceleration phase, the lateral offset of the current sensing frame is calculated based on the first constant acceleration value and the duration of avoidance. The lateral offset of the current sensing frame is positively correlated with the square of the duration of avoidance. If the current sensing frame is in the avoidance deceleration phase, the lateral offset of the current sensing frame is calculated based on the second constant acceleration value and the remaining avoidance time. The lateral offset of the current sensing frame is negatively correlated with the square of the remaining avoidance time. The second constant acceleration value is equal in magnitude and opposite in direction to the first constant acceleration value. When the duration of the current avoidance is the same as the total avoidance time, it is determined that the avoidance deceleration phase has ended and the lateral offset has reached the target lateral offset amplitude. After calculating the lateral offset for each frame, a safety check is performed. If the safety check confirms that there is no obstacle to avoid, the currently calculated lateral offset is used. If there is an obstacle to avoid, the lateral offset obtained from the previous sensing frame is used. The lateral avoidance path is obtained based on the lateral offset used in each sensing frame.

7. The method according to claim 6, characterized in that, The security checks include: Obtain the lateral coordinate range of each obstacle in the Frenet coordinate system; If the lateral coordinate range of the obstacle overlaps with the lane area in the target offset direction, and the longitudinal distance between the obstacle and the vehicle is less than the safe distance determined based on relative speed, then the current avoidance is deemed unsafe and there is an obstacle to avoidance.

8. The method according to claim 6, characterized in that, The method further includes: When the avoidance exit condition is met, the actual lateral offset used in the current perception frame is obtained as the return starting offset; The total return time is determined based on the return start offset, and the return process is divided into a return acceleration segment and a return deceleration segment; In each sensing frame, the current sensing frame is determined to be in the return acceleration phase or the return deceleration phase based on the duration of the return. If the current sensing frame is in the return acceleration segment, the lateral offset is gradually reduced from the return start offset according to the third constant acceleration value; If the current sensing frame is in the return deceleration phase, the lateral offset continues to decrease gradually according to the fourth constant acceleration value, which is equal in magnitude and opposite in direction to the third constant acceleration value. When the duration of the return is the same as the total return duration, or when the lateral offset decreases to less than a preset threshold, the return process is determined to be over and the avoidance operation state is exited.

9. The method according to claim 8, characterized in that, Before controlling the vehicle to perform lateral avoidance according to the lateral avoidance path, the method further includes: Detect whether there are avoidable obstacles on both the left and right sides of the vehicle at the same time; When both avoidable obstacles on the left and right are detected simultaneously, the vehicle is prohibited from performing lateral avoidance according to the lateral avoidance path.

10. A lateral obstacle avoidance device for an autonomous vehicle, characterized in that, The device includes: The planning module is used to obtain lane information of the lane in which the vehicle is located and spatial location information of the target obstacle to avoid. The lane information includes: lane width; The planning module is also used to calculate the degree of lateral intrusion of the target obstacle relative to the lane where the vehicle is located, based on the spatial location information of the target obstacle. The planning module is also used to determine the target avoidance range based on the lane width or the degree of lateral intrusion; The planning module is also used to generate a lateral avoidance path within the lane where the vehicle is located based on the target avoidance range; The control module is used to control the vehicle to perform lateral avoidance according to the lateral avoidance path.