Swarm intelligence-based dynamic obstacle avoidance methods for robots and robots

Abstract

This invention provides a robot dynamic obstacle avoidance method and robot based on swarm intelligence. The method includes: dividing the space into a near-field avoidance zone, a mid-field interaction zone, and a far-field guidance zone based on the robot's current pose and environmental perception data; performing obstacle avoidance search based on obstacle information from the first swarm intelligence and the near-field avoidance zone to obtain a safe movement boundary; performing obstacle bypass search based on obstacle information from the second swarm intelligence and the mid-field interaction zone to obtain a trend guidance direction; and performing obstacle crossing search based on obstacle information from the third swarm intelligence and the far-field guidance zone to obtain a global reference path; performing asynchronous coordination and spatiotemporal coupling based on the safe movement boundary, the trend guidance direction, and the global reference path to obtain the optimal obstacle avoidance trajectory; and mapping the optimal obstacle avoidance trajectory into linear velocity and angular velocity components based on motion constraints to control the obstacle avoidance motion. This invention improves the stability of autonomous obstacle avoidance by the robot in complex dynamic environments.

Description

Technical Field

[0001] This invention relates to the field of robotics, and in particular to a method for dynamic obstacle avoidance of robots based on swarm intelligence, and a robot thereof. Background Technology

[0002] In the field of autonomous navigation for mobile robots, real-time obstacle avoidance in dynamic environments is a core technology that determines their operational safety. Existing technologies often employ schemes based on artificial potential field methods or a single global planning algorithm combined with local dynamic window methods.

[0003] However, these methods have significant drawbacks: First, a single algorithm cannot simultaneously meet the high real-time requirements of near-field emergency obstacle avoidance and the optimal requirements of far-field global path, causing the robot to easily get stuck in local stagnation or produce violent oscillations in complex dynamic environments; Second, in traditional hierarchical schemes, global planning and local obstacle avoidance often run independently or are simply executed sequentially, lacking deep information interaction and constraint coupling, causing local obstacle avoidance actions to frequently deviate from the global goal, or to produce unexecutable trajectories under the interference of dynamic obstacles; Third, existing schemes do not perform hierarchical adaptation for the robot's kinematic characteristics, and the generated obstacle avoidance paths often lack smoothness and are difficult to directly convert into stable motor control commands. Summary of the Invention

[0004] This invention provides a robot dynamic obstacle avoidance method and robot based on swarm intelligence. It aims to search for obstacles in different levels of areas by leveraging the different characteristics of multi-level swarm intelligence. By combining the asynchronous collaboration and spatiotemporal coupling of multi-level swarm intelligence, the robot can achieve dynamic autonomous obstacle avoidance, solve the problems of poor adaptability to dynamic environments and trajectory oscillation, and improve the stability of the robot's autonomous obstacle avoidance in complex dynamic environments.

[0005] In a first aspect, the present invention provides a robot dynamic obstacle avoidance method based on swarm intelligence, comprising: Based on the robot's current pose and environmental perception data, spatial classification is performed to obtain the near-field avoidance area, the mid-field interaction area, and the far-field guidance area. Based on the local repulsion characteristics of the first swarm intelligence combined with obstacle information in the near-field avoidance area, obstacle avoidance search is performed to obtain the safe movement boundary. Based on the dynamic following characteristics of the second swarm intelligence combined with obstacle information in the mid-field interaction area, obstacle detour search is performed to obtain the trend guidance direction. Based on the global topology exploration characteristics of the third swarm intelligence combined with obstacle information in the far-field guidance area, obstacle crossing search is performed to obtain the global reference path. The optimal obstacle avoidance trajectory is obtained by asynchronous coordination and spatiotemporal coupling based on safe motion boundaries, trend guidance directions and global reference paths; Based on the robot's motion constraints, the trajectory characteristics of the optimal obstacle avoidance trajectory are mapped to the linear velocity and angular velocity components of the robot chassis, thereby controlling the robot to perform obstacle avoidance motion.

[0006] In a second aspect, the present invention also provides a robot for implementing the swarm intelligence-based dynamic obstacle avoidance method for robots as described in the first aspect; the robot includes: The spatial grading module is used to perform spatial grading based on the robot's current pose and environmental perception data, resulting in a near-field avoidance area, a mid-field interaction area, and a far-field guidance area. The swarm intelligence collaboration module is used to search for obstacles based on the local repulsion characteristics of the first swarm intelligence combined with obstacle information in the near-field avoidance area to obtain a safe movement boundary; to search for obstacles based on the dynamic following characteristics of the second swarm intelligence combined with obstacle information in the mid-field interaction area to obtain a trend guidance direction; and to search for obstacles based on the global topology exploration characteristics of the third swarm intelligence combined with obstacle information in the far-field guidance area to obtain a global reference path. The obstacle avoidance trajectory planning module is used to asynchronously coordinate and spatiotemporally couple based on safe motion boundaries, trend guidance directions, and global reference paths to obtain the optimal obstacle avoidance trajectory. The obstacle avoidance control module is used to map the trajectory characteristics of the optimal obstacle avoidance trajectory to the linear velocity and angular velocity components of the robot chassis based on the robot's motion constraints, and control the robot to perform obstacle avoidance motion.

[0007] Thirdly, the present invention also provides an electronic device, comprising: a memory for storing computer software programs; and a processor for reading and executing the computer software programs, thereby realizing the robot dynamic obstacle avoidance method based on swarm intelligence as described above.

[0008] Fourthly, the present invention also provides a non-transitory computer-readable storage medium storing a computer software program, which, when executed by a processor, implements the robot dynamic obstacle avoidance method based on swarm intelligence as described above.

[0009] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the robot dynamic obstacle avoidance method based on swarm intelligence as described above.

[0010] The robot dynamic obstacle avoidance method based on swarm intelligence provided in this invention performs spatial hierarchical analysis of the robot's current pose and environmental perception data to obtain a near-field avoidance region, a mid-field interaction region, and a far-field guidance region. This solves the problem in existing technologies where a single algorithm cannot simultaneously achieve near-field real-time performance and far-field optimality. Based on the spatial hierarchical results, a three-layer swarm intelligence characteristic is specifically adopted to perform differentiated obstacle search in different regions. The local exclusion characteristic of the first swarm intelligence is used to process the near-field region to obtain a safe movement boundary for emergency obstacle avoidance, ensuring high real-time performance of near-field obstacle avoidance. The dynamic following characteristic of the second swarm intelligence is used to process the mid-field region to obtain a trend guidance direction for dynamic obstacle avoidance, adapting to the complex changes in the mid-field dynamic environment. The global topology exploration characteristic of the third swarm intelligence is used to process the far-field region to obtain a global reference path for static obstacle crossing, ensuring the optimality of the far-field path. This achieves precise adaptation between obstacle avoidance requirements in different regions and swarm intelligence characteristics, ensuring both the real-time performance of near-field emergency obstacle avoidance and the optimality of the far-field global path. At the same time, the hierarchical search avoids local stagnation or violent oscillations caused by a single algorithm. Based on safe motion boundaries, trend-guided directions, and global reference paths, this invention achieves deep information interaction and constraint coupling between global planning and local obstacle avoidance through asynchronous collaboration and spatiotemporal coupling. This solves the problems of local obstacle avoidance deviating from the global target and trajectory becoming unexecutable, which arise from independent operation or simple sequential execution of these two methods in traditional hierarchical schemes, resulting in the optimal obstacle avoidance trajectory. Based on the optimal obstacle avoidance trajectory and robot kinematic constraints, trajectory features are mapped to chassis linear velocity and angular velocity components. This addresses the problems of existing schemes not adapting to robot kinematic characteristics, insufficient trajectory smoothness, and difficulty in converting them into stable motor control commands, ensuring that the obstacle avoidance trajectory can be directly executed, thereby enabling stable obstacle avoidance movement control of the robot. Therefore, this invention utilizes the different characteristics of multi-level swarm intelligence to search for obstacles in different levels of regions. By combining the asynchronous collaboration and spatiotemporal coupling of multi-level swarm intelligence, it achieves dynamic autonomous obstacle avoidance for the robot, solving the problems of poor adaptability to dynamic environments and trajectory oscillations, and improving the stability of autonomous obstacle avoidance in complex dynamic environments. Attached Figure Description

[0011] Figure 1 This is a flowchart illustrating the robot dynamic obstacle avoidance method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of the robot provided in an embodiment of the present invention; Figure 3 An embodiment diagram of the electronic device provided in this invention; Figure 4 An embodiment diagram of a computer-readable storage medium provided in accordance with the present invention. Detailed Implementation

[0012] 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 embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] Optionally, see Figure 1 , Figure 1 This is a flowchart illustrating the robot dynamic obstacle avoidance method provided by the present invention. In this embodiment of the invention, the robot dynamic obstacle avoidance method based on swarm intelligence is executed by a robot.

[0014] Swarm intelligence-based dynamic obstacle avoidance methods for robots include: Step 10: Based on the robot's current pose and environmental perception data, perform spatial classification to obtain the near-field avoidance area, mid-field interaction area, and far-field guidance area.

[0015] Optionally, the robot collects environmental status information of the surrounding environment through its onboard environmental perception devices (including but not limited to lidar, visual cameras, and ultrasonic sensors), and obtains its current pose through its own positioning module (including but not limited to GPS positioning module, inertial measurement unit, and odometer). The current pose specifically refers to the robot's three-dimensional position coordinates and attitude angles in the world coordinate system. The attitude angles include the robot's heading angle, pitch angle, and roll angle. The heading angle is used to determine the robot's direction of travel, while the pitch angle and roll angle are used to determine whether the robot's body posture is stable.

[0016] The robot uses its current position as the center point and divides its surrounding space into levels according to preset distance and environmental complexity thresholds. The near-field avoidance zone is the area closest to the robot's current position and poses the greatest threat to its operational safety. This zone is primarily defined by distance thresholds, which are determined based on the robot's speed, braking performance, and the response speed of its environmental sensing devices, ensuring the robot can quickly respond to sudden obstacles and achieve emergency avoidance within this area. The mid-field interaction zone is located outside the near-field avoidance zone at a medium distance from the robot's current position. This is the main area where the robot interacts with dynamic obstacles. Its definition considers both distance and environmental complexity thresholds, with a focus on the movement range and trend of dynamic obstacles. The far-field guidance zone is located outside the mid-field interaction zone and furthest from the robot's current position. Its main function is to provide global path guidance for the robot. This zone is primarily defined by distance thresholds, while also considering the location of the global operational target, ensuring that it covers the main path range from the robot's current position to the target position.

[0017] In one embodiment, taking a mobile robot used for indoor warehousing operations as an example, the robot has a rated movement speed of 1.5 m / s, a braking distance of 0.3 m, a lidar response speed of 10 Hz, and a positioning accuracy of ±5 mm.

[0018] When the robot operates in the warehouse environment, it uses LiDAR to collect the location distribution information and motion status information of the surrounding shelves (static obstacles) and handling personnel (dynamic obstacles). The positioning module obtains its current position coordinates in the warehouse coordinate system (X=10.2 meters, Y=5.3 meters, Z=0.5 meters), and the attitude angles are 0° heading angle (traveling along the positive X-axis), 0° pitch angle, and 0° roll angle.

[0019] The distance threshold for the near-field avoidance zone is set at 0.8 meters. That is, the near-field avoidance zone is a spherical space with a radius of 0.8 meters centered on the robot's current position. Obstacles in this zone (including the corner of the shelf 0.5 meters away from the robot and a moving person 0.6 meters away from the robot) are all objects to be avoided in an emergency.

[0020] The distance threshold for the mid-field interaction zone is set to 0.8 meters to 3 meters. That is, the mid-field interaction zone is the part of a spherical space with a radius of 3 meters outside the near-field avoidance zone, centered on the robot's current position, minus the near-field avoidance zone. This area contains multiple slowly moving transport personnel and movable shelves (dynamic obstacles), and is the main area for the robot to interact with dynamic obstacles.

[0021] The distance threshold for the far-field guidance area is set to 3 to 10 meters. That is, the far-field guidance area is the spherical space with a radius of 10 meters outside the mid-field interaction area, centered on the robot's current position, minus the near-field avoidance area and the mid-field interaction area. This area mainly consists of fixed shelves (static obstacles) and includes the robot's target position (X=18.5 meters, Y=6.1 meters, Z=0.5 meters).

[0022] The boundaries of the three areas are clearly defined. The boundary between the near-field avoidance area and the mid-field interaction area is a sphere with a radius of 0.8 meters, and the boundary between the mid-field interaction area and the far-field guidance area is a sphere with a radius of 3 meters, ensuring that any obstacle in any position can be clearly assigned to the corresponding area during the robot's operation.

[0023] Step 20: Based on the local repulsion characteristics of the first swarm intelligence and the environmental state information of the near-field avoidance area, obstacle avoidance search is performed to obtain the safe movement boundary. Based on the dynamic following characteristics of the second swarm intelligence and the environmental state information of the mid-field interaction area, obstacle detour search is performed to obtain the trend guidance direction. Based on the global topology exploration characteristics of the third swarm intelligence and the environmental state information of the far-field guidance area, obstacle crossing search is performed to obtain the global reference path.

[0024] Optionally, for the near-field avoidance zone, the robot employs a first swarm intelligence with local repulsion characteristics. This swarm intelligence, combined with environmental information about the near-field avoidance zone, searches for obstacles to avoid. The goal is to quickly move away from all static and dynamic obstacles within this zone, obtaining a safe movement boundary that ensures the robot stays away from these obstacles. This safe movement boundary clearly indicates the spatial range within which the robot needs to deviate from all static and dynamic obstacles in the near-field avoidance zone, preventing collisions with near-field obstacles. For the mid-field interaction zone, a second swarm intelligence with dynamic following characteristics is employed. This second swarm intelligence, combined with environmental information about the mid-field interaction zone, searches for obstacle avoidance. The goal is to avoid dynamic obstacles within this zone while following the group's movement trend, ultimately obtaining a trend-guided direction. This trend-guided direction clearly indicates the robot's direction of travel, avoiding dynamic obstacles in the mid-field interaction zone and following the group's movement trend, ensuring the robot can smoothly navigate around dynamic obstacles. For the far-field guidance area, a third swarm intelligence with global topology exploration characteristics is adopted. It combines the environmental state information of the far-field guidance area to search for obstacles. The purpose is to avoid static obstacles in the area while taking into account the global operation goal, obtain a global reference path, and clearly instruct the robot to deviate from the path of all static obstacles in the far-field guidance area, so as to provide global directional guidance for the robot, as shown in steps 201 to 213.

[0025] Among them, the environmental status information includes the location distribution information and motion status information of obstacles. Specifically, the obstacle information includes the location distribution information of static obstacles (i.e., the specific location coordinates, size and shape of static obstacles in the spatial coordinate system) and the motion status information of dynamic obstacles (i.e., the current location coordinates, speed, direction and acceleration of dynamic obstacles in the spatial coordinate system).

[0026] Step 30: Asynchronous coordination and spatiotemporal coupling are performed based on the safe motion boundary, trend guidance direction and global reference path to obtain the optimal obstacle avoidance trajectory.

[0027] Optionally, asynchronous cooperative processing refers to the robot's time-sequential cooperative scheduling based on different response priorities of safe motion boundaries, trend guidance directions, and global reference paths. The safe motion boundaries in the near-field avoidance area have the highest priority, ensuring the robot first avoids near-field emergency obstacles. Next, the trend guidance direction in the mid-field interaction area ensures the robot smoothly navigates around dynamic obstacles. Finally, the global reference path in the far-field guidance area ensures the robot does not deviate from the global objective. Spatiotemporal coupling processing refers to the robot fusing the three obstacle avoidance results in both time and space dimensions. In the time dimension, it ensures that each time point of the obstacle avoidance trajectory simultaneously meets the obstacle avoidance requirements of the three areas. In the spatial dimension, it ensures that the spatial path of the obstacle avoidance trajectory smoothly connects the obstacle avoidance needs of the three areas, eliminating trajectory breakpoints and sharp turns to obtain the optimal obstacle avoidance trajectory, as detailed in steps 301 to 304.

[0028] Step 40: Based on the robot's motion constraints, the trajectory characteristics of the optimal obstacle avoidance trajectory are mapped to the linear velocity and angular velocity components of the robot chassis, and the robot is controlled to perform obstacle avoidance motion.

[0029] Optionally, motion constraints refer to the physical and performance limitations imposed on the robot during its movement. These mainly include maximum linear velocity, minimum linear velocity, maximum angular velocity, minimum angular velocity, linear acceleration limits, and angular acceleration limits. The maximum and minimum linear velocities limit the robot's travel speed range, preventing excessive speed from causing untimely braking or excessive speed from affecting work efficiency. The maximum and minimum angular velocities limit the robot's turning speed range, preventing excessively fast turning from causing vehicle vibration or excessively slow turning from failing to avoid obstacles in time. Linear acceleration limits and angular acceleration limits the rate of change of the robot's speed and turning, avoiding the impact of abrupt acceleration, deceleration, or turning on the robot's operational stability.

[0030] The robot analyzes the trajectory features of the optimal obstacle avoidance trajectory. The trajectory features include the spatial path coordinates of the trajectory, the curvature change of the trajectory, the travel direction change of the trajectory, and the time node corresponding to the trajectory. Among them, the spatial path coordinates of the trajectory are used to determine the target position of the robot at different time nodes, the curvature change of the trajectory is used to determine the turning requirements of the robot, the travel direction change of the trajectory is used to determine the heading adjustment requirements of the robot, and the time node corresponding to the trajectory is used to determine the target speed of the robot at different positions.

[0031] Based on its own motion constraints, the robot maps its trajectory features to linear and angular velocity components of the robot chassis. During this mapping process, it's crucial to ensure that both linear and angular velocity components remain within the robot's motion constraints while simultaneously meeting the trajectory's travel requirements: the linear velocity component corresponds to the robot's speed along the trajectory's direction of travel and is dynamically adjusted based on changes in trajectory curvature and time points. When the trajectory curvature is small, the linear velocity component can be appropriately increased; when the trajectory curvature is large (i.e., when turning is required), the linear velocity component is appropriately decreased to ensure smooth turning. The angular velocity component corresponds to the robot's turning speed and is dynamically adjusted based on changes in trajectory curvature and travel direction. The greater the trajectory curvature, the larger the angular velocity component, ensuring the robot can promptly keep up with trajectory turning demands. The more drastic the change in travel direction, the smoother the adjustment rate of the angular velocity component, avoiding vehicle vibration.

[0032] The robot converts the two components into motor control commands and sends them to the drive motors on the robot chassis. The drive motors execute the corresponding rotations according to the control commands, controlling the robot to smoothly perform obstacle avoidance movements along the optimal obstacle avoidance trajectory. During the obstacle avoidance movement, the robot collects its own pose and environmental status information in real time and dynamically fine-tunes the online velocity and angular velocity components to ensure the accuracy and safety of the obstacle avoidance movement.

[0033] In one embodiment, the motion constraints are: maximum linear velocity 1.5 m / s, minimum linear velocity 0.2 m / s, maximum angular velocity 1.2 radians / s, minimum angular velocity 0.1 radians / s, and linear acceleration limited to ±0.5 m / s. 2 Angular acceleration is limited to ±0.3 radians / second. 2 The robot chassis adopts a dual-wheel drive mode, where linear velocity and angular velocity components directly control the speed difference and speed of the two drive motors.

[0034] The robot obtains the optimal obstacle avoidance trajectory in step 30. The trajectory characteristics are as follows: the starting position is the robot's current position (X=10.2 m, Y=5.3 m, Z=0.5 m), and the ending position is a safe position within the mid-field interaction area (X=12.5 m, Y=6 m, Z=0.5 m). The spatial path coordinates of the trajectory are (10.2, 5.3, 0.5), (10.8, 5.5, 0.5), (11.5, 5.7, 0.5), (12.5, 5.9, 0.5), (12.5, 6.0, 0.5). 5), the trajectory curvature changes as follows: the first two path segments have a curvature of 0 (straight-line travel), the third path segment has a curvature of 0.8 radians / meter (slow turning), and the fourth path segment has a curvature of 0.5 radians / meter (slowing down the turning). The corresponding time nodes of the trajectory are: from the starting position to the first intermediate position (10.8, 5.5, 0.5) takes 0.4 seconds, to the second intermediate position (11.5, 5.7, 0.5) takes 0.5 seconds, to the third intermediate position (12, 5.9, 0.5) takes 0.6 seconds, and to the ending position takes 0.5 seconds.

[0035] Trajectory feature mapping is performed based on motion constraints: For the first two straight paths, the linear velocity component is set to 1.2 m / s (within the range of maximum and minimum linear velocity), and the angular velocity component is set to 0.1 radians / s (minimum angular velocity to ensure vehicle stability). For the third path of gradual turning, the linear velocity component is adjusted to 0.8 m / s (reducing speed to match the turning), and the angular velocity component is adjusted to 0.9 radians / s (within the range of maximum angular velocity to match the trajectory curvature). For the fourth path of slowing down the turn, the linear velocity component is adjusted to 1 m / s (gradually increasing speed), and the angular velocity component is adjusted to 0.6 radians / s (to match the slowing trajectory curvature).

[0036] The robot control module converts linear velocity and angular velocity components into motor control commands, which are then sent to the dual drive motors on the chassis. These motors operate at corresponding speeds: in the first two straight sections, the two drive motors rotate at the same speed, ensuring the robot moves in a straight line. In the third and fourth turning sections, the two drive motors create a speed difference, which is dynamically adjusted based on the angular velocity component, ensuring the robot turns along the trajectory.

[0037] During obstacle avoidance, the robot collects real-time environmental information using LiDAR and its own pose using a positioning module. It detects a worker in the central interaction area whose movement direction has changed, with a minimum distance of 0.4 meters from the robot's trajectory. The robot immediately adjusts its linear velocity component to 0.6 m / s and its angular velocity component to 1.0 radians / s, fine-tuning its obstacle avoidance trajectory to ensure it stays away from the worker. After avoiding the worker, the robot restores its linear and angular velocity components to their original mapping values ​​and continues obstacle avoidance, eventually reaching its final position smoothly and completing the obstacle avoidance operation.

[0038] This invention utilizes the different characteristics of multi-level swarm intelligence to search for obstacles in different levels of regions. By combining the asynchronous collaboration and spatiotemporal coupling of multi-level swarm intelligence, it achieves dynamic autonomous obstacle avoidance for robots, thereby improving the stability of autonomous obstacle avoidance in complex dynamic environments.

[0039] Optionally, the processes of steps 201 to 204 include: Step 201: Construct a spherical detection space based on the current pose and the preset safe detection radius, and perform spatial discretization mesh mapping based on the spherical detection space and the local repulsion characteristics of the first swarm intelligence to obtain the initial discrete sampling points in the spherical detection space.

[0040] Optionally, the robot constructs a spherical detection space centered on its current position coordinates according to a preset safe detection radius. The preset safe detection radius refers to the detection range radius pre-set by the robot to cover all potential obstacles within the near-field obstacle avoidance area for near-field emergency obstacle avoidance. This radius setting must consider the robot's own size, movement speed, braking performance, and the detection accuracy of the environmental sensing equipment to ensure that the spherical detection space can completely cover the near-field obstacle avoidance area.

[0041] Based on the local repulsion characteristic of the first swarm intelligence, the robot performs spatial discretization mesh mapping on the spherical detection space. Spatial discretization mesh mapping refers to dividing the continuous spherical detection space into multiple discrete, non-overlapping mesh units according to a preset mesh size. Each mesh unit corresponds to a sampling location, i.e., a discrete sampling point. The local repulsion characteristic of the first swarm intelligence is specifically manifested in that, during mesh mapping, higher-density mesh units are preferentially divided in the region closer to the robot's current position (i.e., the inner region of the spherical detection space), while lower-density mesh units are divided in the region farther from the robot's current position (i.e., the outer region of the spherical detection space).

[0042] The grid size is determined by combining the minimum obstacle size with the detection accuracy of the environmental sensing device, ensuring that the size of each grid cell is less than or equal to the minimum obstacle size. This avoids the inability to accurately capture the position information of small obstacles due to an excessively large grid size. After grid mapping is completed, the robot extracts the center point of each grid cell as an initial discrete sampling point. All initial discrete sampling points together constitute sampling points within the spherical detection space, and each initial discrete sampling point corresponds to a specific spatial location within the spherical detection space.

[0043] In one embodiment, the robot's dimensions (0.8 meters in diameter), movement speed (1.5 meters per second), and lidar detection accuracy (±5 millimeters) are set, with a preset safe detection radius of 0.8 meters, consistent with the radius of the near-field avoidance zone. The robot constructs a spherical detection space with a center point of (10.2 meters, 5.3 meters, 0.5 meters) and a radius of 0.8 meters. This space encompasses all locations within 0.8 meters of the center point. Based on the local repulsion characteristic of the first swarm intelligence, the spherical detection space is discretized into a grid: the inner layer (0-0.4 meters from the center point) uses a grid size of 0.05 meters, and the outer layer (0.4-0.8 meters from the center point) uses a grid size of 0.1 meters.

[0044] The robot extracts the center point of each grid cell as the initial discrete sampling point. A total of 64 initial discrete sampling points are obtained in the inner region and 192 initial discrete sampling points are obtained in the outer region, for a total of 256 initial discrete sampling points. All initial discrete sampling points are evenly distributed in the spherical detection space.

[0045] Step 202: Based on the initial discrete sampling points and the first obstacle position distribution information, determine the static obstacle space occupancy to obtain the first discrete sampling points located inside the static obstacle entity.

[0046] Optionally, the first obstacle location distribution information is the location distribution information of all static obstacles within the near-field avoidance area, including the location coordinates, size, and shape parameters of each static obstacle in the spatial coordinate system. The shape parameters include the length, width, and height (in three-dimensional space) or diameter (for circular obstacles) of the obstacle. The robot compares the spatial location coordinates of each initial discrete sampling point with the physical extent of each static obstacle to determine whether the initial discrete sampling point is located inside the physical extent of the static obstacle.

[0047] Optionally, the determination process in this embodiment of the invention adopts a spatial range comparison method. That is, for each static obstacle, its entity boundary range in the spatial coordinate system is determined based on its position coordinates and shape parameters. If the spatial position coordinates of an initial discrete sampling point are within this boundary range, it is determined that the sampling point is occupied by the static obstacle, that is, it is a sampling point located inside the entity of the static obstacle. If the spatial position coordinates of the initial discrete sampling point are outside the entity boundary ranges of all static obstacles, it is determined that the sampling point is not occupied by the static obstacle. During the determination process, for initial discrete sampling points located in the overlapping area of ​​the entity boundary ranges of multiple static obstacles, it is only necessary to determine that they are occupied by the static obstacle. All initial discrete sampling points determined to be occupied by the static obstacle together constitute the first discrete sampling point, which clearly indicates the specific location occupied by the static obstacle within the spherical detection space.

[0048] Continuing with the above embodiment, the obtained first obstacle location distribution information of the near-field avoidance area is as follows: There is a static obstacle (shelf corner) in the near-field avoidance area, with location coordinates of (10.5 meters, 5.6 meters, 0.5 meters), a cube shape, and size parameters of length 0.3 meters, width 0.3 meters, and height 1 meter. Its physical boundary range is X-axis 10.35-10.65 meters, Y-axis 5.45-5.75 meters, and Z-axis 0.0-1 meter.

[0049] The spatial coordinates of 256 initial discrete sampling points were extracted one by one and compared with the physical boundary range of the corner of the shelf. It was found that the coordinates of 8 initial discrete sampling points were within the range of 10.35-10.65 meters on the X-axis, 5.45-5.75 meters on the Y-axis, and 0.0-1 meters on the Z-axis. They were (10.4 meters, 5.5 meters, 0.5 meters), (10.45 meters, 5.55 meters, 0.5 meters), (10.5 meters, 5.6 meters, 0.5 meters), (10.55 meters, 5.65 meters, 0.5 meters), (10.6 meters, 5.7 meters, 0.5 meters), (10.4 meters, 5.6 meters, 0.5 meters), (10.5 meters, 5.5 meters, 0.5 meters), and (10.5 meters, 5.7 meters, 0.5 meters). These 8 sampling points are the first discrete sampling points located inside the static obstacle.

[0050] Step 203: Perform complementary analysis based on the first discrete sampling point and the initial discrete sampling point to obtain the second discrete sampling point inside the entity that has not entered the static obstacle.

[0051] Optionally, the robot removes sampling points belonging to the first discrete sampling point from the initial discrete sampling points one by one. The remaining sampling points are those that have not entered the interior of the static obstacle, i.e., the second discrete sampling points. Each sampling point in the second discrete sampling point corresponds to a spatial position within the spherical detection space that is not occupied by the static obstacle. These positions are potentially passable locations for the robot during near-field obstacle avoidance.

[0052] Continuing with the above embodiment, from the initial 256 discrete sampling points, the 8 first discrete sampling points are removed one by one, leaving 248 sampling points, which are the second discrete sampling points that have not entered the interior of the static obstacle. These 248 second discrete sampling points are evenly distributed within the spherical detection space and are all located outside the physical boundary of the shelf corner. The spatial position corresponding to each sampling point is a potentially passable location during the robot's near-field obstacle avoidance process.

[0053] Step 204: Based on the second discrete sampling point, the first obstacle position distribution information and the first motion state information, perform spatiotemporal trajectory intersection prediction to obtain the safe motion boundary.

[0054] Optionally, the robot performs spatiotemporal trajectory intersection prediction based on the second discrete sampling point, the first obstacle position distribution information, and the first motion state information to obtain the safe motion boundary, as specifically in steps 2041 to 2044.

[0055] The embodiments of the present invention achieve comprehensive and accurate identification and avoidance of static and dynamic obstacles within the near-field obstacle avoidance area. By utilizing the local repulsion characteristics of the first group intelligence, the high real-time performance of near-field obstacle avoidance is ensured, effectively avoiding the collision risk caused by inaccurate obstacle identification and unclear obstacle avoidance boundaries in the near-field area, and improving the stability of the robot's autonomous obstacle avoidance in complex dynamic environments.

[0056] Optionally, the process of steps 2041 to 2044 includes: Step 2041: Based on the second discrete sampling point, the first obstacle position distribution information and the first motion state information, perform spatiotemporal trajectory intersection prediction to obtain conflicting discrete sampling points that have spatial intersection with the motion trajectory of the dynamic obstacle within a future preset time window.

[0057] Optionally, the preset time window refers to a short prediction period set in advance by the robot based on its own near-field braking duration and the normal movement speed of dynamic obstacles. This period is used to predict potential collisions between the robot and dynamic obstacles within that time period. The duration of this time window is adapted to the real-time requirements of near-field emergency obstacle avoidance.

[0058] The robot uses its current pose as the starting point for motion and calculates the target motion trajectory corresponding to each second discrete sampling point. This trajectory is the spatial path through which the robot travels in a straight line from its current position to the corresponding second discrete sampling point. Combining the first motion state information, the robot infers the real-time motion trajectory of each dynamic obstacle within a preset time window in the future, thus clarifying the spatial movement range of the dynamic obstacle within that time period.

[0059] Subsequently, spatial intersection determination is performed on each robot target motion trajectory and dynamic obstacle motion trajectory. If the robot target motion trajectory corresponding to a certain second discrete sampling point has a spatial overlap area with the motion trajectory of any dynamic obstacle within a future preset time window, the sampling point is determined to be a conflict discrete sampling point. All such sampling points together form a conflict discrete sampling point set.

[0060] In one embodiment, the preset time window is set to 0.3 seconds to accommodate the robot's braking distance of 0.3 meters and movement speed of 1.5 meters per second. The preliminary step acquires 248 second discrete sampling points. Within the near-field avoidance area, there is one dynamic obstacle (the worker). The first motion state information shows the worker's current position coordinates as (10.1 meters, 5.2 meters, 0.5 meters), with movement along the positive Y-axis and a speed of 1 meter per second. The robot then calculates the robot's trajectory corresponding to each of the 248 second discrete sampling points and performs spatial intersection determination with the worker's trajectory within 0.3 seconds. Finally, 12 conflicting discrete sampling points with overlapping trajectories are selected; these 12 points are the sampling points with a risk of dynamic collision in the near field.

[0061] Step 2042: Perform differential analysis based on the conflict discrete sampling points and the second discrete sampling points to obtain target discrete sampling points that do not overlap with the trajectories of static obstacles and do not intersect with the trajectories of dynamic obstacles.

[0062] Optionally, the robot uses the second set of discrete sampling points as the basic complete set, compares and removes all points belonging to the conflicting discrete sampling point set, and the remaining sampling points are the target discrete sampling points. These target discrete sampling points simultaneously meet two safety conditions: first, they have not entered the interior of a static obstacle entity; second, within a preset future time window, the corresponding robot trajectory and the dynamic obstacle trajectory have no spatial intersection, making them candidate points where the robot can safely pass within the near-field avoidance area.

[0063] Continuing with the above embodiment, the total number of second discrete sampling points is 248. Step 2041 yields 12 conflicting discrete sampling points. The robot completely removes the 12 conflicting discrete sampling points from the 248 second discrete sampling points, leaving 236 sampling points as the target discrete sampling points. These 236 points all avoid static obstacles and will not intersect with the trajectory of dynamic workers within the next 0.3 seconds, thus belonging to near-field safe passage points.

[0064] Step 2043: Based on the target discrete sampling points and the local repulsion characteristics of the first swarm intelligence, a boundary expansion evolution is performed to obtain a discrete sampling point cloud that extends outward under local repulsion constraints until it contacts the environmental constraint boundary. During the boundary expansion evolution process, the point sets of discrete sampling points maintain a preset maximum spacing.

[0065] Optionally, the robot uses the target discrete sampling points as the base point set and combines the local repulsion characteristic of the first swarm intelligence to perform boundary expansion evolution. The local repulsion characteristic is manifested here as follows: when the sampling points expand outward, they always move away from the boundaries of static obstacle entities and the trajectories of dynamic obstacles, preferentially extending towards unobstructed open space. The boundary expansion evolution rule is to extend uniformly outward from the target discrete sampling points until reaching the environmental constraint boundary of the near-field avoidance zone, which is the spatial boundary between the near-field avoidance zone and the mid-field interaction zone. During the evolution process, the point sets of discrete sampling points maintain a preset maximum spacing. The preset maximum spacing refers to the maximum interval distance between adjacent sampling points set to ensure the smoothness and continuity of subsequent boundary construction. This spacing is set in conjunction with the robot's own size and environmental perception accuracy. All extended sampling points and the target discrete sampling points together form a discrete sampling point cloud.

[0066] In one embodiment, the maximum spacing is preset to 0.1 meters, which is suitable for the robot's 0.8-meter diameter and the detection accuracy of the lidar. Based on 236 discrete sampling points of the target, the sampling point cloud is expanded to the surrounding area according to the local repulsion characteristics, extending outward until it reaches the environmental constraint boundary with a radius of 0.8 meters of the near-field avoidance area. During the expansion process, the spacing between adjacent sampling points is controlled to not exceed 0.1 meters, and finally a discrete sampling point cloud containing 412 sampling points is generated.

[0067] Step 2044: Construct a convex hull and contour closure connection based on the discrete point set associated with the discrete sampled point cloud to obtain the target boundary line containing each discrete sampled point cloud, and perform relative position reference mapping based on the target boundary line and the robot's pose coordinate system to obtain the safe motion boundary.

[0068] Optionally, the robot uses a convex hull construction algorithm to close the contour of the discrete sampling point cloud. Specifically, the convex hull construction algorithm connects the outermost discrete points of the discrete sampling point cloud in sequence to form a closed convex polygon contour without breaks. The closed convex polygon contour is the target boundary line that contains all discrete sampling point clouds, ensuring that the interior space of the contour completely covers all safe passage points.

[0069] Subsequently, a relative position reference mapping is performed between the target boundary line and the robot's own pose coordinate system. The pose coordinate system is a local coordinate system established with the robot's current position as the origin and the direction of travel as the positive direction. The boundary obtained after mapping is the safe movement boundary. The safe movement boundary is a dynamic spatial constraint relative to the robot itself, explicitly instructing the robot that all movements within the near-field avoidance area must remain within the boundary and must not exceed the boundary range, achieving near-field emergency obstacle avoidance that completely deviates from both static and dynamic obstacles.

[0070] In one embodiment, the robot performs convex hull contour construction on a discrete sampling point cloud composed of 412 sampling points, extracts the outermost sampling points and connects them sequentially to form a closed circular approximate convex polygon target boundary line, completely enclosing all safe sampling points. Subsequently, this boundary line is mapped to a pose coordinate system with the robot's current position as the origin, ultimately generating a closed safe motion boundary within a radius of 0.8 meters. The closed safe motion boundary defines the robot's navigable space in the near field, with obstacles or unsafe areas outside the boundary. The robot can completely avoid near-field static and dynamic obstacles when moving within the boundary.

[0071] This invention achieves dual avoidance of both static and dynamic obstacles within the near-field obstacle avoidance area, transforming obstacle avoidance requirements into concrete spatial motion constraint boundaries. Relying on the local repulsion characteristics of the first group intelligence, it ensures high real-time performance of near-field obstacle avoidance, while making the safety boundary conform to the robot's motion characteristics. This not only avoids near-field collision risks but also provides near-field safety constraints for the subsequent coordinated coupling of global and local obstacle avoidance. It effectively solves the problems of slow response and fuzzy boundary oscillation in near-field emergency obstacle avoidance, and achieves stability of near-field autonomous obstacle avoidance in complex dynamic environments.

[0072] Optionally, steps 205 to 209 include: Step 205: Based on the second motion state information and the current pose, perform motion manifold mapping to obtain the discrete velocity point cloud of the endpoints of the relative velocity vectors of dynamic obstacles relative to the robot within the midfield interaction area.

[0073] Optionally, the second motion state information refers to the motion state information of all dynamic obstacles within the field interaction area, specifically including the current position coordinates, motion speed, motion direction, and motion acceleration of each dynamic obstacle. The robot constructs a relative coordinate system with its current pose as the reference origin, and calculates the relative velocity vector of each dynamic obstacle relative to the robot. The relative velocity vector is a comprehensive representation of the magnitude and direction of the dynamic obstacle's motion speed relative to the robot. The calculation process is as follows: subtract the robot's current motion speed from the absolute motion speed of each dynamic obstacle to obtain the relative motion speed of that dynamic obstacle relative to the robot; combine this with the motion direction to form the relative velocity vector.

[0074] Subsequently, the robot performs motion manifold mapping on the endpoints of each relative velocity vector. Motion manifold mapping refers to mapping the endpoints of continuous relative velocity vectors to a discrete velocity space according to a preset discretization interval, forming discrete velocity points. All these discrete velocity points together constitute a discrete velocity point cloud. The setting of the discretization interval needs to be combined with the motion velocity range of the dynamic obstacle and the robot's environmental perception accuracy to ensure that the discrete velocity points can accurately represent the relative velocity state of each dynamic obstacle.

[0075] In one embodiment, the current movement speed is 1.2 m / s, and the heading angle is 0° (moving along the positive X-axis). There are three dynamic obstacles within the mid-field interaction area, all representing workers. The second movement status information is as follows: The first worker's current position is (11 m, 5.5 m, 0.5 m), movement speed is 1 m / s, and movement direction is along the positive X-axis. The second worker's current position is (10.8 m, 5.8 m, 0.5 m), movement speed is 0.8 m / s, and movement direction is along the positive X-axis. The third worker's current position is (11.2 m, 5.2 m, 0.5 m), movement speed is 1.1 m / s, and movement direction is along the positive X-axis.

[0076] The robot constructs a relative coordinate system with its current position as the reference origin, and calculates the relative velocity vector of each worker relative to itself: the first worker's relative velocity vector is 0.2 m / s (along the negative X-axis), the second is 0.4 m / s (along the negative X-axis), and the third is 0.1 m / s (along the negative X-axis). With a discretization interval of 0.05 m / s, motion manifold mapping is performed on the endpoints of the three relative velocity vectors to obtain three discrete velocity points, which together form a discrete velocity point cloud. Each discrete velocity point corresponds to the relative velocity state of one worker.

[0077] Step 206: Based on the dynamic following characteristics of the discrete velocity point cloud and the second swarm intelligence, perform velocity spatial neighborhood connectivity analysis to obtain initial velocity connectivity domains. In each initial velocity connectivity domain, the velocity difference between dynamic obstacles is less than or equal to a preset difference threshold, and the spatial distance is less than or equal to a preset distance threshold.

[0078] Optionally, the dynamic following characteristic of the second group intelligence is specifically manifested in its ability to identify the commonalities in the movement of a group of dynamic obstacles and group dynamic obstacles with similar movement states into a group.

[0079] The robot performs velocity space neighborhood connectivity analysis. The analysis process is as follows: Each discrete velocity point in the discrete velocity point cloud is selected as a reference point. Within the neighborhood of this reference point in velocity space, all discrete velocity points satisfy two conditions: First, the velocity difference between adjacent discrete velocity points and the corresponding dynamic obstacles is less than or equal to a preset difference threshold. The preset difference threshold is an upper limit set in advance to determine whether the motion velocities of dynamic obstacles are similar, ensuring that the velocity differences of dynamic obstacles within the same connected region are small. Second, the spatial distance between adjacent discrete velocity points and the corresponding dynamic obstacles is less than or equal to a preset distance threshold. The preset distance threshold is an upper limit set in advance to determine whether the spatial positions of dynamic obstacles are adjacent, ensuring that dynamic obstacles within the same connected region are within the same local spatial range. Discrete velocity points satisfying the above two conditions are grouped with the reference point into a connected region. This process is repeated until all discrete velocity points are assigned to corresponding connected regions. Each connected region is an initial velocity connectivity region.

[0080] Therefore, in each initial velocity connectivity domain of the present invention, the dynamic obstacles have similar motion speeds and are spatially adjacent, and can represent a relatively independent group of dynamic obstacles.

[0081] Continuing with step 205, three discrete velocity points are obtained, corresponding to three workers. A preset difference threshold of 0.2 m / s and a preset distance threshold of 0.5 m are set. The robot performs velocity spatial neighborhood connectivity analysis. Taking the discrete velocity point corresponding to the first worker as the reference point, within its neighborhood, the velocity difference of the second worker is 0.2 m / s (equal to the preset difference threshold), and the spatial distance is 0.38 m (less than the preset distance threshold). The velocity difference of the third worker is 0.1 m / s (less than the preset difference threshold), and the spatial distance is 0.42 m (less than the preset distance threshold). Therefore, the three discrete velocity points are grouped into an initial velocity connectivity domain. This initial velocity connectivity domain includes the three workers, whose velocity differences all do not exceed 0.2 m / s and whose spatial distances all do not exceed 0.5 m, forming a dynamic obstacle group.

[0082] Step 207: Based on the initial velocity connected domain and the second obstacle position distribution information, perform static environment visual topology to obtain the target velocity connected domain that is not blocked by static obstacles.

[0083] Optionally, the second obstacle location distribution information refers to the location distribution information of all static obstacles within the field interaction area, specifically including the location coordinates, size, and shape parameters of each static obstacle, used to define the physical extent of the static obstacles. For each initial velocity connected domain, the robot constructs a static environment visual topology. The static environment visual topology refers to analyzing the position of each dynamic obstacle within the initial velocity connected domain with the robot's current pose as the observation point, determining whether the dynamic obstacle is within the robot's visual range, i.e., whether there is a physical obstruction of a static obstacle between the robot and the dynamic obstacle.

[0084] Optionally, the determination process in this embodiment of the invention is as follows: Connect the robot's current position to the position of each dynamic obstacle within the initial velocity connectivity domain to form a line of sight. Check whether this line of sight intersects with the physical extent of any static obstacle. If there is no spatial intersection, it indicates that the dynamic obstacle is within the robot's visual range, and the initial velocity connectivity domain is an observable connectivity domain. If the line of sight intersects with the physical extent of a static obstacle, it indicates that the initial velocity connectivity domain is blocked by the static obstacle, and the robot cannot effectively observe and follow the group; therefore, it needs to be eliminated. Finally, all initial velocity connectivity domains not blocked by static obstacles are the target velocity connectivity domains.

[0085] Continuing with step 206, an initial velocity connectivity region (containing the three workers) is obtained. The second obstacle location distribution information shows that there is one static obstacle (small shelf) in the mid-field interaction area, with location coordinates of (11.5 meters, 5.5 meters, 0.5 meters) and dimensions of 0.4 meters in length, 0.4 meters in width, and 1 meter in height. The robot uses its current position as the observation point and connects its own position with the positions of the three workers, forming three lines of sight. Upon inspection, none of the three lines of sight intersect with the physical area of ​​the small shelf, meaning that the initial velocity connectivity region is not blocked by the static obstacle. Therefore, this initial velocity connectivity region is the target velocity connectivity region.

[0086] Step 208: Perform principal axis analysis on the connected domain of each target velocity to obtain the principal axis direction vector.

[0087] Optionally, the robot performs principal axis analysis for each target velocity domain. The specific process of principal axis analysis is as follows: extract the current position coordinates and relative velocity vectors of all dynamic obstacles within the target velocity domain; calculate the average relative velocity vector of all dynamic obstacles to obtain the average velocity vector of the group. Then, based on this average velocity vector and combined with the spatial distribution of all dynamic obstacles, fit a principal axis that runs through the target velocity domain and is consistent with the average motion direction of the group. The direction of this principal axis is the principal axis direction vector. The principal axis direction vector can accurately characterize the overall motion tendency of the group of dynamic obstacles within the target velocity domain, reflecting the common motion direction of the group.

[0088] Continuing with step 207, a target velocity connected region (containing 3 workers) is obtained. The robot extracts the relative velocity vectors of the 3 workers, which are 0.2 m / s (along the negative X-axis), 0.4 m / s (along the negative X-axis), and 0.1 m / s (along the negative X-axis), respectively. The average velocity vector is calculated to be 0.23 m / s (along the negative X-axis). Combining the spatial distribution of the 3 workers, a principal axis along the negative X-axis is fitted. The direction of this principal axis is the principal axis direction vector, representing the overall movement of the group of 3 workers along the negative X-axis.

[0089] Step 209: Determine the trend guidance direction based on each principal axis direction vector and the current pose.

[0090] Optionally, the robot determines the trend guidance direction based on the direction vector of each main axis and the current pose, as in steps 2091 to 2094.

[0091] This invention utilizes the dynamic following characteristics of the second swarm intelligence to determine the movement trend of a group of dynamic obstacles within the center interaction area. This allows the robot to avoid dynamic obstacles while following the movement trend of the group, adapting to the complex changes in the center dynamic environment and ensuring the robot's smooth movement within the center interaction area. This achieves stable autonomous obstacle avoidance for the robot in complex dynamic environments.

[0092] Optionally, the process of steps 2091 to 2094 includes: Step 2091: Construct the following frustum geometry based on each principal axis direction vector and the current pose to obtain the minimum frustum region with the current pose as the vertex and covering each principal axis direction vector.

[0093] Optionally, the principal axis direction vectors represent the motion tendency of the dynamic obstacle group within the connected domain of each target velocity. The robot constructs the following view cone geometry using its current pose coordinates as the vertex and its current heading angle corresponding to the travel direction as the reference center axis of the view cone. During the construction process, the robot incorporates each principal axis direction vector into the view cone coverage area one by one. By adjusting the cone angle, it ensures that all principal axis direction vectors are completely covered by the view cone region, while continuously reducing the cone angle size until it cannot be reduced further but still covers all principal axis direction vectors. The resulting view cone region is the minimum view cone region. The cone angle size of the minimum view cone region is determined by the distribution range of all principal axis direction vectors. The more concentrated the distribution range, the smaller the cone angle. The more dispersed the distribution range, the larger the cone angle. Its function is to lock the approximate directional range of the robot's movement following the dynamic obstacle group, avoiding excessive directional deviation.

[0094] In one embodiment, the robot's current pose coordinates are (10.2 m, 5.3 m, 0.5 m), and its current heading angle is 0° (traveling along the positive X-axis). A preliminary step yields one principal axis direction vector, oriented along the negative X-axis. Using its current position as the vertex and the positive X-axis as the center axis of the visual cone reference, the robot adjusts the visual cone angle. Since there is only one principal axis direction vector, the cone angle is adjusted to just cover this vector, ultimately obtaining the minimum visual cone region with a cone angle of 30°. This visual cone range covers an area 15° to the left and right of the negative X-axis direction, completely enclosing the principal axis direction vector.

[0095] Step 2092: Based on the boundary geometric constraints of the minimum view cone region and the mid-field interaction region, perform the intersection of feasible passable sectors to obtain a local following region located within the boundary of the mid-field interaction region that does not conflict with static obstacles.

[0096] Optionally, the boundary geometric constraints of the mid-field interaction area refer to the spatial boundary range of the mid-field interaction area, that is, the boundary between the mid-field interaction area and the near-field avoidance area and the far-field guidance area, used to define the spatial boundary of the robot's mid-field following. The second obstacle position distribution information refers to the position distribution information of all static obstacles within the mid-field interaction area, used to determine whether the passage area conflicts with static obstacles.

[0097] The robot projects the minimum view cone region onto the plane of the mid-field interaction region to form an initial passage sector, which is the intersection of the minimum view cone region and the plane of the mid-field interaction region.

[0098] Subsequently, by combining the boundary geometric constraints of the midfield interaction region, the portion of the initial passable sector that exceeds the boundary of the midfield interaction region is removed, resulting in a feasible passable sector located within the boundary of the midfield interaction region.

[0099] Next, combining the information on the location distribution of the second obstacle, it is checked whether there are any areas within the feasible passage sector that overlap with the physical range of the static obstacle. The overlapping parts are eliminated, and the final passage sector that is located within the boundary of the central interaction area and does not conflict with the static obstacle is the local following area. The local following area is the specific spatial range within which the robot can safely follow the movement of the dynamic obstacle group, ensuring that the robot's movement within this area will not exceed the central interaction area and will not collide with the static obstacle.

[0100] After projecting the minimum view cone region obtained in step 2091 onto the mid-field interaction area plane, an initial passage sector is formed (cone angle 30°, covering 15° to the left and right of the negative X-axis direction). The boundary range of the mid-field interaction area is a ring-shaped area with a radius of 0.8 meters to 3 meters centered on the robot's current position. The robot removes the portion of the initial passage sector that exceeds this ring-shaped area, obtaining a feasible passage sector located within the mid-field interaction area. Combining the second obstacle location distribution information (there is a small shelf in the mid-field interaction area, with location coordinates (11.5 meters, 5.5 meters, 0.5 meters)), it is found that the feasible passage sector does not overlap with the small shelf. Therefore, this feasible passage sector is the local following area, with a cone angle of 30°, covering 15° to the left and right of the negative X-axis direction, and completely located within the boundary of the mid-field interaction area.

[0101] Step 2093: Calculate the central angle of the sector based on the local following region and compare it with the maximum value to obtain the optimal guiding sector with the largest central angle value in the local following region.

[0102] Optionally, the robot calculates the sector central angle for each local following area. The sector central angle calculation process is as follows: using the robot's current pose coordinates as the sector center, extract the two boundary directions of the local following area, and calculate the angle between the two boundary directions. This angle is the sector central angle of the local following area. The size of the central angle directly reflects the spatial range of the local following area. The larger the central angle, the wider the passage space of the local following area, and the higher the flexibility of the robot's following movement. The smaller the central angle, the narrower the passage space, and the lower the following flexibility. The robot compares the maximum value of all central angle values, compares the size of the central angle of each local following area one by one, and selects the local following area with the largest central angle value. This local following area is the optimal guiding sector. If only one local following area exists, this local following area is directly used as the optimal guiding sector, providing the robot with the widest safe following space and ensuring that the robot avoids trajectory oscillation or collision due to narrow space when following a group of dynamic obstacles.

[0103] Continuing with step 2092, a local following region is obtained. Using its current position as the center, the robot extracts the two boundary directions of this local following region (15° in the negative X-axis direction and -15° in the negative X-axis direction), calculates the angle between the two boundary directions, and obtains a sector central angle of 30°. Since there is only one local following region, there is no need to compare maximum values; this local following region is the optimal guiding sector, with a central angle of 30°, representing the widest spatial range that the robot can safely follow.

[0104] Step 2094: Perform polar coordinate direction mapping analysis based on the angle bisector direction of the optimal guidance sector and the robot's current pose coordinate system to obtain the trend guidance direction.

[0105] Optionally, the current pose coordinate system is a local coordinate system established with the robot's current position as the origin and the direction of travel corresponding to the current heading angle as the positive direction. It is used to map the angle bisector direction of the sector to a travel direction that the robot can recognize. The angle bisector direction of the optimal guidance sector refers to the symmetrical direction between the two boundary directions of the optimal guidance sector. This direction can most accurately correspond to the movement tendency of the dynamic obstacle group and is located at the center of the optimal guidance sector.

[0106] The robot performs polar coordinate orientation mapping analysis. Specifically, the origin of the current pose coordinate system (the robot's current position) is used as the polar coordinate origin, and the positive direction of the current pose coordinate system is used as the polar axis. The angle bisector direction of the optimal guidance sector is mapped to this polar coordinate system, and the angle of the angle bisector direction relative to the polar axis is determined. The direction corresponding to this angle is the trend guidance direction. Therefore, the trend guidance direction obtained in this embodiment of the invention not only conforms to the movement trend of the dynamic obstacle group but is also located at the center of the optimal guidance sector, ensuring that the robot can avoid dynamic obstacles in the middle of the field.

[0107] Continuing from step 2093, the optimal guiding sector's central angle is 30°, with its two boundary directions being 15° and -15° in the negative X-axis direction, respectively. The angle bisector direction of this sector is calculated to be 0° in the negative X-axis direction (i.e., the positive X-axis negative direction). Using its current position as the origin of the polar coordinate system and the current heading angle of 0° (positive X-axis direction) as the polar axis, the robot maps the angle bisector direction (0° in the negative X-axis direction) to the polar coordinate system, determining that the angle of this direction relative to the polar axis is 180°. The direction corresponding to this angle is the trend guidance direction, instructing the robot to move along the negative X-axis direction, thus both following the movement trend of the dynamic obstacle group and having sufficient safe passage space.

[0108] The embodiments of the present invention achieve precise determination of trend guidance direction, which not only fully conforms to the movement trend of the dynamic obstacle group, but also avoids the limitations of static obstacles and area boundaries in the middle. Relying on the dynamic following characteristics of the second group intelligence, it ensures that the robot can smoothly bypass dynamic obstacles and follow the movement trend of the group in the middle interactive area. It solves the problem of trajectory oscillation or collision risks caused by unclear following direction and narrow passage space in the middle, and improves the stability and adaptability of the robot's autonomous obstacle avoidance in complex dynamic environments.

[0109] Optionally, the process of steps 210 to 213 includes: Step 210: Based on the current pose and the global target pose specified by the task, combined with the boundary of the obstacle-free space in the far-field guidance area, perform bidirectional initialization distribution analysis of the mass group to obtain the initial mass points that start from the current pose and the global target pose and are located in the obstacle-free space.

[0110] Optionally, the global target pose refers to the final position coordinates and attitude angle that the robot needs to reach to complete the current task. This is specified in advance by the task and is the endpoint of the global reference path. The boundary of the obstacle-free space within the far-field guidance area refers to the edge limit of the remaining passable space after removing all static obstacles within the far-field guidance area. This boundary defines the distribution range of the initial mass points, ensuring that all initial mass points are within the passable space. The robot performs bidirectional initialization distribution analysis of the mass point group. Bidirectional initialization refers to simultaneously initializing mass points from two starting points (the robot's current pose and the global target pose) to ensure that subsequent mass point evolution can achieve bidirectional docking, improving the efficiency of global path search. Specifically, using the robot's current pose position coordinates as the first initial starting point and the global target pose position coordinates as the second initial starting point, multiple mass points are uniformly initialized within the boundary range of the obstacle-free space within the far-field guidance area according to a preset mass point distribution density. These mass points are the initial mass points.

[0111] The preset particle distribution density is set based on the spatial range of the far-field guidance area and the distribution density of static obstacles. The larger the spatial range and the denser the obstacle distribution, the higher the particle distribution density, ensuring complete coverage of the obstacle-free space. The smaller the spatial range and the sparser the obstacle distribution, the lower the particle distribution density. All initial particles must meet the conditions of being located within the obstacle-free space of the far-field guidance area and not overlapping with any static obstacle entities. During initialization, the robot verifies the position of each particle in real time, discarding particles that do not meet the conditions, resulting in initial particle sets starting from the current pose and the global target pose, respectively.

[0112] In one embodiment, the global target pose coordinates specified by the task are (18.5, 6.1, 0.5). The far-field guidance area is a ring-shaped area with a radius of 3 to 10 meters centered on the robot's current position. The boundary of the obstacle-free space is the edge of the passable space within the ring-shaped area excluding static obstacles (fixed shelves). Static obstacles are distributed at the edge of the ring-shaped area, and the middle area is an obstacle-free open space.

[0113] The particle distribution density is 3 particles per square meter. Starting from the current pose (10.2 m, 5.3 m, 0.5 m) and the global target pose (18.5 m, 6.1 m, 0.5 m), the particles are initialized within the boundary of the obstacle-free space. A total of 86 initial particles are obtained, of which 42 are from the current pose and 44 are from the global target pose. All initial particles are located within the obstacle-free space and do not overlap with any static obstacles.

[0114] Step 211: Based on the initial mass point, the location distribution information of the third obstacle, and the global topology exploration characteristics of the third swarm intelligence, the obstacle boundary is repulsed and evolved to obtain the target mass point distribution cloud that reaches a preset mechanical equilibrium state in the central area of ​​the channel corresponding to the static obstacle.

[0115] Optionally, the third obstacle location distribution information refers to the location distribution information of all static obstacles within the far-field guidance area, specifically including the location coordinates, size, and shape parameters of each static obstacle, used to define the physical boundaries of static obstacles and the passage area. The global topology exploration characteristic of the third swarm intelligence is specifically manifested in its ability to guide a group of particles to move towards the center region of the passage without obstacles, while being repelled by the boundaries of static obstacles, preventing the particles from approaching the obstacles, and ultimately achieving the mechanical equilibrium of the particle group.

[0116] The robot undergoes a repulsive evolution at obstacle boundaries. The evolution process involves applying two forces to each initial particle: a repulsive force from the static obstacle boundary, the magnitude of which is inversely proportional to the distance from the particle to the boundary (the closer the distance, the stronger the repulsive force), ensuring the particle stays away from the static obstacle boundary; and an attractive force between the particle groups, the magnitude of which is inversely proportional to the distance between the particles, ensuring the particle group remains relatively clustered and preventing excessive dispersion.

[0117] Under the combined action of two forces, the initial particles gradually move towards the center of the passage between static obstacles, continuously adjusting their positions until all particles reach a preset mechanical equilibrium state. The preset mechanical equilibrium state means that the repulsive and attractive forces acting on the particles are equal in magnitude and opposite in direction, and the particles no longer move. At this point, all particles are located in the center of the passage, far from the boundaries of the static obstacles, while maintaining a reasonable distribution density. All particles that have reached mechanical equilibrium together form a target particle distribution point cloud, which can accurately represent the center position of the passable passage within the far-field guidance area.

[0118] Continuing with step 210, 86 initial mass points are obtained. The third obstacle location distribution information shows that there are four fixed shelves (static obstacles) in the far-field guidance area, distributed at positions (12, 5.3, 0.5), (15.0, 4.8, 0.5), (15.0, 6.8, 0.5), and (17, 6.1, 0.5), respectively. Two 1.2-meter-wide passages are formed between the four shelves, which is the central area of ​​the passages. The robot applies repulsive and attractive forces to the 86 initial mass points. The repulsive force moves the mass points away from the boundaries of the four fixed shelves, while the attractive force keeps the mass points clustered together. After evolution and adjustment, all mass points move to the central area of ​​the two passages, reaching a preset mechanical equilibrium state. Finally, a target mass point distribution point cloud containing 86 mass points is formed. The mass points are evenly distributed in the center of the two passages, and the distance from the boundary of each shelf is not less than 0.6 meters to ensure safe passage.

[0119] Step 212: Perform local density analysis based on the target particle distribution point cloud to obtain local density peak points, and perform maximum unobstructed coverage analysis based on each local density peak point and the location distribution information of the third obstacle to obtain the maximum safe passage radius of each local density peak point.

[0120] Optionally, local density analysis refers to calculating the number of particles within a preset neighborhood for each particle. This number is the local density of that particle. The preset neighborhood refers to a circular area centered on each particle with a preset radius. The preset neighborhood radius is set in conjunction with the particle distribution density to ensure accurate reflection of the particle aggregation degree. After calculating the local density of all particles, the robot selects the particle with the highest local density value, which is the local density peak point. The local density peak point is the location where the particle distribution is most dense, accurately representing the core location of the central area of ​​the channel. If multiple particles have the same local density value, all of which are the maximum value, then all such particles are considered local density peak points.

[0121] Subsequently, based on the distribution information of each local density peak point and the location of the third obstacle, the robot performs a maximum obstacle-free coverage analysis. Specifically, it gradually expands the circular coverage area centered on each local density peak point, while simultaneously checking in real time whether the expanded circular area overlaps with the physical area of ​​any static obstacle. When the circular area is about to contact the physical area of ​​a static obstacle without overlap, the expansion stops. The radius of this circular area is the maximum safe passage radius for that local density peak point. The maximum safe passage radius clearly defines the maximum safe passage range around each peak point, ensuring that the robot will not collide with static obstacles within this range.

[0122] Continuing from step 211, the target particle distribution point cloud contains 86 particles. With a preset neighborhood radius of 0.3 meters, the local density of each particle is calculated. It is found that 6 particles have the highest local density (each particle has 5 other particles within its neighborhood). These 6 particles are the local density peak points, evenly distributed in the central area of ​​the two channels. The robot performs maximum obstacle-free coverage analysis centered on each local density peak point, gradually expanding the circular coverage area. Upon inspection, when the circular radius expands to 0.6 meters, the circular area is about to contact the boundary of the fixed shelf without overlap. Therefore, the maximum safe passage radius for each local density peak point is determined to be 0.6 meters. This radius ensures that the robot will not collide with any static obstacles when moving within a 0.6-meter radius around the peak point.

[0123] Step 213: Based on the maximum safe passage radius of each local density peak point, determine the effective connectivity of adjacent points to obtain the global reference path.

[0124] Optionally, the robot determines the connectivity of adjacent points based on the maximum safe passage radius of each local density peak point to obtain a global reference path, as described in steps 2131 to 2134.

[0125] This invention, through the global topology exploration characteristics of third-party collective intelligence, achieves the accurate construction of a global reference path within the far-field guidance area. This ensures that the path avoids all static obstacles while also taking into account the global optimality of the path. It solves the problems of far-field path planning easily deviating from the optimal direction and easily conflicting with static obstacles. It provides far-field path guidance for the asynchronous coordination and spatiotemporal coupling of global obstacle avoidance and local obstacle avoidance, and improves the global adaptability and path optimality of the robot's autonomous obstacle avoidance in complex dynamic environments.

[0126] Optionally, the process of steps 2131 to 2134 includes: Step 2131: Based on the maximum safe passage radius of each local density peak point, determine the effective connectivity of adjacent points to obtain target adjacent node pairs with overlapping areas.

[0127] Optionally, the local density peak point is the particle with the highest local density in the target particle distribution point cloud, representing the core location of the passable passage within the far-field guidance area. The maximum safe passage radius is the largest circular radius around each local density peak point where safe passage is possible without conflict with static obstacles.

[0128] The robot determines the connectivity of adjacent points as follows: Each local density peak point is selected as a reference node, and this reference node is paired with all other local density peak points to form node pairs. For each node pair, two circular safety regions are constructed with the two nodes as centers and their respective maximum safe passage radii as radii. The robot then determines whether these two circular safety regions overlap. If the two circular safety regions overlap, and the straight line connecting the two nodes does not overlap with the physical extent of any static obstacle within the far-field guidance area, the node pair is considered to be effectively connected, i.e., the target adjacent node pair. If the two circular safety regions do not overlap, or the straight line connecting them overlaps with the physical extent of a static obstacle, the node pair is considered ineffectively connected and is discarded.

[0129] Continuing with the above implementation, six local density peak points were obtained, denoted as nodes 1 to 6. The maximum safe passage radius for each node was 0.6 meters. The static obstacles in the far-field guidance area consisted of four fixed shelves distributed on both sides of the aisle. The robot paired each node with other nodes one by one, forming a total of 15 node pairs. For each node pair, a connectivity validity determination was performed: taking node 1 and node 2 as an example, a circular safe area was constructed with the two nodes as the center and a radius of 0.6 meters. The two areas overlapped by 0.2 meters, and the straight line connecting the two nodes was located at the center of the aisle, not overlapping with any fixed shelves; this was determined to be a target adjacent node pair. Taking node 1 and node 4 as an example, the two circular safe areas did not overlap, and the connection was deemed invalid. Finally, eight target adjacent node pairs were selected, all of which met the safe connectivity conditions.

[0130] Step 2132: Construct an undirected connected graph based on the target neighboring node pairs to obtain a far-field region topology connection graph with local density peak points as nodes and target neighboring node pairs as edges.

[0131] Optionally, an undirected connected graph refers to a geometric figure composed of nodes and undirected edges. Undirected edges simply indicate a connection between two nodes, without distinguishing direction, thus meeting the requirement for bidirectional passage within the far-field guidance area. The construction process is as follows: the robot uses each local density peak point as a node in the undirected connected graph, and each pair of target adjacent nodes as an edge. Each edge connects to the two nodes it contains, forming the undirected connected graph, which is the topological connection graph of the far-field region. Each node in the undirected connected graph corresponds to a local density peak point, and each edge corresponds to a connected pair of target adjacent nodes, presenting the connectivity between all nodes and the topological structure of passable channels within the far-field guidance area.

[0132] Continuing with step 2131, we obtain 8 target neighbor node pairs and 6 local density peak points (nodes 1 to 6). The robot uses these 6 nodes as nodes in an undirected connected graph and the 8 target neighbor node pairs as undirected edges to connect the corresponding nodes: node 1 connects to node 2, node 1 connects to node 3, node 2 connects to node 3, node 2 connects to node 4, node 3 connects to node 5, node 4 connects to node 6, and node 5 connects to node 6, thus constructing a far-field region topology connection graph. This graph clearly shows the connectivity between the 6 core nodes, corresponding to the topology of the two far-field channels.

[0133] Step 2133: Based on the far-field region topology connection graph, the starting node corresponding to the current pose and the ending node corresponding to the global target pose, perform a path search with the fewest hops to obtain the target waypoint sequence connecting the starting node and the ending node.

[0134] Optionally, the starting node refers to the local density peak point closest to the robot's current pose. The robot calculates the straight-line distance between each local density peak point and the current pose, and selects the closest node as the starting node. The ending node refers to the local density peak point closest to the global target pose. Similarly, by calculating the straight-line distance between each local density peak point and the global target pose, the closest node is selected as the ending node. The robot performs a minimum hop path search. The minimum hop path refers to the path from the starting node to the ending node with the fewest number of edges. The number of hops is the number of edges in the path. The fewer the hops, the simpler the path, which can improve the efficiency of the robot's far-field motion. The search process is as follows: starting from the starting node, check all nodes directly connected to the starting node (one-hop nodes), then check the nodes directly connected to the one-hop nodes that have not been checked (two-hop nodes), and so on, until the ending node is found. Record the sequence of nodes traversed from the starting node to the ending node. This sequence is the target waypoint sequence corresponding to the minimum hop count. If multiple paths with the same minimum number of hops exist, the robot selects the path with the most overlapping maximum safe passage radii among the nodes to ensure the path's safety.

[0135] Continuing with the above far-field region topology connection diagram, there are 6 nodes (nodes 1 to 6). The starting node is node 1, which is closest to the current pose (10.2 m, 5.3 m, 0.5 m), and the ending node is node 6, which is closest to the global target pose (18.5 m, 6.1 m, 0.5 m).

[0136] The robot performs a path search with the fewest hops. Starting from node 1, the first hop nodes are nodes 2 and 3. Starting from node 2, the second hop node is node 4. Starting from node 4, the third hop node is node 6. Starting from node 3, the second hop node is node 5. Starting from node 5, the third hop node is node 6. Therefore, the minimum number of hops is 3, and there are two paths with the fewest hops: Node 1 → Node 2 → Node 4 → Node 6 and Node 1 → Node 3 → Node 5 → Node 6. The robot selects the path with the most overlapping safe radii (Node 1 → Node 2 → Node 4 → Node 6), and this path is the target waypoint sequence.

[0137] Step 2134: Perform segmented straight-line geometric connection analysis based on the target waypoint sequence to obtain the global reference path that is a series of straight lines connecting the target waypoints.

[0138] Optionally, the target waypoint sequence is the sequence of nodes with the fewest hops from the starting node to the ending node, where each node is a local density peak point located at the core of the far-field passable channel. The robot performs segmented straight-line geometric connection analysis, specifically as follows: according to the order of the target waypoint sequence, adjacent waypoints are connected one by one with straight lines to form multiple straight-line segments. After each straight-line segment is connected, the robot verifies in real time whether the straight-line segment overlaps with the physical extent of any static obstacle within the far-field guidance area. If overlap exists, the connection method of the straight-line segment is adjusted to ensure that the straight-line segment is always located within the obstacle-free space and is at least a preset safety distance from the boundary of the static obstacle (the preset safety distance is at least 1.5 times the robot's own radius). If there is no overlap, the straight-line segment remains unchanged.

[0139] After adjacent waypoints are connected, a continuous path is formed by connecting multiple straight line segments in sequence, which is the global reference path. Each straight line segment of the global reference path is located in an obstacle-free space and passes through the core position of each local density peak point. It indicates the direction of travel for the robot in the far-field guidance area to avoid all static obstacles, ensuring that the robot will not collide with any static obstacles when moving along the path.

[0140] Continuing from step 2133, the target waypoint sequence is Node 1 → Node 2 → Node 4 → Node 6, with the coordinates of the four nodes being (10.8m, 5.3m, 0.5m), (12.5m, 5.3m, 0.5m), (15.5m, 5.8m, 0.5m), and (17.8m, 6.1m, 0.5m), respectively. The robot connects Node 1 with Node 2, Node 2 with Node 4, and Node 4 with Node 6 in sequence, forming three straight line segments. Verification shows that all three straight line segments are located in the center of the passageway, do not overlap with any fixed shelves (static obstacles), and are at least 0.6 meters from the shelf boundary (greater than 1.5 times the robot's own radius of 0.4 meters). Connecting these three straight line segments sequentially forms a continuous global reference path, instructing the robot to start from the starting node, proceed along the straight line through Node 2 and Node 4, and finally reach the ending node, avoiding all far-field static obstacles.

[0141] This invention enables the precise construction of a global reference path, fully leveraging the global topology exploration capabilities of third-party collective intelligence. It transforms nodes of far-field passable channels into explicit travel paths, ensuring that the path avoids all static obstacles within the far-field guidance area while also balancing global optimality and efficiency. This solves the problems of inaccurate far-field path planning, easy conflict with static obstacles, and path redundancy. It provides far-field path guidance for asynchronous coordination and spatiotemporal coupling of global and local obstacle avoidance, improving the global stability and path optimality of the robot's autonomous obstacle avoidance in complex dynamic environments.

[0142] Optionally, the processes of steps 301 to 303 include: Step 301: Using the safe motion boundary as the anti-collision spatial constraint condition when searching the mid-field interaction area, and the trend guidance direction as the obstacle avoidance direction condition when searching the far-field guidance area, the obstacle avoidance decision target is obtained.

[0143] Optionally, two core constraints and directional conditions are defined: First, the safe movement boundary is used as the anti-collision spatial constraint condition when searching the mid-field interaction area. That is, when updating the trend guidance direction and performing obstacle avoidance search in the mid-field interaction area, it must ensure that the travel range corresponding to the new trend guidance direction does not exceed the range of the safe movement boundary, avoiding conflicts between mid-field obstacle avoidance operations and near-field anti-collision constraints, and ensuring the continuity and safety of the robot's near-field and mid-field obstacle avoidance actions. Second, the trend guidance direction is used as the obstacle avoidance directional condition when searching the far-field guidance area. That is, when updating the global reference path in the far-field guidance area, it must ensure that the overall travel direction of the new global reference path is consistent with the trend guidance direction, avoiding excessive deviation between the far-field path and the mid-field following direction, and ensuring the coordination of the robot's mid-field and far-field obstacle avoidance actions.

[0144] The robot integrates the above two constraints and orientation conditions with the optimality requirement of the global reference path to form a unified obstacle avoidance decision objective. This obstacle avoidance decision objective clearly requires: ensuring safety and no collisions in the near field, smoothly following the dynamic group in the mid-field, and conforming to the global optimal path in the far field.

[0145] In one embodiment, the safe motion boundary obtained in the preliminary steps is a closed region with a radius of 0.8 meters centered on the current pose, the trend guidance direction is along the negative X-axis, and the global reference path is a continuous straight line path from node 1 to node 2 to node 4 to node 6. The robot uses the safe motion boundary as a collision avoidance constraint for searching the mid-field interaction area, ensuring that the mid-field trend guidance direction is always within the 0.8-meter safe boundary. The trend guidance direction in the negative X-axis direction is used as the directional condition for updating the far-field path, ensuring that the far-field global reference path extends along the negative X-axis, integrating to form an obstacle avoidance decision objective: near-field avoidance within the 0.8-meter safe boundary, mid-field following the dynamic operator along the negative X-axis, and far-field movement towards the global target pose along a path conforming to this direction, achieving three-layer regional obstacle avoidance coordination.

[0146] Step 302: Based on the obstacle avoidance decision objective, initiate asynchronous collaborative iteration to update each level of the region, obtaining the updated results for each level. Asynchronous collaborative iteration instructions: In the near-field avoidance region, update the safe movement boundary based on real-time perception data in the first update cycle. In the mid-field interaction region, update the trend guidance direction in the second update cycle, using the safe movement boundary generated in the previous cycle as the constraint condition of the search space. In the far-field guidance region, update the global reference path in the third update cycle, using the trend guidance direction generated in the previous cycle as the distribution center of the initial population.

[0147] Optionally, the robot initiates an asynchronous collaborative iteration process based on the obstacle avoidance decision objective. Asynchronous collaborative iteration refers to the three levels of near-field, mid-field, and far-field updating the obstacle avoidance results separately according to their own independent update cycles. At the same time, the update result of the previous cycle of the upper-level region is used as the update constraint or initial condition of the current level region to achieve coordinated linkage of the three levels of regions, rather than synchronous updates. This ensures the real-time performance of near-field emergency obstacle avoidance while also taking into account the stability and optimality of mid-field and far-field obstacle avoidance.

[0148] The specific rules for asynchronous collaborative iteration are as follows: First, in the near-field obstacle avoidance zone, the robot updates the safe movement boundary based on real-time environmental perception data at the first update cycle. The first update cycle refers to the update interval of the safe movement boundary in the near-field obstacle avoidance zone. Since the near field needs to deal with emergency obstacle avoidance, the first update cycle is set to the shortest to adapt to the real-time requirements of the near field. Real-time environmental perception data refers to the information on the position and movement changes of near-field obstacles collected by the robot in real time through devices such as LiDAR and vision sensors. Based on this data, the robot adjusts the safe movement boundary in real time to ensure that the boundary can always avoid all obstacles in the near field.

[0149] Second, in the mid-field interaction area, the robot updates the trend guidance direction using the safe motion boundary generated in the previous cycle as a constraint on the search space during the second update cycle. The second update cycle is longer than the first, adapting to the characteristics of the relatively slow movement speed of dynamic obstacles in the mid-field and the relatively gentle environmental changes. During the update, the safe motion boundary of the previous cycle in the near field is used as a constraint to ensure that the new trend guidance direction does not exceed the safe boundary. Simultaneously, the trend guidance direction is adjusted based on changes in the movement trend of the group of dynamic obstacles in the mid-field to maintain the accuracy of following.

[0150] Third, in the far-field guidance area, the robot updates the global reference path using the trend guidance direction generated in the previous cycle as the initial population distribution center in the third update cycle. The third update cycle is longer than the second update cycle, adapting to the characteristics of fixed positions of static obstacles in the far field and slow environmental changes. During the update, the trend guidance direction from the previous cycle in the mid-field is used as the initial population distribution center, ensuring that the overall direction of the global reference path is consistent with the mid-field following direction. Simultaneously, the path nodes are fine-tuned based on subtle changes in the far-field environment to maintain the path's optimality.

[0151] The three hierarchical regions iterate asynchronously and collaboratively according to their respective update cycles and rules. Each iteration is constrained by the update results of the previous cycle of the previous level, ensuring that the obstacle avoidance results of the three hierarchical regions are always consistent. The iteration continues until the obstacle avoidance results of each level of region tend to stabilize (i.e., the difference between the update results of two consecutive cycles is less than the preset difference threshold), at which point the iteration stops and the updated results of each level of region are obtained, including the updated safe movement boundary, the updated trend guidance direction, and the updated global reference path.

[0152] The first update cycle is 0.1 seconds, the second is 0.3 seconds, and the third is 1.0 second. The robot initiates asynchronous collaborative iteration: In the near field, every 0.1 seconds, based on real-time obstacle information collected by the LiDAR, the safe movement boundary is updated to ensure that the boundary always avoids near-field static shelves and dynamic workers. In the mid-field, every 0.3 seconds, using the safe movement boundary updated in the previous cycle (at 0.1 seconds) as a constraint, and combining subtle changes in the movement trend of dynamic workers, the trend guidance direction is fine-tuned to always remain along the negative X-axis and within the safe boundary. In the far field, every 1.0 second, using the trend guidance direction updated in the previous cycle (at 0.3 seconds) as the initial population distribution center, the node positions of the global reference path are fine-tuned to ensure that the path as a whole is along the negative X-axis and avoids far-field fixed shelves. After three iterations, the update results of each level of region tend to stabilize, resulting in the updated safe movement boundary (still with a radius of 0.8 meters and a slight adjustment of 0.05 meters in position), the updated trend guidance direction (still in the negative X-axis direction and a slight adjustment of 2° in angle), and the updated global reference path (with slight adjustments to node positions and no change in overall direction).

[0153] Step 303: Based on the updated results of each level region, perform multi-timescale temporal trajectory coupling to obtain the optimal obstacle avoidance trajectory.

[0154] Optionally, the robot performs multi-timescale temporal trajectory coupling based on the updated results of each level of region, including the updated safe movement boundary, the updated trend guidance direction, and the updated global reference path. Multi-timescale refers to aligning the obstacle avoidance results of the three levels of regions according to the time dimension, corresponding to different update cycles, ensuring that the obstacle avoidance requirements of the three levels of regions can be met simultaneously at the same time point. Temporal trajectory coupling refers to fusing the obstacle avoidance results of the three levels of regions in both the time and spatial dimensions, eliminating conflicts and contradictions among them, and achieving smooth trajectory connection.

[0155] The specific coupling process is as follows: Based on the update cycles of the three hierarchical regions, the time axis is divided into multiple time segments. Each time segment corresponds to a near-field update cycle, multiple near-field update cycles correspond to a mid-field update cycle, and multiple mid-field update cycles correspond to a far-field update cycle, achieving time scale alignment. Within each time segment, the updated safe motion boundary serves as the near-field constraint, ensuring the trajectory does not exceed the safe boundary within that time segment, thus avoiding near-field collision risks. The updated trend guidance direction serves as the mid-field guide, ensuring the trajectory's direction of travel within that time segment conforms to the trend guidance direction, achieving smooth mid-field following. The updated global reference path serves as the far-field target, ensuring the trajectory's direction of travel within that time segment conforms to the global reference path and does not deviate from the far-field target. The trajectories of adjacent time segments are smoothly connected, eliminating trajectory breakpoints and sharp turns to ensure the continuity and smoothness of the entire trajectory. Simultaneously, it is verified whether the entire trajectory meets the robot's motion constraints. If it does, the trajectory is the optimal obstacle avoidance trajectory. If not, the trajectory's speed and direction are fine-tuned until the motion constraints are met.

[0156] Continuing with step 302, the updated safe motion boundaries, trend guidance direction, and global reference path are obtained. The robot divides the timeline into multiple time segments according to the first update cycle (0.1 seconds): three near-field time segments (0.3 seconds) correspond to one mid-field time segment, and three mid-field time segments (0.9 seconds) correspond to one far-field time segment, achieving time scale alignment. Within each 0.1-second time segment, the robot ensures that the trajectory does not exceed the safe motion boundaries, the direction of travel conforms to the trend guidance direction in the negative X-axis direction, and the overall trajectory extends towards nodes 2, 4, and 6 of the global reference path. The trajectories of adjacent time segments were smoothly connected to eliminate the problem of excessive turning angles. Verification showed that the trajectory met the robot's motion constraints (maximum linear velocity 1.5 m / s, maximum angular velocity 1.2 radians / s). The optimal obstacle avoidance trajectory was finally obtained: starting from the current pose, the robot smoothly starts within a 0.8-meter safety boundary in the near field, follows the dynamic worker along the negative X-axis, passes through nodes 2, 4, and 6 in sequence, and finally reaches the global target pose. The trajectory is continuous and smooth, with no collision risk, and conforms to the global optimal path.

[0157] The embodiments of this invention realize asynchronous coordination and spatiotemporal coupling of safe motion boundaries, trend guidance directions, and global reference paths. This solves the problem of local obstacle avoidance deviating from the global target and trajectory becoming unexecutable due to independent operation or simple serial execution of global planning and local obstacle avoidance. The obtained optimal obstacle avoidance trajectory takes into account near-field safety, mid-field stability, and far-field optimality, providing a precise and executable trajectory for subsequent trajectory mapping and robot control, and improving the stability and coordination of the robot's autonomous obstacle avoidance in complex dynamic environments.

[0158] Furthermore, the robot provided by the present invention will be described below. The robot described below can be referred to in correspondence with the robot dynamic obstacle avoidance method based on swarm intelligence described above.

[0159] Reference Figure 2 , Figure 2 This is a structural schematic diagram of the robot provided by the present invention. The robot includes: The spatial grading module 210 is used to perform spatial grading based on the robot's current pose and environmental perception data to obtain the near-field avoidance area, the mid-field interaction area, and the far-field guidance area. The swarm intelligence collaboration module 220 is used to perform obstacle avoidance search based on the local repulsion characteristics of the first swarm intelligence combined with obstacle information in the near-field avoidance area to obtain a safe movement boundary; to perform obstacle detour search based on the dynamic following characteristics of the second swarm intelligence combined with obstacle information in the mid-field interaction area to obtain a trend guidance direction; and to perform obstacle crossing search based on the global topology exploration characteristics of the third swarm intelligence combined with obstacle information in the far-field guidance area to obtain a global reference path. The obstacle avoidance trajectory planning module 230 is used to asynchronously coordinate and spatiotemporally couple based on safe motion boundaries, trend guidance directions and global reference paths to obtain the optimal obstacle avoidance trajectory. The obstacle avoidance control module 240 is used to map the trajectory characteristics of the optimal obstacle avoidance trajectory to the linear velocity and angular velocity components of the robot chassis based on the robot's motion constraints, and control the robot to perform obstacle avoidance motion.

[0160] This invention utilizes the different characteristics of multi-level swarm intelligence to search for obstacles in different levels of regions. By combining the asynchronous collaboration and spatiotemporal coupling of multi-level swarm intelligence, it achieves dynamic autonomous obstacle avoidance for robots, thereby improving the stability of autonomous obstacle avoidance in complex dynamic environments.

[0161] Please see Figure 3 , Figure 3 An embodiment diagram of an electronic device provided in accordance with the present invention. For example... Figure 3 As shown, an embodiment of the present invention provides an electronic device 300, including a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor 320. When the processor 320 executes the computer program 311, it implements the processes of steps 10 to 40.

[0162] Please see Figure 4 , Figure 4 An embodiment diagram of a computer-readable storage medium provided in accordance with an embodiment of the present invention is shown. Figure 4As shown, this embodiment provides a computer-readable storage medium 400 on which a computer program 311 is stored. When the computer program 311 is executed by a processor, it implements the processes of steps 10 to 40.

[0163] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the robot dynamic obstacle avoidance method based on swarm intelligence provided by the above methods, which includes steps 10 to 40.

Claims

1. A method for dynamic obstacle avoidance of robots based on swarm intelligence, characterized in that, include: Based on the robot's current pose and environmental perception data, spatial classification is performed to obtain the near-field avoidance area, the mid-field interaction area, and the far-field guidance area. Based on the local repulsion characteristics of the first swarm intelligence combined with obstacle information in the near-field avoidance area, obstacle avoidance search is performed to obtain the safe movement boundary. Based on the dynamic following characteristics of the second swarm intelligence combined with obstacle information in the mid-field interaction area, obstacle detour search is performed to obtain the trend guidance direction. Based on the global topology exploration characteristics of the third swarm intelligence combined with obstacle information in the far-field guidance area, obstacle crossing search is performed to obtain the global reference path. The optimal obstacle avoidance trajectory is obtained by asynchronous coordination and spatiotemporal coupling based on safe motion boundaries, trend guidance directions and global reference paths; Based on the robot's motion constraints, the trajectory characteristics of the optimal obstacle avoidance trajectory are mapped to the linear velocity and angular velocity components of the robot chassis, thereby controlling the robot to perform obstacle avoidance motion.

2. The robot dynamic obstacle avoidance method based on swarm intelligence according to claim 1, characterized in that, Determine the safe movement boundaries of the near-field avoidance zone, including: A spherical detection space is constructed based on the current pose and a preset safe detection radius. Based on the spherical detection space and the local repulsion characteristics of the first swarm intelligence, a spatial discretization grid mapping is performed to obtain the initial discrete sampling points within the spherical detection space. Based on the initial discrete sampling points and the first obstacle position distribution information, the static obstacle space occupancy determination is performed to obtain the first discrete sampling points located inside the entity of the static obstacle. Complementary analysis is performed based on the first discrete sampling point and the initial discrete sampling point to obtain a second discrete sampling point inside the entity that has not entered the static obstacle. Based on the second discrete sampling point, the first obstacle location distribution information, and the first motion state information, spatiotemporal trajectory intersection prediction is performed to obtain the safe motion boundary; the safe motion boundary indicates static and dynamic obstacles that deviate from the near-field avoidance area.

3. The robot dynamic obstacle avoidance method based on swarm intelligence according to claim 2, characterized in that, The step of predicting the spatiotemporal trajectory intersection based on the second discrete sampling points, the first obstacle position distribution information, and the first motion state information to obtain the safe motion boundary includes: Based on the second discrete sampling point, the first obstacle location distribution information and the first motion state information, spatiotemporal trajectory intersection prediction is performed to obtain conflicting discrete sampling points that have spatial intersection with the motion trajectory of the dynamic obstacle within a future preset time window. Based on the conflict discrete sampling points and the second discrete sampling points, differential analysis is performed to obtain target discrete sampling points that do not overlap with static obstacles and do not intersect with the trajectories of dynamic obstacles; Based on the target discrete sampling points and the local repulsion characteristics of the first collective intelligence, a boundary expansion evolution is performed to obtain a discrete sampling point cloud that extends outward under local repulsion constraints until it contacts the environmental constraint boundary; during the boundary expansion evolution process, the point sets of discrete sampling points maintain a preset maximum spacing. Based on the discrete point set associated with the discrete sampled point cloud, a convex hull and contour closure connection are constructed to obtain the target boundary line containing each discrete sampled point cloud. Based on the target boundary line and the robot's pose coordinate system, a relative position reference mapping is performed to obtain the safe motion boundary.

4. The robot dynamic obstacle avoidance method based on swarm intelligence according to claim 1, characterized in that, The second swarm intelligence's dynamic following characteristics, combined with obstacle information in the midfield interaction area, perform obstacle avoidance search to obtain trend guidance direction, including: Based on the second motion state information and the current pose, a motion manifold mapping is performed to obtain the discrete velocity point cloud of the endpoints of the relative velocity vector of each dynamic obstacle relative to the robot within the field interaction area. Based on the dynamic following characteristics of the discrete velocity point cloud and the second swarm intelligence, velocity spatial neighborhood connectivity analysis is performed to obtain an initial velocity connectivity domain; the velocity difference between dynamic obstacles in each initial velocity connectivity domain is less than or equal to a preset difference threshold, and the spatial distance is less than or equal to a preset distance threshold. Based on the static environment visual topology of each initial velocity connected domain and the second obstacle position distribution information, the target velocity connected domain that is not blocked by static obstacles is obtained. Based on the connected domain of each target velocity, principal axis analysis is performed to obtain the principal axis direction vector; each principal axis direction vector represents the motion tendency of the dynamic obstacle group within the connected domain of each target velocity. Based on the direction vector of each principal axis and the current pose, a trend guidance direction is determined; the trend guidance direction indicates the direction in which the robot avoids dynamic obstacles in the midfield interaction area and follows the group movement trend.

5. The robot dynamic obstacle avoidance method based on swarm intelligence according to claim 4, characterized in that, The process of determining the trend guidance direction based on each principal axis direction vector and the current pose includes: Based on each principal axis direction vector and the current pose, follow-up frustum geometry is constructed to obtain the minimum frustum region with the current pose as the vertex and covering each principal axis direction vector; Based on the boundary geometric constraints of the minimum view cone region and the mid-field interaction region, the intersection of feasible passable sectors is obtained to obtain a local following region located within the boundary of the mid-field interaction region and not conflicting with static obstacles. Based on the local following region, the central angle of the sector is calculated and compared with the maximum value to obtain the optimal guiding sector with the largest central angle value in the local following region. Based on the angle bisector direction of the optimal guidance sector and the robot's current pose coordinate system, polar coordinate direction mapping analysis is performed to obtain the trend guidance direction.

6. The robot dynamic obstacle avoidance method based on swarm intelligence according to claim 1, characterized in that, The global topology exploration characteristics based on third-party collective intelligence, combined with obstacle information from the far-field guidance area, are used to perform obstacle traversal search and obtain a global reference path, including: Based on the current pose and the global target pose specified by the task, combined with the boundary of the obstacle-free space in the far-field guidance area, a bidirectional initialization distribution analysis of the mass group is performed to obtain initial mass points that start from the current pose and the global target pose respectively and are located in the obstacle-free space. Based on the initial mass point, the location distribution information of the third obstacle, and the global topology exploration characteristics of the third swarm intelligence, the obstacle boundary is repulsed and evolved to obtain a target mass point cloud that reaches a preset mechanical equilibrium state in the central area of ​​the channel corresponding to the static obstacle. Based on the target particle distribution point cloud, local density analysis is performed to obtain local density peak points. Based on each local density peak point and the location distribution information of the third obstacle, maximum unobstructed coverage analysis is performed to obtain the maximum safe passage radius of each local density peak point. The effective connectivity of adjacent points is determined based on the maximum safe passage radius of each local density peak point, resulting in a global reference path. The global reference path is used to indicate static obstacles that deviate from the far-field guidance area.

7. The robot dynamic obstacle avoidance method based on swarm intelligence according to claim 6, characterized in that, The determination of effective connectivity between adjacent points based on the maximum safe passage radius of each local density peak point, to obtain a global reference path, includes: Based on the maximum safe passage radius of each local density peak point, the effective connection of adjacent points is determined, and the target adjacent node pairs with overlapping areas are obtained. Based on the target adjacent node pairs, an undirected connected graph is constructed to obtain a far-field region topology connection graph with local density peak points as nodes and target adjacent node pairs as edges. Based on the far-field region topology connection graph, the starting node corresponding to the current pose and the ending node corresponding to the global target pose, a minimum hop path search is performed to obtain the target waypoint sequence connecting the starting node and the ending node. Based on the target waypoint sequence, a segmented straight-line geometric connection analysis is performed to obtain a global reference path that is a series of straight lines connecting the target waypoints.

8. The robot dynamic obstacle avoidance method based on swarm intelligence according to claim 1, characterized in that, The process of asynchronously coordinating and spatiotemporally coupling based on safe motion boundaries, trend guidance directions, and global reference paths to obtain the optimal obstacle avoidance trajectory includes: Using the safe movement boundary as the collision avoidance space constraint condition when searching the mid-field interaction area, and the trend guidance direction as the obstacle avoidance direction condition when searching the far-field guidance area, the obstacle avoidance decision objective is obtained. Based on the obstacle avoidance decision objective, asynchronous collaborative iteration is initiated to update each level of region, obtaining the updated results for each level of region; the asynchronous collaborative iteration indicates that: in the near-field avoidance region, the safe movement boundary is updated in the first update cycle based on real-time perception data; in the mid-field interaction region, the trend guidance direction is updated in the second update cycle using the safe movement boundary generated in the previous cycle as the constraint condition of the search space; in the far-field guidance region, the global reference path is updated in the third update cycle using the trend guidance direction generated in the previous cycle as the distribution center of the initial population; The optimal obstacle avoidance trajectory is obtained by coupling the updated results of each level region with time-scale temporal trajectories.

9. The robot dynamic obstacle avoidance method based on swarm intelligence according to any one of claims 1 to 8, characterized in that, The obstacle information includes the location distribution information of static obstacles and the motion state information of dynamic obstacles.

10. A robot, characterized in that, For implementing the swarm intelligence-based robot dynamic obstacle avoidance method as described in any one of claims 1 to 9; the robot includes: The spatial grading module is used to perform spatial grading based on the robot's current pose and environmental perception data, resulting in a near-field avoidance area, a mid-field interaction area, and a far-field guidance area. The swarm intelligence collaboration module is used to search for obstacles based on the local repulsion characteristics of the first swarm intelligence combined with obstacle information in the near-field avoidance area to obtain a safe movement boundary; to search for obstacles based on the dynamic following characteristics of the second swarm intelligence combined with obstacle information in the mid-field interaction area to obtain a trend guidance direction; and to search for obstacles based on the global topology exploration characteristics of the third swarm intelligence combined with obstacle information in the far-field guidance area to obtain a global reference path. The obstacle avoidance trajectory planning module is used to asynchronously coordinate and spatiotemporally couple based on safe motion boundaries, trend guidance directions, and global reference paths to obtain the optimal obstacle avoidance trajectory. The obstacle avoidance control module is used to map the trajectory characteristics of the optimal obstacle avoidance trajectory to the linear velocity and angular velocity components of the robot chassis based on the robot's motion constraints, and control the robot to perform obstacle avoidance motion.

Images