A dynamic region collision detection method based on voxelized model

Abstract

The application discloses a kind of dynamic area collision detection methods based on voxelization model, belong to robot autonomous navigation and obstacle avoidance technical field.The application is first by robot three-dimensional model by feature extraction, polyhedral fitting and Boolean operation to construct simplified collision model and voxelization;Rely on simplified model by space transformation and Boolean operation to generate movement swept volume and voxelization, obtain the voxel obstacle avoidance area of uniform specification;Environment sensing point cloud is mapped to same voxel grid, and collision risk is determined by voxel intersection judgment;Offline pre-generation of multiple speed, angular velocity corresponding obstacle avoidance area group, switch on demand during operation.The application overcomes the defects that the modeling precision of existing bounding box is poor, the geometry and voxel format do not match, and the precision and real-time performance are difficult to balance, reduces the computing overhead under the premise of ensuring the fitting precision of robot shape, relies on voxel parallel computing to improve the detection rate, and considers the collision detection accuracy and real-time performance, and the model adaptability is strong.

Description

Technical Field

[0001] This invention relates to a dynamic region collision detection method based on a voxel model, belonging to the field of robot autonomous navigation and obstacle avoidance technology. Background Technology

[0002] In the autonomous navigation process of wheeled robots, collision detection is a key link to achieve dynamic obstacle avoidance. Collision detection is mainly divided into two parts: robot body collision volume modeling and environmental obstacle information processing. The two are combined to complete the robot collision risk assessment.

[0003] Currently, robot collision volume modeling mainly includes three approaches: first, using simple bounding boxes such as spheres, cylinders, and cuboids to simplify the robot's shape; second, generating swept volumes based on bounding box models combined with motion trajectories; and third, directly using a complete triangular mesh of the robot as the collision volume. Common methods for handling environmental obstacles include direct discrimination from raw point clouds, 2D grid / polar coordinate obstacle maps, and 3D voxel maps of obstacles. The most widely used mainstream collision detection scheme is a combination of swept volumes and environmental obstacle object mapping.

[0004] The existing combination schemes mentioned above have several inherent defects in practical applications: First, the swept volume is constructed based on a simple bounding box, which deviates significantly from the actual shape of the robot. In narrow working conditions, it is prone to collision misjudgment or missed detection. Replacing it with a high-precision triangular mesh model would bring huge computational costs, making it difficult to meet the real-time computing needs of the device. Second, environmental obstacles are stored in voxel format, while the robot's swept volume is still expressed in a continuous geometric way. The two data formats are not consistent, and additional data conversion between geometry and voxels is required before intersection calculation, which cannot take advantage of the advantages of voxel parallel computing and hardware acceleration. Third, the obstacle avoidance area is difficult to balance environmental adaptability and real-time performance. The fixed-size obstacle avoidance area cannot adjust its occupied space according to changes in the robot's linear velocity and angular velocity. After the motion conditions change, the detection accuracy drops significantly. Regenerating the swept volume online in real time through geometric Boolean operations is computationally intensive and lacks real-time performance, so it is impossible to simultaneously ensure collision detection accuracy and operational real-time performance. Summary of the Invention

[0005] To simultaneously improve collision detection accuracy and real-time performance, this invention provides a dynamic region collision detection method based on a voxelization model, the technical solution of which is as follows: Step 1: Construct a simplified collision volume model based on the robot's 3D digital model, and perform voxelization to obtain the basic voxel region; Step 2: Based on the simplified collision volume model and robot motion parameters, obtain the motion sweep volume corresponding to the motion state through spatial transformation and Boolean operations; Step 3: Voxelize the motion sweep volume to generate a three-dimensional voxel mesh obstacle avoidance region with the same coordinate system and resolution as the base voxel region; Step 4: Voxelize the point cloud of environmental obstacles and map it to the same 3D voxel mesh, perform voxel intersection with the obstacle avoidance area, and determine the collision risk based on the intersection result; Step 5: Pre-generate multiple sets of obstacle avoidance area voxel meshes with different motion parameters offline. When running online, directly match and switch the obstacle avoidance area of ​​the corresponding level according to the current motion parameters.

[0006] Optionally, step 1 includes: Feature extraction is performed on the three-dimensional digital model of the robot to obtain a set of geometric feature points that characterize its external contour; Based on the nature of the protected components and the degree of collision hazard, a convex hull generation strategy is determined for each component, and an initial convex hull set is generated based on the geometric feature point set. Among them, for critical components with a high degree of collision hazard, a conservative convex hull with an expanded envelope range is generated to ensure safety, while other components generate a compact convex hull that closely resembles the real contour to reduce complexity. Polyhedral fitting is performed on each convex hull in the initial convex hull set to approximate the corresponding geometric contour with the fewest planes within a set fitting error threshold, thereby obtaining the fitted polyhedron for each component. If a convex hull itself is a simple polyhedron, no fitting simplification is needed, and it is directly used as the fitted polyhedron. Based on the kinematic configuration of the robot, at least one combination of Boolean union, Boolean intersection and Boolean difference is selectively performed on the fitted polyhedron to construct a simplified collision volume model with a reduced number of facets and differentiated safety boundaries for different components. The simplified collision volume model is voxelized to obtain the basic voxel region.

[0007] Thus, based on the different degrees of collision damage to components, convex hull strategies are set differently. Conservative convex hulls are used for critical components to improve safety, while tight convex hulls are used for other components to fit the real contour. By combining polyhedral fitting and Boolean combination to construct a simplified collision model, the problem of collision misjudgment and missed detection is effectively reduced. At the same time, the model structure is simplified and the subsequent amount of computation is reduced. While ensuring the accuracy of collision detection, the overall operating efficiency is further improved.

[0008] Optionally, step 1 includes: For a predefined cube voxel mesh, detect all voxels that intersect with the external contour surface provided by the robot's 3D digital model, mark the intersecting voxels as surface voxels, and form a set of surface voxels. Within the space enclosed by the surface voxel set, an internal seed voxel is determined. The determination method is as follows: select a center point of a voxel, and use the ray intersection parity rule or the signed distance function to determine whether the point is located inside the entity. If it is, then the voxel is taken as the internal seed voxel; otherwise, traverse the voxels along a predetermined direction until a voxel located inside the entity is obtained as the internal seed voxel. Starting with the internal seed voxel, check the adjacent voxels of the current internal voxel in the specified neighborhood. If the adjacent voxel does not belong to the surface voxel set and has not been marked as an internal voxel, then mark the adjacent voxel as an internal voxel and add it to the growth queue. Repeat this process until all non-surface voxels connected to the internal seed voxel are marked as internal voxels. The surface voxel set is merged with all the marked internal voxels to obtain all the voxels of the robot 3D entity in the cube voxel mesh, which serve as the base voxel region.

[0009] In this way, by completing the voxelization of the model through surface voxel marking, internal seed voxel determination and voxel growth algorithm, the spatial range of the robot entity can be accurately restored. The inner and outer contours of the model can be completely reproduced, ensuring the integrity and accuracy of the basic voxel region, avoiding collision judgment deviation caused by missing voxels, and quickly completing voxel modeling, providing a reliable data foundation for subsequent collision detection.

[0010] Optionally, the process of generating the motion sweep volume includes: Based on the simplified collision volume model, the motion trajectory is determined according to the robot's linear velocity and angular velocity. The robot performs pose sampling and spatial unfolding at multiple moments along the trajectory, and obtains the total space occupied by the robot during the motion process through geometric Boolean union operation.

[0011] Optionally, the environmental obstacle point cloud voxelization process includes: After preprocessing the obstacle point cloud acquired by the sensor, it is projected onto a three-dimensional voxel grid with the same resolution and coordinate system as the obstacle avoidance area, and the voxel units occupied by the obstacles are marked.

[0012] Optionally, the obstacle point cloud is derived from data acquired by multiple sensors, and the preprocessing of the obstacle point cloud includes the fusion of point cloud data from multiple sensors.

[0013] Optionally, step 4 specifically includes: performing voxel intersection operation on the map of environmental obstacles and the voxel grid of the obstacle avoidance area corresponding to the current motion state in a unified three-dimensional grid space, and determining whether each voxel is simultaneously occupied by an obstacle and a dynamic obstacle avoidance area; if there is a non-empty intersection, it is determined that there is a collision risk for the robot to move according to the current motion state; if there is no intersection, it is determined that the current trajectory is safe.

[0014] Optionally, step 5 includes: The robot's linear velocity and angular velocity are predefined offline to form multiple sets of motion parameter combinations. For each set of motion parameters, steps 2 to 3 are repeated to pre-calculate and store the corresponding three-dimensional voxel mesh obstacle avoidance areas in batches, thus constructing a multi-level dynamic obstacle avoidance area voxel mesh group.

[0015] Optionally, the environmental obstacle point cloud is acquired through a three-dimensional sensor. The three-dimensional sensor refers to a device that can collect and output three-dimensional geometric information of the environment or target object, including but not limited to LiDAR, depth cameras (such as ToF cameras and structured light cameras), binocular stereo vision cameras, and other sensor elements that can be used for three-dimensional environment reconstruction.

[0016] Three-dimensional sensors are used to perceive the spatial distribution of obstacles in the robot's workspace in real time, and to obtain the position, size and contour information of the obstacles in order to construct a figurative model of the environmental obstacles.

[0017] Secondly, the present invention provides a dynamic region collision detection system based on a voxelization model, the system being configured to implement the dynamic region collision detection method based on a voxelization model as described in any of the preceding claims, comprising: The model simplification voxelization module is configured to construct a simplified collision volume model based on the robot's 3D digital model and perform voxelization to obtain the basic voxel region. The sweep volume calculation module is configured to obtain the motion sweep volume of the corresponding motion state through spatial transformation and Boolean operation based on the simplified collision volume model and robot motion parameters. The obstacle avoidance region generation module is configured to voxelize the motion sweep volume and generate a three-dimensional voxel mesh obstacle avoidance region with the same coordinate system and resolution as the basic voxel region. The collision determination module is configured to voxelize the point cloud of environmental obstacles and map it to the same three-dimensional voxel mesh, perform voxel intersection with the obstacle avoidance area, and determine the collision risk based on the intersection result. The partitioned storage and calling module is configured to pre-generate multiple sets of obstacle avoidance area voxel meshes corresponding to different motion parameters offline, and directly match and switch the corresponding obstacle avoidance area according to the current motion parameters during online operation.

[0018] Thirdly, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the dynamic region collision detection method based on the voxelization model as described in any of the preceding claims.

[0019] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the steps of the dynamic region collision detection method based on a voxelization model as described in any of the preceding claims.

[0020] The beneficial effects of this invention are: This invention overcomes the shortcomings of traditional simple bounding box shape fitting accuracy and large computational load of precision triangular mesh model by extracting features from the robot's three-dimensional digital model, constructing a simplified collision model, and then performing body voxelization processing. It can not only fit the robot's real concave and convex contours and avoid collision misjudgment and missed detection problems in narrow scenes, but also effectively reduce the model's computational load and improve the efficiency of body modeling.

[0021] This invention transforms the robot's sweep volume into a three-dimensional voxel mesh with the same specifications and coordinate system as the environment, thereby achieving data format homogeneity between the robot's obstacle avoidance area and environmental obstacles. This eliminates the additional computational overhead of converting between geometry and voxels, fully utilizes the characteristics of voxel data to achieve parallel accelerated computation, and significantly improves the computational speed of collision intersection.

[0022] This invention addresses the shortcomings of existing technologies, such as excessive computational load in real-time online generation of swept volume and the inability of fixed obstacle avoidance areas to adapt to changes in motion state, by pre-storing multiple groups of voxel obstacle avoidance regions corresponding to different linear velocities and angular velocities offline and retrieving and switching them online as needed. While dynamically adjusting the obstacle avoidance range according to the robot's real-time motion parameters and ensuring detection accuracy, it eliminates the need for online Boolean operations and real-time voxelization, significantly reducing the overall computational load and saving computing power. It also reduces the requirements for hardware performance while balancing collision detection accuracy and real-time system performance.

[0023] This invention only requires replacing the robot's simplified collision volume model and related parameters to adapt to different models of equipment, without the need for underlying algorithm reconstruction. It has good versatility and expands the scope of application of the algorithm. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart of the dynamic region collision detection method based on the voxelization model in Embodiment 2 of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0027] Example 1: This embodiment provides a dynamic region collision detection method based on a voxelization model, and the technical solution is as follows: Step 1: Construct a simplified collision volume model based on the robot's 3D digital model, and perform voxelization to obtain the basic voxel region; Step 2: Based on the simplified collision volume model and robot motion parameters, obtain the motion sweep volume corresponding to the motion state through spatial transformation and Boolean operations; Step 3: Voxelize the motion sweep volume to generate a 3D voxel mesh obstacle avoidance region with the same coordinate system and resolution as the base voxel region; Step 4: Convert the point cloud of environmental obstacles into voxels and map it to the same 3D voxel mesh. Perform voxel intersection with the obstacle avoidance area and determine the collision risk based on the intersection results. Step 5: Pre-generate multiple sets of obstacle avoidance area voxel meshes with different motion parameters offline. When running online, directly match and switch the obstacle avoidance area of ​​the corresponding level according to the current motion parameters.

[0028] Example 2: This embodiment provides a dynamic region collision detection method based on a voxelization model, which is applied to the autonomous navigation and obstacle avoidance scenarios of wheeled robots. The complete implementation steps are as follows.

[0029] Step 1: Construct a simplified collision volume model based on the accurate numerical model and complete the basic voxelization.

[0030] Import a complete and accurate 3D digital model of the wheeled robot, extract key geometric features, and retain the contour features affecting passage, such as the chassis, drive wheels, driven wheels, sensor protrusions, and convex structures. Using convex hull generation, polyhedral fitting, and Boolean union / intersection / difference operations, the high-precision mesh model is simplified into a simplified 3D collision volume model that accurately approximates the real shape and is compatible with concave and convex features. This simplified collision volume serves as the basic collision area for dynamic obstacle avoidance. The specific implementation steps are as follows: Step 11: Extract features from the robot's 3D digital model to obtain a set of geometric feature points that characterize its outline; Step 12: Based on the nature of the protected component and its collision hazard level, determine the convex hull generation strategy for each component, and generate an initial convex hull set based on the geometric feature point set. Among them, for key components with high collision hazard level, a conservative convex hull with expanded envelope range is generated to ensure safety, and other components generate a compact convex hull that closely resembles the real contour to reduce complexity. Step 13: Perform polyhedral fitting on each convex hull in the initial convex hull set to approximate the corresponding geometric contour with the fewest planes within a set fitting error threshold, thereby obtaining the fitted polyhedron for each component; where, if a convex hull itself is a simple polyhedron, no fitting simplification is needed, and it is directly used as the fitted polyhedron. Step 14: Based on the robot's kinematic configuration, selectively perform at least one combination of Boolean union, Boolean intersection, and Boolean difference on the fitted polyhedron to construct a simplified collision volume model with a reduced number of facets and differentiated safety boundaries for different components. Step 15: Voxelize the simplified collision volume model to obtain the basic voxel region.

[0031] In this embodiment, the basic collision volume model is simultaneously discretized into three-dimensional voxels. A fixed-resolution regular voxel mesh is used for partitioning, and the voxel side length is set according to the robot's minimum feature size and scene accuracy requirements to complete the basic voxel modeling of the robot body, providing a unified geometric representation benchmark for subsequent motion sweep calculations.

[0032] Step 2: Calculate the motion sweep volume under a specific motion state based on the simplified collision volume model.

[0033] Based on the simplified collision volume model generated in step 1, the prediction time and motion trajectory are determined according to the robot's linear velocity and angular velocity. Through spatial transformations such as translation, rotation, and pose recursion, the simplified model is sampled and spatially unfolded at multiple time points along the planned trajectory.

[0034] By employing geometric Boolean union operations and fusing simplified models from all sampled poses, the sweep volume of the robot's motion in the future time period under the given velocity and angular velocity combination is calculated, fully covering the entire spatial range occupied by the robot during its motion, forming a dynamic envelope region with a continuous geometric representation.

[0035] Step 3: Motion sweep volume voxelization to generate a 3D voxel mesh obstacle avoidance area.

[0036] The continuous geometric sweep volume obtained in step 2 is discretized into regular three-dimensional voxels using the same voxel resolution and coordinate system as the basic model. The continuous envelope region is transformed into an occupied voxel mesh, resulting in a three-dimensional voxel obstacle avoidance region that matches the robot's current motion state.

[0037] Step 4: Environmental obstacle object simplification and unified spatial mapping.

[0038] The point cloud of obstacles in the scene is acquired by 3D sensors such as LiDAR and depth camera, and projected onto a 3D voxel mesh with the same resolution and coordinate system as the obstacle avoidance area to generate an environmental obstacle voxel occupancy map. All voxel units occupied by obstacles are marked to complete the 3D voxel representation of environmental information.

[0039] Step 5: Voxel-level intersection determination and collision risk identification.

[0040] The voxel intersection operation is performed on the map of environmental obstacles and the voxel grid of the obstacle avoidance area corresponding to the current motion state in a unified three-dimensional grid space. Each voxel is then checked to determine whether it is simultaneously occupied by obstacles and dynamic obstacle avoidance area.

[0041] If a non-empty intersection exists, it is determined that there is a risk of collision if the robot continues to move in its current state; if there is no intersection, the current trajectory is determined to be safe, and navigation can continue.

[0042] Step 6: Pre-generation and dynamic switching of multi-level obstacle avoidance area groups during runtime.

[0043] Typical graded ranges of robot linear velocity and angular velocity are predefined offline to form multiple sets of motion parameter combinations. For each set of parameters, steps 2 to 3 are repeated to pre-calculate and store the corresponding three-dimensional voxel obstacle avoidance regions in batches, and to construct a multi-level dynamic obstacle avoidance region voxel mesh group.

[0044] When the robot is running online, it reads the current linear velocity and angular velocity in real time, and directly matches and loads the voxel mesh of the obstacle avoidance area corresponding to the pre-stored mesh group. This eliminates the need to perform geometric Boolean operations and voxel discretization in real time, thus achieving a balance between motion state adaptation and real-time collision detection.

[0045] Example 3: This embodiment provides a dynamic region collision detection method based on a voxelization model. The difference between this embodiment and Embodiment 2 is that the method of constructing the basic voxel region in step 1 is different. The basic voxel region construction process in this embodiment is as follows.

[0046] Step 11: For the predefined cube voxel mesh, detect all voxels that intersect with the external contour surface provided by the robot's 3D digital model, mark the intersecting voxels as surface voxels, and form a set of surface voxels.

[0047] Step 12: Determine an internal seed voxel within the space enclosed by the surface voxel set.

[0048] In this embodiment, the method for determining the internal seed voxel is as follows: select a voxel center point, and use the ray intersection parity rule or the sign distance function to determine whether the point is located inside the entity. If it is, then the voxel is taken as the internal seed voxel; otherwise, traverse the voxels along the predetermined direction until a voxel located inside the entity is obtained as the internal seed voxel.

[0049] Step 13: Starting from the internal seed voxel, check the adjacent voxels of the current internal voxel in the specified neighborhood. If the adjacent voxel does not belong to the surface voxel set and has not been marked as an internal voxel, mark the adjacent voxel as an internal voxel and add it to the growth queue. Repeat this process until all non-surface voxels connected to the internal seed voxel are marked as internal voxels. Step 14: Merge the surface voxel set with all the marked internal voxels to obtain all the voxels of the robot 3D entity in the cube voxel mesh, which will serve as the base voxel region.

[0050] Example 4: This embodiment provides a dynamic region collision detection system based on a voxelization model, including: The model simplification voxelization module is configured to construct a simplified collision volume model based on the robot's 3D digital model and perform voxelization to obtain the basic voxel region. The sweep volume calculation module is configured to obtain the motion sweep volume of the corresponding motion state through spatial transformation and Boolean operations based on the simplified collision volume model and robot motion parameters. The obstacle avoidance region generation module is configured to voxelize the motion sweep volume and generate a three-dimensional voxel mesh obstacle avoidance region with the same coordinate system and resolution as the base voxel region. The collision determination module is configured to voxelize the point cloud of environmental obstacles and map it to the same three-dimensional voxel mesh, perform voxel intersection with the obstacle avoidance area, and determine the collision risk based on the intersection result. The partitioned storage and calling module is configured to pre-generate multiple sets of obstacle avoidance area voxel meshes corresponding to different motion parameters offline, and directly match and switch the corresponding obstacle avoidance area according to the current motion parameters during online operation.

[0051] Some steps in the embodiments of the present invention can be implemented using software, and the corresponding software program can be stored in a readable storage medium, such as an optical disc or a hard disk.

[0052] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A dynamic region collision detection method based on a voxelization model, characterized in that, The method includes: Step 1: Construct a simplified collision volume model based on the robot's 3D digital model, and perform voxelization to obtain the basic voxel region; Step 2: Based on the simplified collision volume model and robot motion parameters, obtain the motion sweep volume corresponding to the motion state through spatial transformation and Boolean operations; Step 3: Voxelize the motion sweep volume to generate a three-dimensional voxel mesh obstacle avoidance region with the same coordinate system and resolution as the base voxel region; Step 4: Voxelize the point cloud of environmental obstacles and map it to the same 3D voxel mesh, perform voxel intersection with the obstacle avoidance area, and determine the collision risk based on the intersection result; Step 5: Pre-generate multiple sets of obstacle avoidance area voxel meshes with different motion parameters offline. When running online, directly match and switch the obstacle avoidance area of ​​the corresponding level according to the current motion parameters.

2. The dynamic region collision detection method based on a voxelization model according to claim 1, characterized in that, Step 1 includes: Feature extraction is performed on the three-dimensional digital model of the robot to obtain a set of geometric feature points that characterize its external contour; Based on the nature of the protected components and the degree of collision damage, a convex hull generation strategy is determined for each component, and an initial convex hull set is generated based on the geometric feature point set. Among them, for key components with a high degree of collision damage, a conservative convex hull with an expanded envelope range is generated, and for other components, a compact convex hull that closely resembles the real contour is generated. Polyhedral fitting is performed on each convex hull in the initial convex hull set to approximate the corresponding geometric contour with the fewest planes within a set fitting error threshold, thereby obtaining the fitted polyhedral for each component; wherein, if a certain convex hull itself is a simple polyhedral, it is directly used as the fitted polyhedral. Based on the kinematic configuration of the robot, Boolean combination operations are performed on the fitted polyhedron to construct a simplified collision volume model with differentiated safety boundaries; The simplified collision volume model is voxelized to obtain the basic voxel region.

3. The dynamic region collision detection method based on a voxelization model according to claim 1, characterized in that, Step 1 includes: For a predefined cube voxel mesh, detect all voxels that intersect with the external contour surface provided by the robot's 3D digital model, mark the intersecting voxels as surface voxels, and form a set of surface voxels. Within the space enclosed by the surface voxel set, an internal seed voxel is determined. The determination method is as follows: select a center point of a voxel, and use the ray intersection parity rule or the signed distance function to determine whether the point is located inside the entity. If it is, then the voxel is taken as the internal seed voxel; otherwise, traverse the voxels along a predetermined direction until a voxel located inside the entity is obtained as the internal seed voxel. Starting with the internal seed voxel, check the adjacent voxels of the current internal voxel in the specified neighborhood. If the adjacent voxel does not belong to the surface voxel set and has not been marked as an internal voxel, then mark the adjacent voxel as an internal voxel and add it to the growth queue. Repeat this process until all non-surface voxels connected to the internal seed voxel are marked as internal voxels. The surface voxel set is merged with all the marked internal voxels to obtain all the voxels of the robot 3D entity in the cube voxel mesh, which serve as the base voxel region.

4. The dynamic region collision detection method based on a voxelization model according to claim 1, characterized in that, The process of generating the motion sweep volume includes: Based on the simplified collision volume model, the motion trajectory is determined according to the robot's linear velocity and angular velocity. The robot performs pose sampling and spatial unfolding at multiple moments along the trajectory, and obtains the total space occupied by the robot during the motion process through geometric Boolean union operation.

5. The dynamic region collision detection method based on a voxelization model according to claim 1, characterized in that, The voxelization process for the point cloud of environmental obstacles includes: After preprocessing the obstacle point cloud acquired by the sensor, it is projected onto a three-dimensional voxel grid with the same resolution and coordinate system as the obstacle avoidance area, and the voxel units occupied by the obstacles are marked.

6. The dynamic region collision detection method based on a voxelization model according to claim 1, characterized in that, Step 4 specifically includes: performing voxel intersection operation on the map of environmental obstacles and the voxel grid of the obstacle avoidance area corresponding to the current motion state in a unified three-dimensional grid space, and determining whether each voxel is simultaneously occupied by an obstacle and a dynamic obstacle avoidance area; if there is a non-empty intersection, it is determined that there is a risk of collision if the robot moves according to the current motion state; if there is no intersection, it is determined that the current trajectory is safe.

7. The dynamic region collision detection method based on a voxelization model according to claim 1, characterized in that, Step 5 includes: The robot's linear velocity and angular velocity are predefined offline to form multiple sets of motion parameter combinations. For each set of motion parameters, steps 2 to 3 are repeated to pre-calculate and store the corresponding three-dimensional voxel mesh obstacle avoidance areas in batches, thus constructing a multi-level dynamic obstacle avoidance area voxel mesh group.

8. A dynamic region collision detection system based on a voxelization model, characterized in that, The system is configured to implement the dynamic region collision detection method based on a voxelization model as described in any one of claims 1-7, including: The model simplification voxelization module is configured to construct a simplified collision volume model based on the robot's 3D digital model and perform voxelization to obtain the basic voxel region. The sweep volume calculation module is configured to obtain the motion sweep volume of the corresponding motion state through spatial transformation and Boolean operations based on the simplified collision volume model and robot motion parameters. The obstacle avoidance region generation module is configured to voxelize the motion sweep volume and generate a three-dimensional voxel mesh obstacle avoidance region with the same coordinate system and resolution as the basic voxel region. The collision determination module is configured to voxelize the point cloud of environmental obstacles and map it to the same three-dimensional voxel mesh, perform voxel intersection with the obstacle avoidance area, and determine the collision risk based on the intersection result. The partitioned storage and calling module is configured to pre-generate multiple sets of obstacle avoidance area voxel meshes corresponding to different motion parameters offline, and directly match and switch the corresponding obstacle avoidance area according to the current motion parameters during online operation.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the dynamic region collision detection method based on the voxelization model as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the dynamic region collision detection method based on the voxelization model as described in any one of claims 1 to 7.

Images