A dynamic obstacle avoidance method for a mechanical arm based on PRM and timing prediction driving DLite

Abstract

This invention discloses a dynamic obstacle avoidance method for robotic arms based on Probability of Recognition (PRM) and temporal prediction-driven DLite algorithm. First, it acquires workspace perception data and constructs a 3D occupancy probability grid map. Then, based on the 3D occupancy probability grid map, it determines a 3D free-form configuration space and generates an offline static global topology map. When a dynamic obstacle is detected, the real-time pose of the obstacle is determined based on the 3D occupancy probability grid map. Combined with the offline static global topology map, the trajectory sequence and four-dimensional spatiotemporal risk penalty edge weights of the obstacle are determined. Based on the DLite algorithm, local path incremental reconstruction is performed, and the safe discrete path is updated. Finally, the safe discrete path is smoothed, and the final obstacle avoidance trajectory is issued to the underlying driver, driving the robotic arm to execute actions according to the final obstacle avoidance trajectory. This method enables the robotic arm to make decisions to avoid collisions when facing dynamic obstacles, thereby achieving dynamic obstacle avoidance in 3D space.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, and in particular to a dynamic obstacle avoidance method for robotic arms based on PRM and time-series prediction-driven DLite. Background Technology

[0002] With the rapid development of intelligent manufacturing and human-machine collaboration technologies, robotic arms are increasingly being used in industrial production, medical assistance, and services. Obstacle avoidance path planning, as a core bottleneck for robots to achieve autonomous operation, has always been a focus of industry attention. Based on environmental characteristics, obstacle avoidance planning can be categorized into three main types: static obstacle avoidance, semi-dynamic obstacle avoidance, and fully dynamic environment obstacle avoidance. As application scenarios continue to expand, robotic arms need to frequently cope with highly complex dynamic disturbances, such as random pedestrian traffic and the intervention of mobile devices. However, existing research on dynamic obstacle avoidance largely focuses on reactive avoidance in the pure spatial dimension, and traditional sequential decoupled trajectory planning methods often struggle to handle high-frequency coupled spatiotemporal conflicts. In practical applications, these planners often experience a surge in computational load when facing sudden environmental changes, failing to meet the millisecond-level rapid response requirements; simultaneously, they are prone to causing the robotic arm's trajectory to become zigzag and jittery, or even falling into local deadlock.

[0003] Based on this, in order to overcome the limitations of traditional algorithms in terms of computing power and time and space, and to enable planners to balance rapid response, efficient time and space avoidance, and smooth underlying trajectory in complex dynamic environments, some improved obstacle avoidance methods have emerged. For example, CN116117822A proposes an RRT robotic arm trajectory planning method based on sampling of the non-obstacle space probability potential field; however, it relies on the random tree expansion of the RRT algorithm, making it difficult to handle path planning in narrow spaces and dynamic obstacle avoidance planning. CN121857774A proposes a robot dynamic obstacle avoidance method that incorporates action prediction and Markov game theory. It integrates the robot's own state and action set with the state and action set of the dynamic target into a joint matrix for calculation, but this method requires higher computing power. CN121413824A discloses a method that integrates topology map and D... The Lite algorithm performs path planning to ensure dynamic obstacle avoidance, but it employs a D-axis approach in local planning when facing obstacle collisions. The scheme combining the Lite algorithm and the DWA algorithm cannot consider the movement trajectory of obstacles when facing dynamic obstacles. Furthermore, due to the discontinuity of the physical instructions between the two algorithms, there is a possibility of getting stuck in local deadlock or producing a tortuous trajectory. Summary of the Invention

[0004] To address some or all of the problems in existing technologies and to overcome the limitations of existing path planning methods in solving dynamic obstacle avoidance in high-dimensional spaces, the first aspect of this invention provides a dynamic obstacle avoidance method for robotic arms based on PRM and temporal prediction-driven DLite, comprising: Acquire workspace perception data and construct a 3D occupancy probability grid map; Based on the three-dimensional occupancy probability grid map, a three-dimensional free configuration space is determined, and an offline static global topology map is generated based on the three-dimensional free configuration space; When a dynamic obstacle is detected, the real-time pose of the dynamic obstacle is determined based on the three-dimensional occupancy probability grid map. Combined with the offline static global topology map, the trajectory sequence of the dynamic obstacle and the four-dimensional spatiotemporal risk penalty edge weight are determined. Based on D The Lite algorithm performs incremental reconstruction of local paths and updates safe discrete paths; The safe discrete path is smoothed, the final obstacle avoidance trajectory of the underlying drive is issued, and the robotic arm is driven to perform actions according to the final obstacle avoidance trajectory.

[0005] Furthermore, an offline static global topology graph is generated based on a probabilistic state machine-driven 3D hybrid bias sampling strategy, including: Obtain the three-dimensional workspace boundary of the robotic arm and set the safety margin for each dimension; Different sampling engines are activated by a random decision factor, wherein the sampling engines include global uniform sampling, obstacle boundary Gaussian sampling, and start-end corridor sampling; In the three-dimensional free configuration space, a three-dimensional hybrid bias sampling strategy is adopted to generate an offline static global topology map.

[0006] Furthermore, the random decision factor of the globally uniform sampling is less than 0.4, and its proportion is 40%. The random decision factor for the Gaussian sampling of the obstacle boundary ranges from 0.4 to 0.7, and accounts for 30% of the total. The random decision factor for the sampling of the origin and destination corridors is greater than 0.7, and its proportion is 30%.

[0007] Furthermore, generating an offline static global topology graph also includes: Construct a simplified algebraic collision detection model; After sampling, interference nodes are eliminated using the algebraic collision detection model, and nearest neighbor connections are made to generate an undirected connected graph with static risk avoidance awareness.

[0008] Furthermore, the simplified algebraic collision detection model includes: Map the workspace and construct a mapping between the 3D workspace mesh and the sample points of the robotic arm joint configuration space; Simplify the cubic mesh representing obstacles and transform it into a sphere; The radius of the cylinder representing the robotic arm link is superimposed onto the corresponding sphere to obtain an enlarged new sphere, thereby simplifying the collision detection between the robotic arm link and the obstacle into the determination of the positional relationship between the straight line segment and the sphere.

[0009] Furthermore, a long short-term memory network is used to determine the trajectory sequence of the dynamic obstacle, wherein the trajectory sequence includes the spatiotemporal three-dimensional coordinate sequence and timestamp set of the dynamic obstacle in the next K prediction periods, where K is a natural number.

[0010] Furthermore, the four-dimensional spacetime risk penalty boundary weights are determined to include: For each candidate edge in the offline static global topology graph, determine the expected physical time for the robotic arm to reach the midpoint of the candidate edge; By combining the trajectory sequence of the dynamic obstacles, the expected spatial distance and the time difference of arrival are determined, and a composite total edge weight cost function is constructed to determine the four-dimensional spatiotemporal risk penalty edge weight.

[0011] Furthermore, local path incremental reconstruction includes: When the local edge weights fluctuate and shift, through D The Lite algorithm re-estimates the local look-ahead cost of associated nodes; If the actual cost of a node is inconsistent with the local look-ahead cost, the cost is corrected, and the difference in total revenue between the new candidate path and the currently executing historical path is calculated. When the difference in total revenue is higher than the hysteresis safety threshold, a path switch is performed, and the increment is updated.

[0012] Furthermore, the safe discrete path is smoothed using a line-of-sight shortcut method and a cubic Catmull-Rom spline high-order smoothing method.

[0013] Based on the aforementioned dynamic obstacle avoidance method for robotic arms, the second aspect of this invention provides a method based on PRM and time-series prediction driving D. Lite's robotic arm dynamic obstacle avoidance system includes: The environmental perception and spatial modeling module is used to acquire workspace perception data, construct a 3D occupancy probability grid map, and output the real-time pose of 3D free-form space and dynamic obstacles. The global topology network construction module is used to generate an offline static global topology graph based on a three-dimensional free configuration space, resulting in a low-level collision-free topology skeleton. The spatiotemporal prediction and cost mapping module is used to determine the trajectory sequence of the dynamic obstacle based on the real-time pose of the dynamic obstacle and the underlying collision-free topological skeleton, and to determine the four-dimensional spatiotemporal risk penalty edge weights. The incremental replanning and stability control module is used to perform incremental reconstruction of local paths based on dynamic penalty edge weights and update safe discrete paths. The trajectory smoothing and control execution module is used to smooth the safe discrete path and drive the robotic arm to perform actions.

[0014] This invention provides a dynamic obstacle avoidance method for robotic arms based on PRM and temporal prediction-driven DLite, which combines global static topology preconstruction with local dynamic spatiotemporal incremental replanning. Based on environmental perception, an offline global topology navigation map (PRM) is first constructed using 3D hybrid bias sampling to address connectivity issues in narrow passages. When a dynamic obstacle intervenes, its future trajectory is calculated using a spatiotemporal prediction model (such as LSTM), and the collision risk is transformed into a soft and hard constraint penalty cost including a time decay factor and mapped onto the topology map. Subsequently, based on these affected dynamic edge weights, a DLite-driven method with exponential smoothing and hysteresis anti-oscillation constraints is adopted. The Lite algorithm performs incremental replanning, effectively eliminating high-frequency jitter while maintaining fast response; finally, high-order spline interpolation outputs a smooth and accurate terminal execution trajectory. This dynamic obstacle avoidance method for robotic arms enables them to make decisions on avoiding collisions when facing dynamic obstacles, thus solving the dynamic obstacle avoidance problem in three-dimensional space. Attached Figure Description

[0015] To further illustrate the above and other advantages and features of the various embodiments of the present invention, a more specific description of the various embodiments of the present invention will be presented with reference to the accompanying drawings. It is to be understood that these drawings depict only typical embodiments of the invention and are therefore not intended to limit its scope. In the drawings, identical or corresponding parts will be indicated by identical or similar reference numerals for clarity.

[0016] Figure 1 This diagram illustrates a flow chart of a dynamic obstacle avoidance method for a robotic arm based on PRM and time-series prediction-driven DLite according to an embodiment of the present invention. Figure 2 A flowchart illustrating a method for constructing an offline global topology navigation map according to an embodiment of the present invention is shown. Figure 3 A flowchart illustrating a spatiotemporal prediction and cost determination method according to an embodiment of the present invention is shown. Figure 4 A flowchart illustrating an embodiment of the incremental replanning method of the present invention is shown. Figure 5 This invention illustrates an embodiment of a PRM-based and timing prediction-driven D... A schematic diagram of the dynamic obstacle avoidance system for the Lite robotic arm. Detailed Implementation

[0017] In the following description, the invention is described with reference to various embodiments. However, those skilled in the art will recognize that the embodiments may be practiced without one or more specific details or in conjunction with other alternatives and / or additional methods. In other instances, well-known operations are not shown or described in detail so as not to obscure the inventive points of the invention. Similarly, for illustrative purposes, specific configurations are set forth to provide a comprehensive understanding of embodiments of the invention. However, the invention is not limited to these specific details.

[0018] In this specification, references to "an embodiment" or "this embodiment" mean that a particular feature, structure, or characteristic described in connection with that embodiment is included in at least one embodiment of the invention. The phrase "in one embodiment" appearing throughout this specification does not necessarily refer to the same embodiment in all instances.

[0019] It should be noted that the embodiments of the present invention describe the method steps in a specific order; however, this is only for illustrating the specific embodiment and not for limiting the order of the steps. On the contrary, in different embodiments of the present invention, the order of the steps can be adjusted according to actual needs.

[0020] In this invention, the modules of the system according to the invention can be implemented using software, hardware, firmware, or a combination thereof. When a module is implemented using software, its function can be implemented through computer program flow. For example, the module can be implemented using code segments (such as code segments in languages ​​like C and C++) stored in a storage device (such as a hard disk, memory, etc.), wherein the corresponding function of the module can be implemented when the code segment is executed by a processor. When a module is implemented using hardware, its function can be implemented by setting a corresponding hardware structure. For example, the module's function can be implemented by hardware programming a programmable device such as a field-programmable gate array (FPGA), or by designing an application-specific integrated circuit (ASIC) that includes multiple transistors, resistors, capacitors, and other electronic devices. When a module is implemented using firmware, the module's function can be written into a read-only memory such as an EPROM or EEPROM in the form of program code, and the corresponding function of the module can be implemented when the program code is executed by a processor.

[0021] To achieve dynamic obstacle avoidance planning for robotic arms in dynamic obstacle environments, this invention provides a dynamic obstacle avoidance method for robotic arms based on PRM and temporal prediction-driven DLite. This method integrates an offline global topology navigation map (PRM) with a temporal prediction-driven DLite. The integration of Lite technology allows the robotic arm to autonomously plan its path when facing dynamic obstacles, thus avoiding collisions and solving the problem of dynamic obstacle avoidance in three-dimensional space.

[0022] The technical solution of the present invention will be further described below with reference to the accompanying drawings of the embodiments.

[0023] Figure 1 This diagram illustrates a flow chart of a dynamic obstacle avoidance method for a robotic arm based on PRM and time-series prediction-driven DLite, according to an embodiment of the present invention. Figure 1 As shown, a dynamic obstacle avoidance method for a robotic arm based on PRM and time-series prediction-driven DLite includes: First, in step 101, a three-dimensional occupancy probability grid map is constructed. Workspace perception data is acquired, and a three-dimensional occupancy probability grid map is constructed. In one embodiment of the invention, the OctoMap method is used to map the workspace, thereby constructing a precise mapping between the three-dimensional workspace grid and the sample points of the robotic arm joint configuration space. Through this precise mapping, collision detection can be performed, thus ensuring the safety of the robotic arm during movement. OctoMap is an efficient probabilistic 3D mapping framework based on an octree. Each node in the octree has three states: occupied, uncertain, and unoccupied. These three states can be represented by floating-point numbers between 0 and 1. For example, a value of 0.5 represents uncertain, and a value between 0.5 and 1 represents occupied, with a larger value indicating a higher probability of occupation. A value between 0 and 0.5 represents unoccupied, with a smaller value indicating a lower probability of unoccupied. Based on this, when the child nodes of a node have the same state (i.e., all are occupied, uncertain, or unoccupied), these nodes can be pruned, leaving only one parent node to save storage space.

[0024] For observations of dynamic environments, the smallest unit of the OctoMap, the square, may sometimes be observed as occupied, and sometimes be observed as unoccupied due to changes in the environment. Assuming that... At what time, the observed data is The information recorded in each leaf node can then be represented as: , in, This represents the prior probability of whether a leaf node is occupied before observation, and its preferred value is 0.5.

[0025] The information recorded in the leaf nodes can be represented using logarithmic probability notation as follows: , in, ; Next, in step 102, an offline static global topology map is generated. Based on the three-dimensional occupancy probability grid map, a three-dimensional free configuration space is determined, and an offline static global topology map is generated based on the three-dimensional free configuration space. In one embodiment of the present invention, a three-dimensional hybrid bias sampling strategy driven by a probabilistic state machine is used to generate the offline static global topology map. Figure 2 This is a flowchart illustrating a method for constructing an offline global topology navigation map according to an embodiment of the present invention. Figure 2 As shown, firstly, based on the three-dimensional occupancy probability grid map and the start and end coordinates, the three-dimensional workspace boundary of the robotic arm is determined, and a safety margin for each dimension is set, such as an inward offset of several units, to define the effective feasible region. Next, random decision factors are generated through modules such as the system's internal pseudo-random number generator. To activate different sampling engines, a three-dimensional hybrid bias sampling strategy is finally adopted to construct the topology navigation map offline. This three-dimensional hybrid bias sampling strategy abandons single global uniform sampling in the configuration space, instead dividing it into three parallel bias sampling channels: global uniform sampling, obstacle boundary Gaussian sampling, and start-end corridor sampling. The global uniform sampling has a random decision factor less than 0.4 and accounts for 40% of the sampling; it is used to perform standard uniform distribution sampling within the feasible region boundary to establish the basic connectivity skeleton. The obstacle boundary Gaussian sampling has a random decision factor ranging from 0.4 to 0.7 and accounts for 30% of the sampling; it is used to randomly extract targets from the static obstacle set, expand a specific safety range outward in three dimensions based on its bounding box (e.g., 0.6 to 1.0 unit length), and forcibly drop nodes within this expansion interval, following a zero-mean Gaussian distribution with a specified variance. To increase the node density in narrow passages, a random decision factor greater than 0.7 is used in the sampling of the start and end corridors, accounting for 30% of the total sampling. This factor is used to extract the coordinates of the start and end points, perform linear interpolation, and apply a limited-amplitude three-dimensional random noise perturbation to the space using an orthogonal method to generate high-density guide corridor nodes. The hybrid bias sampling strategy synchronously triggers multiple sampling modules through a pseudo-random factor, achieving adaptive tilting of sampling resources towards topologically narrow passages and guide areas. It actively generates high-density nodes in the external region of static obstacles, thereby improving the high success rate of global optimization from the bottom layer.

[0026] In one embodiment of the present invention, such as Figure 2As shown, after sampling, interference nodes are further eliminated using an algebraic collision detection model, and nearest neighbor connections are established to generate an undirected connected graph with static risk avoidance awareness. Specifically, in OctoMap, obstacles are represented as cubic meshes. To simplify the calculation process and improve detection efficiency, the cubic mesh representing obstacles can be simplified by using spherical bounding boxes, transforming the collision detection problem into determining the positional relationship between cylinders and spheres in space. In one embodiment of the invention, to further simplify the collision detection calculation, the radius of the cylinder can be superimposed on the relevant sphere to form an enlarged new sphere, simplifying the collision detection between the robotic arm link and the obstacle to determining the positional relationship between a straight line segment and a sphere, thus obtaining a simplified algebraic collision detection model. The coordinates of the two ends of the simplified cylindrical straight line segment are respectively... and Then the straight line segment It can be represented as: , For a sphere, the center The coordinates are , radius is The endpoints of the line segment are The distances from the center of the circle to the endpoints of the line segment are respectively... and This indicates that the distance from the center of the circle to the perpendicular line is... If a point is defined, then the intersection relationship between a line segment in space and a sphere can be represented by four cases: when , as well as When the line segment does not intersect with the obstacle, there is no physical collision interference; when or When the straight line intersects with the obstacle, there is physical collision interference; when , but If at this time Point and Point located at If the points are on the same side, the straight line segment does not intersect with the obstacle, and there is no physical collision interference; when , but If at this time Point and Point located at If the points are on opposite sides, the straight line segment intersects with the obstacle, resulting in physical collision interference.

[0027] Candidate nodes with physical collision interference are discarded, while candidate nodes without physical collision interference are retained and moved to the legal node pool. When the number of candidate nodes in the legal node pool reaches a preset value, nearest neighbor connections are established, the 3D occupancy probability grid map and the origin and destination coordinates are updated, and an undirected connected graph with static risk avoidance awareness is generated. In one embodiment of the invention, nearest neighbor connections are established for legal nodes based on an adaptive kd-tree. Next, in step 103, spatiotemporal prediction and cost determination. When a dynamic obstacle is detected, the real-time pose of the dynamic obstacle is determined based on the three-dimensional occupancy probability grid map. Combined with the offline static global topology map, the trajectory sequence of the dynamic obstacle and the four-dimensional spatiotemporal risk penalty edge weights are determined. Figure 3 This diagram illustrates a flow chart of a spatiotemporal prediction and cost determination method according to an embodiment of the present invention. Figure 3 As shown, in one embodiment of the present invention, when a dynamic obstacle is detected, a Long Short-Term Memory (LSTM) network is used to predict the trajectory of the dynamic obstacle based on its coordinates and the edge set of the underlying topology graph. The output is a sequence of the obstacle's spatiotemporal three-dimensional coordinates and a set of timestamps over the next K prediction periods, where K is a natural number. Next, candidate edges affected by the obstacle are extracted from the topology graph. For each candidate edge in the PRM topology graph... Calculate the expected physical time for the robotic arm to reach the midpoint of the edge. By combining the predicted trajectory sequence of dynamic obstacles, the expected spatial distance between the midpoint of the candidate edge and the predicted coordinates of the obstacle is calculated. Arrival time difference, which is the time difference between the expected physical time and the predicted timestamp of the obstacle. And construct a composite total edge weight cost function: , in, Candidate edges The fundamental physical path Euclidean cost and static environment penalty; Candidate edges The hard constraint is triggered when the spatial distance is less than the safe envelope radius and the time windows coincide, thus forcing the hard constraint to take the value of infinity. Candidate edges Soft constraints, in which For spatial distance, For time difference, when the spatial distance is close but the time difference is greater than the threshold, only nonlinear spatiotemporal soft penalty is accumulated, giving the robotic arm the ability to avoid collision risk actions.

[0028] The composite total edge weight cost function is used to assign costs to candidate edges. When the spatial distance is greater than the absolute safety radius, the candidate edge is considered safe and assigned a basic physical path cost. When the spatial distance is less than the absolute safety radius, it is further determined whether the time difference is less than the critical overlap window. If so, a hard constraint is triggered, indicating a fatal collision and assigning an infinite cost. If not, a soft constraint is triggered, calculating a soft risk penalty based on time decay, and adding the calculated penalty to the basic cost to obtain a dynamic topological edge containing spatiotemporal weights. In one embodiment of the invention, when determining collision risk, a dynamic decay factor characterizing the physical arrival time difference is calculated and introduced. By nonlinearly fusing spatial interference gradient penalty with time factor, four-dimensional spatiotemporal risk is accurately reduced to one-dimensional search cost.

[0029] Based on the composite total edge weight cost function, a time decay penalty function is introduced to significantly reduce the penalty cost when the robotic arm and the obstacle have spatial overlap but there is a significant time difference in the time of arrival at the intersection point. This enables the robotic arm to make a decision to avoid obstacles when faced with dynamic interference, effectively avoiding problems such as sudden stop waiting or ineffective detour deadlock. Next, in step 104, local path incremental reconstruction is performed. Based on D The Lite algorithm performs incremental reconstruction of local paths and updates safe discrete paths. In one embodiment of the invention, when the spatiotemporal risk mapping in the aforementioned steps leads to the lower-level local edge weights... When fluctuations occur, D The Lite algorithm only considers the local look-ahead cost of the affected and associated nodes. A reassessment will be conducted. Figure 4 A flowchart illustrating an embodiment of the incremental replanning method of the present invention is shown. Figure 4 As shown, in one embodiment of the present invention, based on the changed spatiotemporal penalty edge weights, the endpoints corresponding to the topological edges whose edge weights have changed are first located. To prevent high-frequency divergence of the planned path due to low-level perception noise, in one embodiment of the present invention, a low-pass filter based on exponentially moving weighted (EMA) edge weights is applied beforehand. That is, a digital filter is introduced before the edge weight input priority queue. Specifically, exponentially moving weighted smoothing is introduced before positioning, and the correction cost is calculated using a smoothing function. , in, This is a smoothing factor.

[0030] Next, the local forward-looking cost of the short point is reassessed based on the cost of the adjacent node. and the actual cost of the node Comparison, if the actual cost of the node and If the results are consistent, the node is skipped without triggering a network-level update; otherwise, it is pushed into a priority queue for reverse wavefront propagation calculation to generate candidate replanning paths. The difference between the total reward of the new candidate path and the reward of the currently executing path is calculated, and incremental updates are performed to obtain the updated safe discrete path. To prevent robot chassis oscillation, in one embodiment of the invention, a reward evaluation threshold of hysteresis gain constant is introduced post-processed. That is, a hysteresis mechanism is introduced after determining the reward difference, setting a strict relative optimization threshold and target distance locking line for path switching. By introducing a hysteresis safety threshold, the difference between the total reward of the new candidate path and the currently executing historical path is calculated. And only when the total revenue of the new path significantly exceeds the hysteresis safety threshold, i.e. Only when the time is right will a path switching action be confirmed, thus ensuring the absolute stability of physical execution from the root. By introducing filtering and hysteresis anti-oscillation threshold mechanisms, high-frequency fluctuations in local costs can be effectively smoothed. While ensuring path safety, the probability of high-frequency path switching is reduced, thereby improving the steady-state level of servo motor control and reducing invalid jitter of the underlying actuators; Finally, in step 105, the path is smoothed. The updated safe discrete path is smoothed, the final obstacle avoidance trajectory of the underlying driver is issued, and the robotic arm is driven to perform actions based on the final obstacle avoidance trajectory. In one embodiment of the invention, a line-of-sight shortcut method and a cubic Catmull-Rom spline high-order smoothing method are employed. Specifically, for D... The discrete polyline path with abrupt turns output by Lite is first processed using a line-of-sight shortcut operation. This involves a greedy algorithm combined with the aforementioned collision detection mechanism to skip redundant nodes and directly connect distant nodes, thus eliminating meaningless turning points. Then, the simplified skeleton control node set is processed... Mathematical interpolation fitting was performed using cubic Catmull-Rom spline curves. Adjacent control points... and curve segments between The parametric matrix equation is: , Cubic Catmull-Rom spline curves enable the generated smooth trajectory to accurately pass through safety control nodes, avoiding the risk of secondary collisions caused by convex hull interpolation offsets, while ensuring the first-order continuity of the trajectory tangential velocity. This eliminates actuator vibration, which helps the robotic arm to smoothly pass through complex obstacle areas.

[0031] Based on the dynamic obstacle avoidance method of the robotic arm described above. Figure 5 This invention illustrates an embodiment of a PRM-based and timing prediction-driven D... A schematic diagram of the dynamic obstacle avoidance system for the Lite robotic arm. (See diagram below.) Figure 5 As shown, a method based on PRM and time-series prediction-driven D Lite's robotic arm dynamic obstacle avoidance system includes an environmental perception and spatial modeling module 501, a global topology network construction module 502, a spatiotemporal prediction and cost mapping module 503, an incremental replanning and stability control module 504, and a trajectory smoothing and control execution module 505. The environmental perception and spatial modeling module 501 acquires workspace perception data and constructs a 3D occupancy probability grid map, outputting the real-time poses of the 3D free-form space and dynamic obstacles. The global topology network construction module 502 generates an offline static global topology map based on the 3D free-form space, outputting a bottom-level collision-free topology skeleton. The spatiotemporal prediction and cost mapping module 503 determines the trajectory sequence of the dynamic obstacles based on their real-time poses and the bottom-level collision-free topology skeleton, and determines the four-dimensional spatiotemporal risk penalty edge weights, outputting the dynamic edge weights of the topology affected by collisions. The incremental replanning and stability control module 504 performs incremental local path reconstruction and updates safe discrete paths based on dynamic penalty edge weights and a hysteresis filtering mechanism. The trajectory smoothing and control execution module 505 is used to smooth the safe discrete path using methods such as high-order spline interpolation algorithms and drive the robotic arm to perform actions.

[0032] Although various embodiments of the invention have been described above, it should be understood that they are presented by way of example only and not as limitations. It will be apparent to those skilled in the art that various combinations, modifications, and alterations can be made without departing from the spirit and scope of the invention. Therefore, the breadth and scope of the invention disclosed herein should not be limited by the exemplary embodiments disclosed above, but should be defined solely by the appended claims and their equivalents.

Claims

1. A method for dynamic obstacle avoidance of a DLite-based robotic arm driven by PRM and timing prediction, characterized in that, include: Acquire workspace perception data and construct a 3D occupancy probability grid map; Based on the three-dimensional occupancy probability grid map, a three-dimensional free configuration space is determined, and an offline static global topology map is generated based on the three-dimensional free configuration space; When a dynamic obstacle is detected, the real-time pose of the dynamic obstacle is determined based on the three-dimensional occupancy probability grid map. Combined with the offline static global topology map, the trajectory sequence of the dynamic obstacle and the four-dimensional spatiotemporal risk penalty edge weight are determined. Based on D The Lite algorithm is used to perform local path incremental reconstruction and update the safe discrete path. The safe discrete path is smoothed, the final obstacle avoidance trajectory of the underlying drive is issued, and the robotic arm is driven to perform actions according to the final obstacle avoidance trajectory.

2. The method of claim 1, wherein, A probabilistic state machine-driven 3D hybrid bias sampling strategy is used to generate the offline static global topology graph, which includes: Obtain the three-dimensional workspace boundary of the robotic arm and set the safety margin for each dimension; Different sampling engines are activated by a random decision factor, wherein the sampling engines include global uniform sampling, obstacle boundary Gaussian sampling, and start-end corridor sampling; In the three-dimensional free configuration space, a three-dimensional hybrid bias sampling strategy is adopted to generate an offline static global topology map.

3. The method of claim 2, wherein, The random decision factor of the global uniform sampling is less than 0.4, and its proportion is 40%. The random decision factor for the Gaussian sampling of the obstacle boundary ranges from 0.4 to 0.7, and accounts for 30% of the total. The random decision factor for the sampling of the origin and destination corridors is greater than 0.7, and its proportion is 30%. 4.The method of claim 2, wherein, Generating an offline static global topology graph also includes: Construct a simplified algebraic collision detection model; After sampling, interference nodes are eliminated using the algebraic collision detection model, and nearest neighbor connections are made to generate an undirected connected graph with static risk avoidance awareness.

5. The method of claim 4, wherein, Constructing a simplified algebraic collision detection model includes: Map the workspace and construct a mapping between the 3D workspace mesh and the sample points of the robotic arm joint configuration space; Simplify the cubic mesh representing obstacles and transform it into a sphere; The radius of the cylinder representing the robotic arm link is superimposed onto the corresponding sphere to obtain an enlarged new sphere, thereby simplifying the collision detection between the robotic arm link and the obstacle into the determination of the positional relationship between the straight line segment and the sphere.

6. The method of claim 1, wherein, The trajectory sequence of the dynamic obstacle is determined by a long short-term memory network, wherein the trajectory sequence includes the spatiotemporal three-dimensional coordinate sequence and timestamp set of the dynamic obstacle in the next K prediction periods, where K is a natural number.

7. The method of claim 1, wherein, The four-dimensional spatiotemporal risk penalty edge weights are determined through the following steps: For each candidate edge in the offline static global topology graph, determine the expected physical time for the robotic arm to reach the midpoint of the candidate edge; By combining the trajectory sequence of the dynamic obstacles, the expected spatial distance and the difference between arrival time are determined, and a composite total edge weight cost function is constructed to determine the four-dimensional spatiotemporal risk penalty edge weight. 8.The method of claim 1, wherein, Local path incremental reconstruction includes: When the local edge weights fluctuate and shift, through D The Lite algorithm re-estimates the local look-ahead cost of associated nodes; If the actual cost of a node is inconsistent with the local look-ahead cost, the cost is corrected, and the difference in total revenue between the new candidate path and the currently executing historical path is calculated. When the difference in total revenue is higher than the hysteresis safety threshold, a path switch is performed, and the increment is updated.

9. The method of claim 1, wherein, The safe discrete path is smoothed using a line-of-sight shortcut method and a cubic Catmull-Rom spline high-order smoothing method.

10. A method of driving a D The mechanical arm dynamic obstacle avoidance system of the ROS include: The environmental perception and spatial modeling module is configured to acquire workspace perception data, construct a 3D occupancy probability grid map, and output the real-time pose of 3D free-form space and dynamic obstacles. The global topology network construction module is configured to generate an offline static global topology graph based on a three-dimensional free configuration space, resulting in a low-level collision-free topology skeleton. The spatiotemporal prediction and cost mapping module is configured to determine the trajectory sequence of the dynamic obstacle and the four-dimensional spatiotemporal risk penalty edge weights based on the real-time pose of the dynamic obstacle and the underlying collision-free topology skeleton. The incremental replanning and stability control module is configured to perform incremental reconstruction of local paths based on dynamic penalty edge weights and update safe discrete paths. A trajectory smoothing and control execution module is configured to smooth the safe discrete path and drive the robotic arm to perform actions.

Images