A dynamic path planning method and system for cooperative work of double industrial robots

Abstract

The application relates to a dynamic path planning method and system for cooperative work of double industrial mechanical arms, which comprises the following steps: firstly, an initial feasible path connecting starting points and target endpoints of the mechanical arms is generated by using a global path search algorithm; secondly, key nodes of the initial path are extracted and estimated occupation time windows are calculated to form space-time occupation information which is published to a shared state pool; then, each mechanical arm retrieves space-time occupation information of other mechanical arms from the shared state pool and takes the information as a dynamic constraint to generate a candidate trajectory segment set meeting cooperative avoidance requirements; finally, a dynamic window method is used to sample and evaluate the candidate trajectory segments in a speed-angle velocity space, and an optimal trajectory segment is selected as a current execution path. The application can improve path planning efficiency in a dynamic scene, eliminate collision risks in cooperative work and realize beat synchronization, thereby effectively solving the problems of low search efficiency, high collision risk and asynchronous beats in double-arm cooperative work.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for industrial robots, and in particular to a dynamic path planning method and system for collaborative operation of two industrial robotic arms. Background Technology

[0002] In the field of modern intelligent manufacturing, the collaborative operation of dual or multiple robotic arms has become a key technology for improving production flexibility and efficiency. Path planning, as the core component of robotic arm motion control, directly determines the safety and cycle time of the entire collaborative system.

[0003] Currently, path planning technologies for dual-arm collaborative operations have the following limitations: 1. Traditional offline programming methods pre-set fixed trajectories for the robotic arms. This approach lacks the ability to perceive and adapt to environmental changes. Once unexpected dynamic obstacles appear in the working environment or the workstation shifts slightly, collisions or operation failures are easily triggered, failing to meet the needs of flexible production. 2. Some online planning methods use master-slave control or fixed priority strategies for conflict management, where one robotic arm (slave arm) completely avoids the other robotic arm (master arm). While this static coordination mechanism can avoid collisions, it often leads to unnecessary waiting or large-scale detours for the slave arm, wasting time and energy, reducing the overall system's collaborative operation efficiency, and potentially causing deadlocks in complex tasks. 3. Other reactive planning methods based on local obstacle avoidance, such as the artificial potential field method, although fast in response, are prone to getting trapped in local optima due to the lack of global path guidance. This can cause path oscillations or failure to reach the target point in areas with dense obstacles.

[0004] Therefore, existing technologies generally lack a planning framework that can balance global path optimization and local dynamic response capabilities. In particular, they are significantly inadequate in achieving proactive and predictive spatiotemporal coordination between the two arms, making it difficult to maximize the utilization of the shared workspace and achieve tight work cycle synchronization while ensuring safety. Summary of the Invention

[0005] To help solve the technical problems existing in the prior art, the present invention provides a dynamic path planning method and system for collaborative operation of two industrial robotic arms, which can realize dynamic, conflict-free collaborative operation and synchronized operation rhythm of the two arms in a shared space, thereby effectively solving the problems of low search efficiency, high collision risk and asynchronous operation rhythm in collaborative operation of two arms.

[0006] This invention discloses a dynamic path planning method for collaborative operation of two industrial robotic arms, comprising the following steps: (a) Global path generation steps: Obtain the starting point, target endpoint and environmental map information of each industrial robotic arm, and use a global path search algorithm based on random sampling to search within the configuration space to generate an initial feasible path connecting the starting point and the target endpoint; (b) Spatiotemporal information sharing step: extract the critical path nodes in the initial feasible path, and calculate the estimated time window corresponding to each critical path node in combination with the preset running speed, thereby forming spatiotemporal occupancy information, and publish the spatiotemporal occupancy information to the preset shared state pool; (c) Collaborative obstacle avoidance planning steps: Each industrial robotic arm retrieves the spatiotemporal occupancy information of other robotic arms from the shared state pool at a preset period, converts the retrieved information into dynamic spatial conflict constraints, and combines the current environmental obstacle distribution information to generate a set of candidate trajectory segments that meet the collaborative avoidance requirements in the neighborhood of the initial feasible path. (d) Local path optimization steps: A rolling window is constructed using the dynamic window method (DWA). The candidate trajectory segment set is sampled in the velocity-angular velocity space, and the sampled trajectory is evaluated in real time through a multi-objective cost function. The optimal trajectory segment is selected as the current execution path to achieve dynamic obstacle avoidance and operation rhythm synchronization of the two arms in the shared space.

[0007] It is understood that the technical solution of this invention aims to construct a closed-loop control process of global planning guidance, information sharing coordination, and local dynamic optimization to ensure that the two arms can operate in parallel efficiently and safely within a shared space. The execution process of this method begins with (a) the global path generation step, which is the foundation of the entire planning. The system first obtains the specific starting point, target endpoint, and environmental map information containing static obstacles (such as workbenches, fixtures, equipment frames, etc.) for each robotic arm. Based on this information, a global path search algorithm based on random sampling explores the configuration space of the robotic arm (i.e., the multi-dimensional space composed of all joint angles), with the goal of quickly generating a collision-free initial feasible path from the starting point to the target endpoint. This path provides a macroscopic and directional guiding benchmark for the overall movement of the robotic arm. Subsequently, the system enters the spatiotemporal information sharing step, which is the key to realizing "mutual perception" and collaboration between the two arms. The system does not directly share the entire geometric path generated in the previous step, but processes the initial feasible path and extracts the key path nodes that have a decisive influence on the movement. Combining the robot arm's preset operating speed or dynamics model, the system can calculate the estimated time window for the robot arm to arrive at, occupy, and leave each critical path node and its vicinity. Thus, a purely geometric path is transformed into four-dimensional spatiotemporal occupancy information containing both spatial occupancy range and temporal occupancy interval. This timestamped information is then published to a preset shared state pool accessible to all robot arms, serving as a declaration of the robot arm's future motion intention. Next, (c) the collaborative obstacle avoidance planning step transforms the motion intentions of other robot arms into its own behavioral constraints. Each robot arm actively retrieves spatiotemporal occupancy information published by other robot arms from the shared state pool at a preset short cycle. This information is interpreted as a dynamic, time-varying virtual obstacle in the current robot arm's planning module. The planning module combines these dynamic spatial conflict constraints with static obstacle information in the current environment, performing a local search within the vicinity of the initial feasible path generated in step (a), generating a set of multiple candidate trajectory segments that meet the collaborative avoidance requirements. These candidate trajectory segments can all bypass static obstacles and the predetermined trajectories of other robot arms within a local range. Finally, the system executes step (d) local path optimization to make the optimal real-time decision from numerous candidate trajectories. This step employs the Dynamic Window Method (DWA) to construct a forward prediction rolling window. Within the currently reachable velocity-angular velocity space of the robotic arm, the set of candidate trajectory segments generated in step (c) is densely sampled. The system uses a multi-objective cost function to quantitatively evaluate multiple performance indicators such as safety, smoothness, and execution efficiency for each sampled trajectory in real time. Ultimately, the trajectory segment with the optimal cost function score is selected as the path to be executed immediately within the current control cycle.By repeatedly executing steps (c) and (d) at a high frequency (i.e. the update frequency of the scrolling window), the system achieves continuous response to the movements of other robotic arms and dynamic changes in the environment, thereby ensuring dynamic obstacle avoidance by both arms while achieving close synchronization of the work cycle.

[0008] According to a dynamic path planning method for collaborative operation of two industrial robotic arms of the present invention, in step (a), the global path search algorithm is an improved fast expanding random tree (RRT) algorithm that introduces a gravity bias and adaptive step size mechanism, specifically including: a gravity bias sampling mechanism: during the search tree expansion iteration process, with a preset probability... Force the generation of sampling points in the direction of the target point, and use the gravity bias vector. The gravity bias vector is calculated to guide the search tree to converge toward the target direction as follows:

[0009] In the formula, Indicates the coordinates of the current node. Indicates the coordinates of the target point. This represents the preset gravity weighting coefficient; Adaptive step size control mechanism: the expansion step size of the search tree As the distance between the current node and the nearest obstacle The changes occur in real time, and the calculation method is as follows:

[0010] In the formula, Indicates the maximum allowed expansion step size. This represents the step size adjustment coefficient, which is used to automatically reduce the step size in areas with dense obstacles to improve obstacle avoidance accuracy, and increase the step size in open areas to improve search efficiency.

[0011] Understandably, this scheme introduces an improved Rapid Expanding Random Tree (RRT) algorithm with a gravity bias and adaptive step size mechanism. This aims to address the problems of traditional RRT algorithms in industrial robotic arm path planning, such as strong search blindness, slow convergence speed, and poor path quality in areas with dense obstacles.

[0012] Specifically, regarding the gravitational bias sampling mechanism, the traditional RRT algorithm expands the search tree by uniformly random sampling throughout the configuration space. This approach leads to a large number of invalid explorations without explicit guidance, especially when the target point is far away, resulting in low convergence efficiency. The gravitational bias sampling mechanism introduced in this scheme expands the search tree by sampling with a preset probability... Sampling points are generated in the direction of the target point, and the gravitational bias vector is used. Correcting the sampling direction effectively guides the search process. Its core lies in the formula for calculating the gravitational bias vector:

[0013] This formula is not a simple linear pointing direction, but rather constructs a normalized direction vector whose magnitude is determined by the gravitational weighting coefficient. Control. This design is remarkably innovative in its technology: firstly, This ensures that regardless of the distance between the current node and the target point, the gravitational force always provides directional guidance within a unit length, avoiding the problem of step size loss due to excessive distance and guaranteeing the stability of the guidance. Secondly, probability is introduced. and weighting coefficients This mechanism is not a mandatory greedy strategy, but rather achieves a technical balance between global exploration (random sampling) and goal orientation (gravitational bias). This balance allows the algorithm to converge quickly to the target region while retaining the ability of the RRT algorithm to escape local optima, resolving the inherent contradictions of traditional algorithms and representing a non-obvious technical improvement.

[0014] Regarding the adaptive step size control mechanism, traditional RRT algorithms use a fixed step size for expansion. When the step size is large, it is prone to planning failure or path jitter in areas with dense obstacles due to the inability to find a suitable obstacle avoidance path; when the step size is small, it significantly reduces search efficiency in open areas. The adaptive step size control mechanism proposed in this scheme aims to dynamically adjust the expansion step size to adapt to environments of different complexities. Its core lies in the step size calculation formula: The technical innovation of this formula lies in combining the step size with the distance from the current node to the nearest obstacle. A direct correlation was established. This was achieved through adjusting the coefficient. This allows for precise control over the sensitivity of the step size as it changes with distance. The `min` function sets an upper limit for the step size. This ensures that the step size does not grow indefinitely in completely open areas, thus guaranteeing the overall stability and predictability of path planning. This design allows the search tree to explore with finer granularity when near obstacles, improving the success rate and smoothness of finding feasible paths in confined spaces; while in safe areas far from obstacles, it expands rapidly with larger step sizes, significantly improving global search efficiency. This is a specific and effective quantitative solution to the core contradiction between efficiency and safety in path planning, and not a conventional choice for those skilled in the art.

[0015] According to a dynamic path planning method for collaborative operation of two industrial robotic arms of the present invention, in step (b), the spatiotemporal occupancy information specifically includes: spatial dimension information, which includes the geometric bounding volume coordinates corresponding to the end effector and each link segment of each industrial robotic arm during the movement, to define the dynamic occupancy range of the robotic arm in physical space; and temporal dimension information, which includes the start timestamps corresponding to the arrival, occupation, and departure of each industrial robotic arm from the geometric bounding volume coordinates. With end timestamp This is used to form a predicted time window that can be retrieved by other robotic arms.

[0016] It is understood that this solution aims to provide a precise and unambiguous information exchange foundation for dual-arm collaboration, thereby achieving efficient collaborative obstacle avoidance. Specifically, regarding spatial dimension information, traditional methods may only treat the end effector of the robotic arm or simplify it to a single point, which is ineffective in scenarios where collisions may occur between link segments. This solution explicitly proposes using the geometric bounding volume coordinates corresponding to each link segment, meaning that each part of the robotic arm is modeled as a geometry (such as a capsule or cube) with a defined volume and position. This method can greatly improve the accuracy of collision detection because it reflects the actual physical occupancy of the robotic arm during movement, ensuring that a safe distance is maintained between any two links, which is the foundation for achieving safety in high-density collaborative operations. Secondly, regarding temporal dimension information, spatial occupancy information alone is insufficient to support dynamic obstacle avoidance, as two robotic arms may need to traverse the same spatial area at different times. This solution introduces a start timestamp. With end timestamp This elevates static space occupancy information to a four-dimensional spatiotemporal level. This constitutes a predictive occupancy time window, clearly indicating during which a specific spatial area will be unavailable in the future. Other robotic arms, when planning their own paths, can use this time window as a hard constraint to plan staggered paths, achieving efficient collaboration that avoids direct avoidance or waiting, thus significantly improving the utilization rate of shared space and the overall operational cycle time.

[0017] According to a dynamic path planning method for collaborative operation of two industrial robotic arms of the present invention, in step (b), the shared state pool adopts a data storage architecture based on timestamp segmentation, supports concurrent read and write operations of multiple arms, and uses a semaphore lock mechanism to control data consistency; the spatiotemporal information sharing step includes updating the planned path in real time through the trajectory writing interface and marking key occupied nodes to the shared state pool by timestamp.

[0018] It is understandable that the technical objective of this solution is to ensure the real-time nature, concurrency, and data consistency of information sharing, which is a technical guarantee for the stable operation of multi-agent systems. First, regarding the timestamp-based data storage architecture described in this solution, this architecture is not a simple database. Storing data in timestamp segments means that the data is logically organized in chronological order. When a robotic arm needs to query whether there are conflicts in a future time period, the system can directly access the data partition for the corresponding time period without traversing all historical or future path information. This architecture greatly improves the efficiency of information retrieval and is key to achieving real-time collaborative planning. Second, regarding the semaphore locking mechanism described in this solution, in dual-arm or multi-arm systems, all robotic arms need to read and update the shared state pool almost simultaneously. Without effective concurrency control, read-write conflicts or dirty reads can easily occur, leading to incorrect obstacle avoidance decisions. The semaphore locking mechanism is a proven and effective technical means to solve the problem of multi-threaded / multi-process resource contention. By locking the relevant data segments during write operations, it ensures the atomicity of data updates while allowing multiple read operations to be performed in parallel, thereby maximizing the system's concurrent processing capabilities and response speed while ensuring data consistency.

[0019] According to a dynamic path planning method for collaborative operation of two industrial robotic arms of the present invention, the collaborative obstacle avoidance planning step further includes a cycle synchronization mechanism. The cycle synchronization mechanism is implemented by introducing a time window penalty term into the cost function of path planning. The time window penalty term is calculated based on the expected time difference between the two arms in completing the current operation, so as to constrain the collaborative cycle error between the two arms to within 80ms.

[0020] It is understandable that this solution further introduces a cycle synchronization mechanism, the technical purpose of which is to solve the problem of not only avoiding collisions but also maintaining time coordination in dual-arm collaborative operations, especially in assembly line tasks. The cycle synchronization mechanism is implemented by introducing a time window penalty term into the cost function of path planning. This means that when planning a candidate trajectory, in addition to evaluating its path length, obstacle avoidance risk, and other conventional indicators, the time difference between the trajectory and the completion of related processes by another robotic arm is calculated. If this time difference is too large, a penalty term is added to the cost of the trajectory, putting it at a disadvantage in the evaluation. Constraining the collaborative cycle error to within 80ms is a technically innovative numerical choice. This value is not arbitrarily set but is an optimization result derived from comprehensive analysis and extensive experimental testing of the industrial robot control system response cycle, robotic arm dynamics characteristics, and typical assembly process time requirements. If the error constraint is too large (e.g., 500ms), the synchronization is lost, the close connection of processes cannot be guaranteed, and production efficiency is affected. If the error constraint is too small (e.g., 10ms), it may exceed the actual execution capabilities of the robot controller and mechanical structure, making it difficult for the path planning algorithm to find a solution that meets the conditions, or causing the generated path to frequently exceed the limit due to small disturbances during actual execution, resulting in system instability. Therefore, the value of 80ms represents a non-obvious engineering balance between ensuring the success rate of assembly tasks and improving production cycle time, reflecting the profound understanding and precise quantification of practical industrial application scenarios in this solution.

[0021] According to a dynamic path planning method for collaborative operation of two industrial robotic arms of the present invention, in the collaborative obstacle avoidance planning step (c), when a spatiotemporal overlap conflict is detected between the paths of the two arms, the system activates the following collaborative conflict resolution logic: Priority assessment sub-step: Assessing the urgency of the task Path length and the degree of conflict prediction The priority of each robotic arm is calculated using a weighted model. :

[0022] In the formula, These are the preset adjustment factors for each item; Conflict Execution Sub-Step: Tasks with lower priority scores execute a coordination and waiting strategy until the conflict area is released; or trigger a local path replanning module by adjusting the execution timestamp sequence. This enables non-blocking collaborative obstacle avoidance.

[0023] It is understood that this solution aims to provide a set of specific conflict resolution logic when a spatiotemporal conflict is detected in the path. This logic aims to transform from passive waiting or stopping to proactive and intelligent coordination, so as to improve the overall stability and operational efficiency of the system.

[0024] Specifically, regarding the priority evaluation sub-step described in this scheme, traditional methods typically handle conflicts using a fixed master-slave model or a simple first-come, first-served principle, which lacks flexibility. The priority evaluation model proposed in this scheme integrates three key dimensions through a weighted summation: Dimension 1: Task Urgency This can reflect the importance and time constraints of the task in the entire production process.

[0025] Dimension 2: Path Length This can indirectly reflect the complexity and execution cost of the task.

[0026] Dimension 3: Degree of Conflict Prediction It can quantify the severity of conflicts, such as the length of overlapping time and the distance between them.

[0027] Its weighting model formula is: The innovation of this formula lies in its transformation of scheduling decisions from qualitative judgment to quantitative calculation. This is achieved by adjusting the weighting factors. , , It can flexibly customize scheduling behavior based on different production strategies (e.g., prioritizing critical tasks or selecting the option with the lowest total cost). This is a highly configurable and dynamic decision-making mechanism, far exceeding the fixed priority settings in this field.

[0028] Secondly, regarding the conflict execution sub-step described in this scheme, this step provides two differentiated execution strategies. Strategy one is coordinated waiting, which adopts a waiting strategy for low-priority tasks; this is simple, effective, and has low computational overhead. Strategy two is local path replanning, which adjusts the execution timestamp sequence. This is achieved by translating or stretching the path along the timeline without altering its geometry. The technical significance of this is that it enables non-blocking obstacle avoidance. This is crucial for maintaining job continuity and reducing the additional computational burden and execution time caused by replanning the geometric path.

[0029] According to the dynamic path planning method for collaborative operation of two industrial robotic arms of the present invention, the mathematical expression of the multi-objective cost function F in the local path optimization step (d) is set as follows:

[0030] In the formula, L is the path length, T is the execution time, and E is the energy consumption. For the path acceleration variance, The minimum safe distance between the robotic arm and the obstacle. to These are the weighting coefficients for each performance objective; among them, the smoothness weighting coefficient is increased. The value of is chosen to reduce the variance of path acceleration. It also suppresses end-effector jitter and reduces attitude error.

[0031] It is understood that this scheme aims to provide a specific mathematical expression for the multi-objective cost function F used in the local path optimization step (d). This function is the core of trajectory evaluation using the Dynamic Window Method (DWA), and its design directly determines the quality of the local path. Its mathematical expression is: The ingenuity of this formula lies in its comprehensive definition of the optimal path. It does not pursue the extreme of a single metric, but rather seeks a balance between multiple conflicting performance objectives, which is crucial in industrial applications. In the formula, L (path length) and T (execution time) are efficiency metrics, and E (energy consumption) is an operating cost metric. (Minimum safe distance) is a safety indicator. (Path acceleration variance) is a crucial yet often overlooked technical metric. In path planning, most algorithms focus on the absolute values ​​of velocity or acceleration, but rarely on their variance. A small acceleration variance indicates smooth acceleration changes, resulting in stable, shock-free, and vibration-free robotic arm movement. This has direct and significant technical benefits for assembly tasks requiring high-precision positioning, reducing mechanical wear, extending equipment lifespan, and minimizing end-effector attitude errors. Introducing path acceleration variance as an independent and weighted objective into the cost function is a major technical innovation of this solution. Furthermore, by adjusting its weighting coefficients… It can directly control the smoothness of the generated path, which is a direct and quantitative engineering method to solve the problem of end effector jitter in robotic arms.

[0032] According to a dynamic path planning method for collaborative operation of two industrial robotic arms of the present invention, the local path optimization step (d) further includes a weight adaptive adjustment logic step: adjusting the smoothness weight coefficient in real time according to the type of task. Specifically, when the task is a high-precision assembly task, the smoothness weighting coefficient is increased. The value of is chosen to reduce the variance of path acceleration. And reduce the end-effector attitude error; when the task is energy-saving handling, reduce the smoothness weighting coefficient. The value of is chosen to reduce unit energy consumption E / L by allowing for larger acceleration fluctuations.

[0033] Understandably, this solution aims to further provide logic for adaptive weight adjustment, enabling the path optimization algorithm to move beyond static adjustments and dynamically adjust its optimization preferences based on task requirements, thus demonstrating a higher level of intelligence. In this solution, the smoothness weight coefficient is adjusted in real-time according to the accuracy requirements of the task. Specifically, in high-precision assembly tasks, the attitude stability and positioning accuracy of the end effector are primary objectives. This is achieved by increasing the smoothness weighting coefficient. The value of makes the cost function more inclined to select trajectories with extremely gentle acceleration changes, even if this may slightly increase path length or execution time. This is an intelligent trade-off that sacrifices some other performance for maximizing the key performance indicator (accuracy). In energy-saving material handling tasks, the core objective is to move objects quickly and at low cost, and the requirement for path smoothness is relatively low. By reducing the smoothness weight coefficient... The system considers both the value of the parameter and the energy consumption per unit (E / L). It selects the most energy-efficient path, allowing for some speed variation along the path to reduce overall energy consumption. This means it lowers the E / L by allowing for larger acceleration fluctuations. This adaptive adjustment logic enables a single path planning algorithm to handle various industrial tasks, greatly enhancing the system's flexibility and applicability, and avoiding the complexity of developing different algorithms or maintaining multiple sets of parameters for different tasks.

[0034] According to a dynamic path planning method for collaborative operation of two industrial robotic arms of the present invention, the method further includes a dynamic obstacle prediction step, specifically including: using a depth camera to collect point cloud data of the working environment, identifying dynamic obstacles through point cloud clustering; using a long short-term memory network (LSTM) to predict the motion trajectory of the dynamic obstacle in the next 0.5s, and feeding the predicted trajectory as a dynamic path constraint back to the local path optimization step (d) in real time, so as to achieve feedforward local dynamic obstacle avoidance, and control the path reconstruction delay to within 100ms.

[0035] Understandably, this solution aims to add a dynamic obstacle prediction step, elevating the system's obstacle avoidance capability from a passive response to an active prediction level. First, it utilizes LSTM for trajectory prediction. Traditional dynamic obstacle avoidance typically relies on linear extrapolation based on the obstacle's current speed and position, a method that performs poorly for obstacles undergoing non-linear motion (such as turning, acceleration, and deceleration). This solution employs a Long Short-Term Memory (LSTM) network, a deep learning model specifically designed for processing and predicting time-series data. LSTM can learn the complex motion patterns of dynamic obstacles, thus providing more accurate future trajectory predictions. Second, regarding the feedforward local dynamic obstacle avoidance described in this solution, the predicted future trajectory is directly fed back as a dynamic constraint to the local path optimization step, representing a form of feedforward control. When planning the current path, the robotic arm already anticipates the obstacle's position in the short term, allowing for the pre-planning of a smooth detour path instead of waiting until an impending collision to apply emergency braking or steering. Furthermore, in this solution, the 0.5s prediction timeframe represents a crucial balance between prediction accuracy and response time. A time domain that is too short is insufficient for effective advance planning; a time domain that is too long will cause prediction errors to increase dramatically, rendering it useless. 0.5s is, under current technological conditions, a time window that allows for high prediction confidence and practical planning significance for the movement of dynamic obstacles (such as AGVs and other robots) in typical industrial scenarios. A control path reconstruction delay of 100ms places higher demands on the computational power and algorithm efficiency of the entire system. It ensures that the entire closed loop from perception to prediction to replanning and executing the new path can be completed before the obstacle undergoes significant displacement. Achieving this metric means that the algorithm itself and its operating hardware platform have been highly optimized, guaranteeing that the system's feedforward obstacle avoidance capability in real dynamic environments is truly effective, not just theoretical. The combination of these two values ​​enables a higher-performance predictive obstacle avoidance system.

[0036] In addition, the present invention also provides a dynamic path planning system for collaborative operation of two industrial robotic arms, comprising: a global path generation module for executing the global path generation step in the dynamic path planning method of the present invention; a shared state pool module for executing the spatiotemporal information sharing step in the dynamic path planning method of the present invention to achieve real-time synchronization of occupancy information between the two arms; a collaborative obstacle avoidance planning module for executing the collaborative obstacle avoidance planning step in the dynamic path planning method of the present invention; and a local path optimization module for executing the local path optimization step in the dynamic path planning method of the present invention.

[0037] The technical advantages of the dynamic path planning method and system for collaborative operation of two industrial robotic arms of the present invention include: First, by constructing a two-layer planning framework of global path generation and local path optimization, this method effectively combines the advantages of two planning approaches. Global path generation ensures the completeness and goal orientation of the planning, avoiding the problem of pure local planning methods easily getting trapped in local optima; while local optimization gives the system high flexibility and real-time response capability, enabling it to effectively cope with dynamically changing environments.

[0038] Secondly, the introduction of a spatiotemporal information sharing mechanism transforms obstacle avoidance between robotic arms from passive collision detection to proactive intent prediction and coordination. By sharing spatiotemporal occupancy information including time windows, one robotic arm can obtain the future motion plan of another, thereby generating motion trajectories that are staggered in the time dimension. This approach achieves non-blocking cooperative paths, replacing the inefficient obstacle avoidance strategy that forces one robotic arm to pause its movement to wait for another to pass, thus greatly improving the utilization rate of shared space and overall operational efficiency.

[0039] Finally, a dynamic window method combined with a multi-objective cost function is used for real-time evaluation and selection of local paths, ensuring that the final executed trajectory is not only safe and collision-free, but also achieves comprehensive optimization in multiple dimensions such as motion smoothness, energy consumption, and execution time. This refined local optimization capability is of great significance for improving the positioning accuracy of the robotic arm, reducing mechanical impact, and extending equipment life. In summary, the present invention forms a complete dynamic collaborative planning system from macro-level guidance to micro-level decision-making, effectively solving the problems of path conflict and cycle time mismatch in dual-arm collaborative operations, and significantly enhancing the system's intelligence, safety, and production efficiency. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0041] Figure 1 This is a flowchart of the collaborative operation process of the present invention; Figure 2 This is a visualization diagram of the dynamic obstacle prediction and path update of the present invention; Figure 3 This is a graph showing the relationship between path smoothness and energy consumption in this invention; Figure 4 This is a schematic diagram of the overall structure of the improved RRT-DWA joint algorithm of the present invention; Figure 5 This is a schematic diagram of the improved RRT sampling method of the present invention; Figure 6This is a schematic diagram of the shared state pool structure and synchronization mechanism of the present invention; Figure 7 This is a flowchart of the collaborative path scheduling process of the present invention; Figure 8 This is a schematic diagram of the simulation scene of the present invention; Figure 9 These are simulation diagrams of typical scenarios of the present invention; Figure 10 This is a schematic diagram of the collaborative operation of two industrial robotic arms in the intelligent manufacturing unit of this invention. Figure 11 This is a schematic diagram of the key scheduling logic and dynamic path switching process of the system of the present invention. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0043] This embodiment provides a dynamic path planning method for collaborative operation of two industrial robotic arms. The method is based on a simulation platform using ROS2 and Gazebo, which is configured with two UR series six-DOF industrial robotic arms, hereinafter referred to as robotic arm A and robotic arm B. The working environment is set as a shared workspace of 1.2m × 1.2m, containing static obstacles and one or more dynamic obstacles, to simulate a typical collaborative assembly unit of metal parts. Figure 8 The screenshot of this simulation scene clearly shows the layout of two industrial robotic arms, interactive target points, and dynamic and static obstacles.

[0044] like Figure 1 As shown, this is the overall process of dual-arm collaborative operation in this embodiment. This process clearly demonstrates the entire control logic from receiving tasks, decomposing tasks, generating paths, to updating the shared path pool, forming the basic framework of this method. Specifically, the dynamic path planning method for dual industrial robotic arm collaborative operation in this embodiment includes the following steps: First, the global path generation step is executed. In this step, the system acquires the work tasks of each industrial robotic arm (robotic arm A and robotic arm B), including the clear start point, target end point, and environmental map information containing the positions of static obstacles such as workbenches and fixtures. Combined with... Figure 4The diagram illustrates the overall algorithm structure, with the planning process entering the global layer. At this layer, the system employs a global path search algorithm based on random sampling to search within the robotic arm's configuration space. The goal of this search algorithm is to generate an initial feasible path connecting the starting point and the target endpoint. This initial feasible path is a macroscopic geometric path that does not interfere with known static obstacles, providing a benchmark for subsequent local dynamic adjustments.

[0045] After generating the initial feasible path, the system executes step (b) of spatiotemporal information sharing. The system analyzes the initial feasible path generated in the previous step, extracting a series of critical path nodes. Subsequently, combining the robot arm's preset operating speed and dynamic parameters, the system calculates the estimated time window for the robot arm to move to each critical path node and its surrounding area. This time window includes a start timestamp and an end timestamp. Through this calculation, purely geometric path information is transformed into spatiotemporal occupancy information containing spatial coordinates and time intervals. (Refer to...) Figure 6 The diagram illustrates the shared state pool structure and synchronization mechanism. This spatiotemporal occupancy information is published in real-time to a pre-defined shared state pool that is stored in segments based on timestamps, via a trajectory writing interface. For example, robotic arm A writes the spatiotemporal occupancy information of its planned path into the shared state pool for robotic arm B to retrieve. This step quantifies and publicizes the motion intentions of each robotic arm, providing a data foundation for subsequent cooperative obstacle avoidance.

[0046] After the shared state pool is established, each industrial robotic arm (robotic arm A and robotic arm B) independently and in parallel executes cooperative obstacle avoidance planning at a preset short period (tens of milliseconds in this embodiment), i.e., executing cooperative obstacle avoidance planning step (c). Specifically, robotic arm B retrieves the spatiotemporal occupancy information published by robotic arm A from the shared state pool through the path reading interface. The system directly converts the retrieved information into dynamic spatial conflict constraints, i.e., specific spatial areas are considered impassable within a specific time period. The planning module fuses this dynamic spatial conflict constraint with the static obstacle distribution information in the current environment, and generates a set of candidate trajectory segments that meet the cooperative obstacle avoidance requirements within the neighborhood of the initial feasible path generated in step (a). Figure 11 The diagram illustrates the key scheduling logic and dynamic path switching process of the system. When robotic arm B predicts a conflict with robotic arm A or a dynamic obstacle in its execution path, this step generates new alternative path segments. These path segments constitute a set of candidate trajectory fragments to ensure the continuity of motion.

[0047] Finally, the local path optimization step (d) is executed, which is the decision-making step for the candidate trajectories. The system uses the Dynamic Window Method (DWA) to construct a forward prediction rolling window. Within the velocity-angular velocity space allowed by the current physical limitations of the robotic arm, the system densely samples the set of candidate trajectory segments generated in step (c). Subsequently, each sampled trajectory is quantitatively evaluated in real time using a preset multi-objective cost function. This cost function comprehensively considers multiple performance indicators such as path length, execution time, safety, and smoothness. The system ultimately selects the trajectory segment with the best score under the multi-objective cost function evaluation as the final execution path for the current control cycle. By repeatedly looping this step, the robotic arm can continuously and smoothly adjust its trajectory. Figure 9 The simulation results of the typical scenario shown are as follows: Figure 9 The demonstration shows that in a high-density obstacle environment, the motion trajectory curves of the two robotic arms were clearly separated through the planning of this method, effectively avoiding potential collision zones. Figure 9 Figure b demonstrates how, in a dynamic obstacle avoidance scenario, the algorithm automatically avoids obstacles based on predicted obstacle trajectories, generating new detour paths. These results intuitively prove that the planning method of this invention can achieve dynamic obstacle avoidance and synchronized operation rhythm between the two arms in a shared space.

[0048] In one embodiment, in step (a), the global path search algorithm is an improved Rapid Expanding Random Tree (RRT) algorithm that incorporates a gravity bias and adaptive step size mechanism. (Refer to...) Figure 5 The diagram shows the improved RRT sampling, which integrates a gravity bias sampling mechanism and an adaptive step size control mechanism on the basis of traditional RRT.

[0049] Regarding the gravity bias sampling mechanism, during the iterative process of search tree expansion, the algorithm uses a preset probability. Sampling points are forcibly generated in the direction of the target point. This process is achieved through a gravity bias vector. This vector is used to guide the search tree to converge quickly towards the target direction, thereby reducing ineffective exploration in irrelevant regions. The specific calculation method for the gravity bias vector is as follows:

[0050] In the formula, Indicates the coordinates of the current node. Indicates the coordinates of the target point. This represents the preset gravity weighting coefficient. For example... Figure 5 As shown by the thick black arrow in the middle, this vector provides a clear direction for expanding the search tree.

[0051] Regarding the adaptive step size control mechanism, in order to adapt to the complexity of obstacle distribution in the environment, the expansion step size of the search tree is adjusted. As the distance between the current node and the nearest obstacle The changes occur in real time, and the calculation method is as follows:

[0052] In the formula, Indicates the maximum allowed expansion step size. This represents the step size adjustment coefficient, and the effect of this mechanism is as follows: Figure 5 As shown, in areas with dense obstacles (the area closed by the dashed line in the figure), due to Decrease, step size This also reduces the size accordingly, allowing the search tree to explore with more precise steps, thereby improving obstacle avoidance accuracy; while in open areas (such as... Figure 5 (As shown by the thick dashed line segment in the lower right of the middle) Larger, step size Increase and limited This improves search efficiency.

[0053] By combining the two mechanisms mentioned above, the initial feasible path generated by the improved RRT algorithm is optimized in terms of both convergence speed and path smoothness, providing higher quality input for subsequent local planning.

[0054] In step (b), the spatiotemporal occupancy information specifically includes spatial dimension information and temporal dimension information.

[0055] Regarding spatial dimension information, to accurately define the dynamic occupancy range of the robotic arm in physical space, the spatial dimension information includes the coordinates of the end effector and the geometric bounding volume corresponding to each link segment of the industrial robotic arm during its movement. In this embodiment, each link of the robotic arm is modeled as a capsule. At each key node of the path, the system calculates the three-dimensional spatial coordinates and orientation of all capsules. These coordinate sets together constitute the accurate physical occupancy model of the robotic arm at that moment. Compared to simplifying the robotic arm into a point or line model, this approach greatly improves the reliability of collision detection and effectively prevents interference between link segments.

[0056] Regarding the time dimension information, in order to achieve dynamic collaboration, the time dimension information includes the start timestamps corresponding to the arrival, occupation, and departure of each industrial robotic arm from the aforementioned geometric bounding volume coordinates. With end timestamp These two timestamps together form a predicted time window that can be retrieved by other robotic arms. When a robotic arm plans its path, it can query the shared state pool to obtain the space areas that other robotic arms will occupy in various future time periods. This allows the planning algorithm to generate time-staggered paths, achieving efficient dynamic avoidance rather than simple spatial avoidance.

[0057] Furthermore, in step (b), the shared state pool adopts a timestamp-based segmented data storage architecture, supports multi-arm concurrent read and write operations, and utilizes a semaphore locking mechanism to control data consistency. (Refer to...) Figure 6 The diagram illustrates the shared state pool structure and synchronization mechanism. In this embodiment, the shared state pool is constructed centrally within the system. The core data structure of this shared state pool uses a timestamp-based approach. The system maintains the occupied area records of each robotic arm in a segmented manner. This architecture allows the system to efficiently query the space occupancy status within a specific future time period without traversing the entire dataset. The spatiotemporal information sharing step specifically includes updating the planned path in real time via a trajectory writing interface and marking key occupancy nodes with timestamps in the shared state pool. Within each planning cycle, after the upstream robotic arm (robotic arm A) completes path planning, it updates the spatiotemporal occupancy information corresponding to its latest planned path to the shared state pool via the trajectory writing interface. Simultaneously, the downstream robotic arm (robotic arm B) concurrently obtains this information via a path reading interface. To address potential data conflicts caused by concurrent read / write operations between the two arms, this embodiment employs a semaphore lock mechanism to control data consistency. When a robotic arm performs a write operation, the relevant memory area is locked until the write is complete, thus preventing data corruption or incomplete readings. This mechanism ensures data accuracy while supporting efficient concurrent reading, guaranteeing the high responsiveness and stability of the entire collaborative system.

[0058] Furthermore, the collaborative obstacle avoidance planning step further includes a timing synchronization mechanism. Specifically, in tasks with strict time coordination requirements, such as collaborative assembly by both arms, this embodiment introduces a timing synchronization mechanism. This timing synchronization mechanism is implemented by introducing a time window penalty term into the cost function of path planning. When the system plans a candidate trajectory for one of the robotic arms, it not only evaluates its length, obstacle avoidance, and other performance characteristics, but also calculates the estimated completion time of the trajectory. The time window penalty term is calculated based on the estimated time difference between the two arms completing the current associated work steps. Specifically, the system calculates the time difference between the robotic arm and the other robotic arm in completing their respective task steps if a candidate trajectory is adopted. If this difference exceeds a preset threshold, the cost of the candidate trajectory is significantly penalized, putting it at a disadvantage in subsequent optimal trajectory selection. In this way, the planning algorithm is guided to select trajectories that make the movements of the two arms more coordinated and the time difference smaller. In this embodiment, by adjusting the weight of the penalty term, the collaborative timing error between the two arms can be constrained to within 80ms. This value ensures that in tightly coupled tasks such as gripping and fitting metal parts, the movements of the two robotic arms can be highly synchronized, avoiding production interruptions or reduced efficiency caused by one robotic arm completing its action too early or too late, thereby improving the overall production cycle time and stability.

[0059] Furthermore, in the cooperative obstacle avoidance planning step (c), when a spatiotemporal overlap conflict is detected between the two arm paths, the system activates the following cooperative conflict resolution logic. (Refer to...) Figure 7 The cooperative path scheduling process shown begins with the system retrieving the execution paths for each robotic arm from the task queue and performing path conflict detection. Once a temporal and spatial overlap conflict is detected between two paths, the cooperative conflict resolution logic is activated, and the system employs the following cooperative conflict resolution logic: Resolution Logic (1), Priority Evaluation Sub-step: The system enters the priority evaluation module. This module does not use a fixed master-slave relationship, but dynamically calculates the priority score of the current task of each robotic arm. Priority scoring takes into account task urgency. Path length and the degree of conflict prediction The calculation is performed using a weighted model:

[0060] In the formula, These are the preset adjustment factors for each item. Through this model, the system can quantitatively and dynamically prioritize conflicting parties based on current production needs (such as prioritizing high-urgency tasks or selecting the path with lower total cost).

[0061] (2) Conflict Resolution Logic: After completing the priority assessment, the system executes the corresponding conflict resolution strategy based on the scoring results. For tasks with lower priority scores, the system will execute a coordinated waiting strategy, i.e., the robotic arm will pause execution or slow down and wait in front of the conflict area until the information in the shared state pool shows that the conflict area has been released by a higher-priority robotic arm. Alternatively, the system triggers the local path replanning module. This module adjusts the execution timestamp sequence of the robotic arm. Without altering the path geometry, non-blocking collaborative obstacle avoidance is achieved by delaying the planned execution time of key state points (such as waypoints and start / stop points) along the path. This time-adjusted obstacle avoidance method has lower computational overhead and faster response speed compared to recalculating the geometric path, thus better maintaining the continuity of operations.

[0062] Furthermore, the mathematical expression for the multi-objective cost function F in the local path optimization step (d) is set as follows:

[0063] In the local path optimization step (d), the Dynamic Window Method (DWA) evaluates candidate trajectories using a multi-objective cost function F. This function is designed to comprehensively evaluate multiple key performance indicators of the path. The specific structure of the function is as follows: L is the path length, T is the execution time, and E is the energy consumption. For the path acceleration variance, The minimum safe distance between the robotic arm and the obstacle. to These are the weighting coefficients for each performance objective. By adjusting these weighting coefficients, the optimization tendency of path planning can be changed.

[0064] Among them, the path acceleration variance This is a key smoothness metric. A smaller acceleration variance means smoother velocity changes during the robotic arm's movement, thus reducing physical shock and vibration. In this embodiment, the smoothness weighting coefficient is increased. 4. It can effectively reduce the variance of path acceleration, and its direct technical effect is to suppress the jitter of the robotic arm's end effector and reduce attitude error. (Refer to...) Figure 3 The curve showing the relationship between path smoothness and energy consumption increases... Although it may lead to a slight increase in energy consumption per unit (E / L), the significant decrease in path acceleration variance demonstrates its effectiveness in improving motion smoothness, which is crucial for tasks such as high-precision assembly.

[0065] Furthermore, the local path optimization step (d) also includes the following adaptive weight adjustment logic, which adjusts the smoothness weight coefficient in real time according to the accuracy requirements of the task. In this embodiment, when the system's task scheduler issues a task to the path planning module, it includes a type tag for the task, such as "high-precision assembly" or "energy-saving handling." Upon receiving this tag, the local path optimization module triggers adaptive weight adjustment logic.

[0066] Specifically, increasing the precision in high-precision assembly tasks. This reduces path acceleration variance and end-effector attitude error. Specifically, when the task label is "high-precision assembly," the system automatically increases the smoothness weighting coefficient. The value of . This allows the multi-objective cost function to evaluate the path acceleration variance when assessing candidate trajectories. The algorithm imposes a higher penalty on certain items. Therefore, it tends to choose trajectories with gentler acceleration changes, even if this may slightly sacrifice path length or execution time. The end result is more stable robotic arm movement, effectively suppressing end effector jitter and keeping the average attitude error within ±2.3mm, meeting the stringent precision requirements of precision fitting and other operations.

[0067] In energy-saving handling tasks, it reduces The optimization objective is to optimize unit energy consumption (E / L). That is, when the task label is "energy-saving transportation," the system's optimization goal shifts to efficiency and energy consumption. At this point, the adaptive weight adjustment logic is reduced. The value of is reduced, decreasing the requirement for path smoothness, while relatively increasing the weight of energy consumption term E and path length L in the cost function. This allows the algorithm to select trajectories with more direct acceleration and deceleration changes and shorter paths, thereby optimizing unit energy consumption E / L while ensuring safety and reducing the system's operating costs.

[0068] This adaptive weight adjustment logic enables a single planning algorithm to flexibly adapt to various production needs, enhancing the system's flexibility and versatility.

[0069] Preferably, the method of this embodiment further includes a dynamic obstacle prediction step to achieve feedforward local dynamic obstacle avoidance. Specifically, refer to... Figure 2 The diagram illustrates dynamic obstacle prediction and path update on the ROS2 platform. This embodiment adds a separate dynamic obstacle prediction module to the overall planning framework. The specific steps include: (1) Use a depth camera to collect point cloud data of the working environment and identify dynamic obstacles through point cloud clustering: The RealSense depth camera configured in the system continuously collects three-dimensional point cloud data of the working environment at a high frequency. This data is sent to a processing node based on the PCL library. This node uses point cloud clustering algorithms (such as Euclidean clustering segmentation) to identify and segment independent dynamic obstacles in the environment and extract their position, size and other information.

[0070] (2) Predicting the trajectory of the dynamic obstacle in the next 0.5s using a Long Short-Term Memory (LSTM) network: For each identified dynamic obstacle, the system tracks its position sequence in consecutive time frames. This time series data is input into a pre-trained Long Short-Term Memory (LSTM) network model. The LSTM model is able to learn and capture the complex motion patterns of the obstacle and predict its trajectory in the next 0.5s accordingly.

[0071] (3) The predicted trajectory is fed back as a dynamic path constraint to the local path optimization step (d) in real time: The predicted future trajectory, along with its confidence interval, is sent to the local path optimization module in real time. In the optimization module, these predicted future positions are regarded as dynamic constraints that change over time. Therefore, when DWA evaluates candidate trajectories, it also checks whether these trajectories will collide with the predicted path of the obstacle within the next 0.5s. This feedforward obstacle avoidance mechanism enables the robotic arm to plan its bypass actions in advance, rather than waiting until the obstacle is close before reacting. This embodiment, through hardware and software co-optimization, can control the path reconstruction delay of the entire "perception-prediction-replanning" process to within 100ms, ensuring the system's rapid response and safe avoidance capability when facing sudden dynamic obstacles.

[0072] Based on the above, this embodiment also provides a dynamic path planning system for collaborative operation of two industrial robotic arms, implementing the above method. This system, through modular design, decomposes the complex planning process into multiple collaborative functional units to ensure the efficient and stable operation of the entire planning process. (Refer to...) Figure 4 The diagram shows the overall structure of the improved RRT-DWA joint algorithm. Logically, the system can be divided into the following core modules.

[0073] Global Path Generation Module: This module is responsible for executing the global path generation step. After the system receives the task, the global path generation module obtains the robotic arm's starting point, target endpoint, and environmental map information. In this embodiment, this module integrates an improved Rapid Expanding Random Tree (RRT) algorithm, which includes a gravity bias and adaptive step size mechanism. The module's output is an initial feasible path connecting the starting point and the target endpoint, avoiding static obstacles. This path provides macroscopic guidance for subsequent local planning.

[0074] Shared State Pool Module: This module executes the spatiotemporal information sharing steps to achieve real-time synchronization of occupancy information between the two arms. This module maintains a central data storage area, employing a timestamp-based segmented data storage architecture and utilizing a semaphore lock mechanism to ensure data consistency during concurrent reads and writes. The paths generated by the global path generation module are processed to form spatiotemporal occupancy information including geometric bounding volume coordinates and predicted occupancy time windows, and published through the write interface of this module.

[0075] Collaborative Obstacle Avoidance Planning Module: This module executes the collaborative obstacle avoidance planning steps. It retrieves spatiotemporal occupancy information of other robotic arms from the shared state pool module at preset intervals and transforms this information into dynamic spatial conflict constraints. Combining static obstacle information, this module generates a series of candidate trajectory segments that meet the collaborative avoidance requirements within the neighborhood of the initial feasible path. Furthermore, this module integrates a beat synchronization mechanism and collaborative conflict resolution logic to handle time synchronization and path overlap issues.

[0076] Local Path Optimization Module: This module executes the local path optimization step. It employs the Dynamic Window Method (DWA), sampling and evaluating the candidate trajectory fragment set generated by the cooperative obstacle avoidance planning module in real-time within a rolling window using a multi-objective cost function. After evaluation, the optimal trajectory fragment is selected as the execution path for the current control cycle and output to the robotic arm controller. This module can also integrate adaptive weight adjustment logic and receive dynamic obstacle prediction information from external sources to further improve path quality and the system's environmental adaptability.

[0077] In actual operation, these four modules work closely together to form a complete closed loop from global to local, from planning to execution. The global path generation module provides a long-term planning blueprint, the shared state pool module acts as an information exchange bus between the two arms, the collaborative obstacle avoidance planning module generates diverse local solutions based on shared information, and finally the local path optimization module makes the final and optimal real-time decision, thus jointly realizing the efficient and safe collaborative operation of the two industrial robotic arms in complex dynamic environments.

[0078] Finally, through Figure 10This invention demonstrates a typical industrial scenario where the method can be applied: the collaborative operation of two industrial robotic arms in a smart manufacturing unit. In this scenario, robotic arm A picks up metal components from the left-hand loading area, while robotic arm B performs insertion or fixing operations on parts on tooling fixtures in the right-hand assembly area. The two robotic arms need to work synchronously within a limited shared space. By applying the dynamic path planning and collaborative method described in this invention, the two robotic arms can maintain smooth paths while avoiding each other and the processing fixtures in the center of the worktable, thereby efficiently and safely completing continuous picking-and-fitting tasks.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A dynamic path planning method for collaborative operation of two industrial robotic arms, characterized in that, Includes the following steps: (a) Global path generation steps: Obtain the starting point, target endpoint and environmental map information of each industrial robotic arm, and use a global path search algorithm based on random sampling to search within the configuration space to generate an initial feasible path connecting the starting point and the target endpoint; (b) Spatiotemporal information sharing step: extract the critical path nodes in the initial feasible path, and calculate the estimated time window corresponding to each critical path node in combination with the preset running speed, thereby forming spatiotemporal occupancy information, and publish the spatiotemporal occupancy information to the preset shared state pool; (c) Collaborative obstacle avoidance planning steps: Each industrial robotic arm retrieves the spatiotemporal occupancy information of other robotic arms from the shared state pool at a preset period, converts the retrieved information into dynamic spatial conflict constraints, and combines the current environmental obstacle distribution information to generate a set of candidate trajectory segments that meet the collaborative avoidance requirements in the neighborhood of the initial feasible path. (d) Local path optimization steps: A rolling window is constructed using the dynamic angular velocity method (DWA). The candidate trajectory segment set is sampled in the velocity-angular velocity space, and the sampled trajectory is evaluated in real time through a multi-objective cost function. The optimal trajectory segment is selected as the current execution path to achieve dynamic obstacle avoidance and operation rhythm synchronization of the two arms in the shared space.

2. The dynamic path planning method for collaborative operation of two industrial robotic arms according to claim 1, characterized in that, In step (a), the global path search algorithm is an improved Rapid Expanding Random Tree (RRT) algorithm that incorporates a gravity bias and adaptive step size mechanism, specifically including: Gravity bias sampling mechanism: During the search tree expansion iteration process, sampling is performed with a preset probability. Force the generation of sampling points in the direction of the target point, and use the gravity bias vector. The gravity bias vector is calculated to guide the search tree to converge toward the target direction as follows: In the formula, Indicates the coordinates of the current node. Indicates the coordinates of the target point. This represents the preset gravity weighting coefficient; Adaptive step size control mechanism: the expansion step size of the search tree As the distance between the current node and the nearest obstacle The changes occur in real time, and the calculation method is as follows: In the formula, Indicates the maximum allowed expansion step size. This represents the step size adjustment coefficient, which is used to automatically reduce the step size in areas with dense obstacles to improve obstacle avoidance accuracy, and increase the step size in open areas to improve search efficiency.

3. The dynamic path planning method for collaborative operation of two industrial robotic arms according to claim 1, characterized in that, In step (b), the spatiotemporal occupancy information specifically includes: Spatial dimension information, which includes the geometric bounding volume coordinates of the end effector and each link segment of each industrial robotic arm during the movement process, is used to define the dynamic occupancy range of the robotic arm in physical space. The time dimension information includes the start timestamps corresponding to the arrival, occupation, and departure coordinates of each industrial robotic arm within the geometric bounding volume. With end timestamp This is used to form a predicted time window that can be retrieved by other robotic arms.

4. The dynamic path planning method for collaborative operation of two industrial robotic arms according to claim 1, characterized in that, In step (b), the shared state pool adopts a data storage architecture based on timestamp segmentation, supports multi-arm concurrent read and write, and uses a semaphore lock mechanism to control data consistency; the spatiotemporal information sharing step includes updating the planned path in real time through the trajectory writing interface and marking key occupied nodes to the shared state pool by timestamp.

5. The dynamic path planning method for collaborative operation of two industrial robotic arms according to claim 3, characterized in that, The collaborative obstacle avoidance planning step further includes a rhythm synchronization mechanism, which is implemented by introducing a time window penalty term into the cost function of path planning. The time window penalty term is calculated based on the estimated time difference between the two arms completing the current operation, so as to constrain the collaborative rhythm error between the two arms to within 80ms.

6. The dynamic path planning method for collaborative operation of two industrial robotic arms according to claim 1, characterized in that, In the cooperative obstacle avoidance planning step (c), when a spatiotemporal overlap conflict is detected between the two arm paths, the system activates the following cooperative conflict resolution logic: Priority assessment sub-step: Assessing the urgency of the task Path length and the degree of conflict prediction The priority of each robotic arm is calculated using a weighted model. : In the formula, These are the preset adjustment factors for each item; Conflict Execution Sub-Step: Tasks with lower priority scores execute a coordination and waiting strategy until the conflict area is released; or trigger a local path replanning module by adjusting the execution timestamp sequence. This enables non-blocking collaborative obstacle avoidance.

7. The dynamic path planning method for collaborative operation of two industrial robotic arms according to claim 1, characterized in that, The mathematical expression for the multi-objective cost function F in the local path optimization step (d) is set as follows: In the formula, L is the path length, T is the execution time, and E is the energy consumption. For the path acceleration variance, The minimum safe distance between the robotic arm and the obstacle. to These are the weighting coefficients for each performance objective; among them, the smoothness weighting coefficient is increased. The value of is chosen to reduce the variance of path acceleration. It also suppresses end-effector jitter and reduces attitude error.

8. The dynamic path planning method for collaborative operation of two industrial robotic arms according to claim 7, characterized in that, The local path optimization step (d) also includes a weight adaptive adjustment logic step: The smoothness weight coefficient is adjusted in real time according to the type of task. Specifically: When the task is a high-precision assembly task, increase the smoothness weighting coefficient. The value of is chosen to reduce the variance of path acceleration. And reduce end-effector attitude error; When the task is energy-saving material handling, reduce the smoothness weighting coefficient. The value of is chosen to reduce unit energy consumption E / L by allowing for larger acceleration fluctuations.

9. The dynamic path planning method for collaborative operation of two industrial robotic arms according to claim 1, characterized in that, The method further includes a dynamic obstacle prediction step, specifically comprising: Point cloud data of the working environment is collected using a depth camera, and dynamic obstacles are identified through point cloud clustering. The Long Short-Term Memory (LSTM) network is used to predict the trajectory of the dynamic obstacle in the next 0.5s, and the predicted trajectory is fed back to the local path optimization step (d) in real time as a dynamic path constraint to achieve feedforward local dynamic obstacle avoidance, and the control path reconstruction delay is within 100ms.

10. A dynamic path planning system for collaborative operation of two industrial robotic arms, characterized in that, include: Global path generation module: used to perform the global path generation step in the method as described in any one of claims 1 to 3; The shared state pool module is used to execute the spatiotemporal information sharing step in the method as described in claim 1 or 4, so as to realize the real-time synchronization of occupancy information between the two arms; A collaborative obstacle avoidance planning module is used to perform the collaborative obstacle avoidance planning step in the method as described in any one of claims 1, 5 or 6; A local path optimization module is used to perform the local path optimization step in the method as described in any one of claims 1, 7, or 8.

Images