Industrial robot dynamic path planning method and system based on AI

Abstract

The invention relates to the technical field of industrial robot dynamic path planning, and particularly discloses an AI-based industrial robot dynamic path planning method and system. Comprising the steps of fusion perception data set generation, dynamic parameter set construction, trajectory prediction input signal generation, obstacle motion trajectory prediction, collision risk map generation, path optimization objective function generation, dynamic path re-planning execution and joint control instruction conversion. According to the method, a perception data set is fused in real time, environment changes and the dynamic state of the robot are captured, a dynamic parameter set is generated, a collision risk map is constructed, meanwhile, the weight is optimized by dynamically adjusting a path, a target function is generated to execute re-planning, and finally a re-planning output path is obtained; dependence of a traditional method on a static map is avoided, efficient operation in a changeable industrial scene is ensured, safety and efficiency are balanced, an active safety barrier is provided for an industrial robot, and the efficiency and quality of an industrial automation process are optimized.

Description

Technical Field

[0001] The present invention belongs to the technical field of industrial robot dynamic path planning, and relates to an AI-based industrial robot dynamic path planning method and system. Background Art

[0002] Traditional industrial robot path planning methods suffer from significant drawbacks when faced with moving obstacles in dynamic environments. These methods typically rely on offline calculations to generate fixed trajectories and are unable to perceive and respond to environmental changes in real time. When obstacles suddenly appear or shift in position, the robot may be unable to adjust its path in a timely manner, causing mission interruptions or operational delays. Furthermore, traditional planning algorithms rely on simple geometric models or rule bases and lack in-depth analysis of the robot's own dynamic state. Overall, this problem stems from the lag in environmental perception and insufficient adaptability, making it difficult for industrial robots to maintain a balance between safety and efficiency during dynamic operations.

[0003] Traditional solutions primarily rely on a predefined path combined with a correction mechanism based on sensor feedback. This mechanism achieves local obstacle avoidance by building a map of the environment and recalculating the shortest path when obstacles are detected. In practice, the system sets a safety distance threshold, triggering path replanning when an obstacle approaches. However, this entire planning process is often independent of the robot's real-time motion state, focusing solely on external environmental factors while ignoring the robot's own dynamic characteristics.

[0004] The drawbacks of traditional methods are their delayed response and insufficient optimization, making them incapable of efficiently handling highly dynamic scenarios. First, path correction based on predefined rules can only handle simple obstacle movements, such as uniform linear movement. However, when faced with complex nonlinear movements or interactions between multiple obstacles, the prediction accuracy drops sharply, frequently leading to misjudgments or collision risks. Secondly, these methods do not integrate the robot's internal state data, resulting in the planned path avoiding obstacles but potentially causing arm jitter, increased energy consumption, or extended task execution time. At the same time, the offline replanning process is computationally intensive, and in real-time systems with millisecond-level response requirements, it can easily cause processing delays, affecting the overall production rhythm. Summary of the Invention

[0005] In view of this, in order to solve the problems raised in the above background technology, an AI-based industrial robot dynamic path planning method and system are proposed.

[0006] The objectives of the present invention can be achieved through the following technical solutions: The first aspect of the present invention provides an AI-based industrial robot dynamic path planning method, including: S1, fusion perception data set generation: obtaining the real-time motion data and environmental perception data of the industrial robot to generate a fusion perception data set.

[0007] S2. Dynamic parameter set construction: construct a motion state vector based on the fused perception data set to generate a dynamic parameter set.

[0008] S3. Trajectory prediction input signal generation: parsing the dynamic parameter set to generate a trajectory prediction input signal.

[0009] S4. Obstacle motion trajectory prediction: Generate an obstacle motion trajectory based on the trajectory prediction input signal and the fused perception data set.

[0010] S5. Collision risk map generation: combining the dynamic parameter set and the obstacle motion trajectory to generate a collision risk map.

[0011] S6. Generate a path optimization objective function: Based on the collision risk map and the trajectory prediction input signal, adjust the path optimization weights to generate an objective function.

[0012] S7. Dynamic path replanning execution: Apply the objective function to execute dynamic path replanning and generate a replanned output path.

[0013] S8. Joint control instruction conversion: convert the replanned output path to generate a set of joint control instructions.

[0014] The second aspect of the present invention provides an AI-based industrial robot dynamic path planning system, including: a fusion perception data set generation module, which obtains the real-time motion data and environmental perception data of the industrial robot and generates a fusion perception data set.

[0015] The dynamic parameter set construction module constructs a motion state vector based on the fused perception data set to generate a dynamic parameter set.

[0016] The trajectory prediction input signal generating module parses the dynamic parameter set and generates a trajectory prediction input signal.

[0017] The obstacle motion trajectory prediction module generates an obstacle motion trajectory based on the trajectory prediction input signal and the fused perception data set.

[0018] The collision risk map generation module combines the dynamic parameter set and the obstacle motion trajectory to generate a collision risk map.

[0019] The path optimization objective function generation module adjusts the path optimization weights according to the collision risk map and the trajectory prediction input signal to generate an objective function.

[0020] The dynamic path replanning execution module applies the objective function to execute dynamic path replanning and generates a replanned output path.

[0021] The joint control instruction conversion module converts the replanned output path to generate a set of joint control instructions.

[0022] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects: (1) The present invention can capture environmental changes and the robot's own dynamic state, such as the obstacle motion trajectory and the acceleration vector of the end effector, by fusing the perception data set in real time, thereby constructing a more comprehensive spatial model, avoiding the traditional method's reliance on static maps, and enabling the robot to maintain efficient operation in changing industrial scenarios without interrupting the task for manual reconfiguration.

[0023] (2) This invention dynamically adjusts the path optimization weights based on the motion smoothness index, ensuring path continuity while reducing vibration and sudden stops of the robot arm. Compared with traditional methods, this technology balances the dual objective functions of safety and efficiency, avoids overly conservative path design, and thus reduces energy consumption and execution time.

[0024] (3) This invention constructs a collision risk map, marks the time windows and position coordinates of potential conflict points in advance, and generates an objective function based on this to constrain path search. This solves the problem of response delay in traditional path planning, ensures dynamic replanning within millisecond timescales, improves the reliability of the system in safety-critical environments, provides an active safety barrier for industrial robots, and reduces maintenance costs and operational risks.

[0025] (4) This invention directly converts the replanning output into joint control instructions through discretized paths and inverse kinematics solutions, ensuring the consistency and accuracy of trajectory execution. This avoids the common problem of disconnection between path and execution in traditional methods. In practice, pulse width modulation instructions are generated based on real-time data to drive the servo motor to achieve smooth motion, enhancing the system's end-to-end integration capabilities and reducing the need for manual intervention, thereby optimizing the efficiency and quality of the entire industrial automation process. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0027] Figure 1 Schematic diagram of the method steps of the present invention.

[0028] Figure 2 It is a schematic diagram of the system structure connection of the present invention.

[0029] Figure 3Schematic diagram of the steps for generating a dynamic parameter set of the present invention. DETAILED DESCRIPTION

[0030] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0031] See also Figure 1 As shown, the first aspect of the present invention provides an AI-based industrial robot dynamic path planning method, including: S1, fusion perception data set generation: obtaining the real-time motion data and environmental perception data of the industrial robot to generate a fusion perception data set.

[0032] In a specific embodiment of the present invention, the specific process of generating a fusion perception data set is: collecting angle data and angle change rate data of the six-axis joints of the industrial robot through joint encoders.

[0033] Acceleration data and angular velocity data are collected through the inertial measurement unit fixed to the end of the robotic arm.

[0034] Obstacle point cloud coordinate data is obtained by scanning the working area with LiDAR.

[0035] A unified time source is used to assign timestamps to the angle data, the angle change rate data, the acceleration data, the angular velocity data, and the obstacle point cloud coordinate data to generate the fused perception data set.

[0036] It should be noted that after the angle data, the angle change rate data, the acceleration data, the angular velocity data and the obstacle point cloud coordinate data are collected, a unified time source is used to assign an accurate timestamp to each data point to ensure that all data are aligned in the time dimension, thereby generating a fused perception data set containing timestamp-aligned joint angle data, angle change rate data, acceleration data, angular velocity data and obstacle point cloud coordinate data.

[0037] It should be further clarified that joint encoders are sensor devices used to measure the angular position and rate of change of robot joints. The characteristic attribute of angle data is a physical quantity measured in radians, and the characteristic attribute of angle rate data is a physical quantity measured in radians per second. These parameters are based on the standard sensor configuration for industrial robot control systems. An inertial measurement unit (IMU) refers to a sensor device fixedly attached to the end of a robotic arm and used to measure linear acceleration and angular velocity. The characteristic attribute of acceleration data is a physical quantity measured in meters per second squared, and the characteristic attribute of angular velocity data is a physical quantity measured in radians per second. These parameters are based on the dynamic load test specifications for industrial robots. LiDAR refers to a sensor device used to scan the environment and generate three-dimensional coordinate points. The characteristic attribute of obstacle point cloud coordinate data is a data structure, with each point containing X-, Y-, and Z-axis values. These parameters are based on the LiDAR factory calibration parameters. Timestamp alignment is the process of assigning each sensor data point the same time reference. Its characteristic attribute is a time synchronization mechanism, and its configuration is based on the real-time operating system clock synchronization protocol. A fused perception dataset refers to a structured data set formed by integrating all synchronized data. Its characteristic attributes are in database or array format and include timestamps, joint angle values, angle change rate values, acceleration values, angular velocity values, and point cloud coordinate values.

[0038] S2. Dynamic parameter set construction: construct a motion state vector based on the fused perception data set to generate a dynamic parameter set.

[0039] See also Figure 3 As shown, in a specific embodiment of the present invention, the specific process of generating a dynamic parameter set is: extracting angle data from the fused perception data set, and determining the real-time position of an end effector through forward kinematics calculation.

[0040] Based on the difference in the real-time position of the end effector at consecutive time stamps, a linear velocity vector and an end linear acceleration vector are calculated and generated.

[0041] The joint angular velocity change rate data in the fused perception data set are integrated to generate an angular acceleration vector.

[0042] The real-time position, the linear velocity vector, the terminal linear acceleration vector, and the angular acceleration vector are combined to generate the dynamic parameter set.

[0043] It should be noted that joint angle data, including the angle values ​​of the six-axis joints of the industrial robot, is extracted from the fused perception dataset. The real-time position of the end effector in three-dimensional space is determined through a forward kinematics calculation process. Forward kinematics refers to the mathematical calculation process that derives the end effector's spatial coordinates from the joint angles based on the robot's geometric structure. Next, the end effector position data from consecutive timestamps, such as the difference in end position coordinate values ​​between time points T0 and T1, is combined to calculate a linear velocity vector. The linear velocity vector is obtained by dividing the position difference by the time difference, and is measured in meters per second. The end effector linear acceleration vector is then generated based on the difference in linear velocity vectors from consecutive timestamps. The end effector linear acceleration vector is obtained by dividing the velocity difference by the time difference. Furthermore, the joint angular velocity change rate data from the fused perception dataset is integrated. This data represents the rate of angle change. By calculating the change in its consecutive timestamps, an angular acceleration vector is generated. The angular acceleration vector is obtained by dividing the angular velocity change rate difference by the time difference, and is measured in radians per second squared. Finally, the real-time position vector, linear velocity vector, terminal linear acceleration vector and angular acceleration vector are combined into a structured data set to generate a dynamic parameter set including position, velocity and acceleration.

[0044] It should be further explained that forward kinematics refers to the algorithmic process of calculating the position of the end effector in the Cartesian coordinate system using the robot's joint angle parameters. The setting is based on the link parameters provided by the robot manufacturer and the standard DH coordinate system modeling. The end effector refers to the tool device at the end of the industrial robot's mechanical arm used to perform tasks. The real-time position refers to the three-dimensional spatial coordinate value of the end effector at a specific point in time. The linear velocity vector refers to the set of components of the end effector's motion velocity in the X-axis, Y-axis, and Z-axis directions. The end linear acceleration vector refers to the set of components of the end effector's motion acceleration in the X-axis, Y-axis, and Z-axis directions. The angular acceleration vector refers to the set of components of the end effector's rotational acceleration on the rotation axis.

[0045] S3. Trajectory prediction input signal generation: parsing the dynamic parameter set to generate a trajectory prediction input signal.

[0046] In a specific embodiment of the present invention, the specific process of generating a trajectory prediction input signal is: extracting the terminal linear acceleration vector and the angular acceleration vector from the dynamic parameter set.

[0047] It should be noted that the terminal linear acceleration vector and angular acceleration vector are extracted from the dynamic parameter set. The terminal linear acceleration vector includes acceleration component values ​​in the X-axis, Y-axis, and Z-axis directions, and the angular acceleration vector includes acceleration component values ​​in the rotation axis direction.

[0048] The amplitudes of the terminal linear acceleration vector and the angular acceleration vector are calculated, and the amplitudes are mapped into a motion smoothness index.

[0049] It should be noted that the specific method of calculating the amplitude of the terminal linear acceleration vector and the angular acceleration vector is: the amplitude is obtained by the Euclidean norm formula, which is: ,in represents the amplitude of the terminal linear acceleration vector, represents the X-axis acceleration component, represents the Y-axis acceleration component, Represents the Z-axis acceleration component. The angular acceleration vector amplitude is calculated in a similar way. The formula is: ,in represents the angular acceleration vector amplitude, represents the acceleration component of the X-axis, represents the acceleration component of the Y rotation axis, Represents the acceleration component of the Z rotation axis.

[0050] According to the motion smoothness index, a smoothness threshold of a trajectory prediction algorithm is adjusted against a preset smoothness critical value.

[0051] It should be noted that the terminal linear acceleration vector amplitude and angular acceleration vector amplitude are mapped to the motion smoothness index, which is obtained by normalizing the comprehensive amplitude value. The formula is: ,in Represents the motion stability index, is the reference amplitude, the reference amplitude It is set to 1.0 m / s² and is obtained through calibration of the smooth operation test of an industrial robot. The lower the value of this indicator, the more unstable the movement.

[0052] It should also be noted that the specific method of adjusting the smoothness threshold of a trajectory prediction algorithm is: comparing the motion smoothness index with the preset smoothness critical value; if the motion smoothness index is less than the preset smoothness critical value, adjusting the smoothness threshold to the set value; if the motion smoothness index is greater than or equal to the preset smoothness critical value, keeping the smoothness threshold unchanged.

[0053] In a specific embodiment of the present invention, the preset stability critical value includes but is not limited to 0.8, and the set value includes but is not limited to 2.0. The preset stability critical value 0.8 and the set value 2.0 are determined based on multiple simulated trajectory conflict test experiments.

[0054] The motion smoothness indicator and the adjusted smoothness threshold are combined to generate the trajectory prediction input signal.

[0055] It should be noted that the trajectory prediction input signal is generated, which includes the motion smoothness index value, the adjusted smoothness threshold, and the original acceleration vector value, and is used to be input into the trajectory prediction process.

[0056] S4. Obstacle motion trajectory prediction: Generate an obstacle motion trajectory based on the trajectory prediction input signal and the fused perception data set.

[0057] In a specific embodiment of the present invention, the specific process of generating an obstacle motion trajectory is: extracting an obstacle point cloud coordinate sequence from the fused perception data set.

[0058] It should be noted that the obstacle point cloud coordinate sequence is extracted from the fused perception dataset. This sequence contains the three-dimensional spatial coordinate data points of the obstacles in the industrial robot's operating area. Each data point has a precise timestamp, forming a time-ordered coordinate set.

[0059] Applying the smoothness threshold in the trajectory prediction input signal, filtering is performed on the obstacle point cloud coordinate sequence to generate a filtered coordinate sequence.

[0060] It should be noted that the smoothness threshold in the trajectory prediction input signal is applied to the point cloud coordinate sequence filtering processing. The filtering processing uses a moving average algorithm to smooth the coordinate point sequence to reduce sensor noise. The smoothness threshold controls the size of the moving average window, and the window size is equal to the smoothness threshold multiplied by the basic window value.

[0061] In a specific embodiment of the present invention, the basic window value includes but is not limited to 3, and the setting basis is obtained by optimizing based on multiple measured data.

[0062] Based on the filtered coordinate sequence, the obstacle coordinate points in multiple future time steps are predicted by velocity vector interpolation to generate the obstacle motion trajectory.

[0063] It should be noted that the velocity vector of the obstacle is calculated based on the filtered point cloud coordinate sequence. The velocity vector is obtained by dividing the difference of the filtered coordinates of consecutive timestamps by the time difference. Finally, the velocity vector interpolation is used to predict the coordinate points of the obstacle trajectory in multiple time steps in the future. The prediction formula is: ,in Represents the coordinate points of the future obstacle trajectory, Represents the current filter coordinate point, represents the velocity vector, represents the time step, Represents an integer value of the number of future steps. Generates the obstacle trajectory output, which contains a sequence of 3D coordinates at multiple future time points.

[0064] S5. Collision risk map generation: combining the dynamic parameter set and the obstacle motion trajectory to generate a collision risk map.

[0065] In a specific embodiment of the present invention, the specific process of generating a collision risk map is: deducing a future trajectory of the end effector according to the linear velocity vector in the dynamic parameter set.

[0066] It should be noted that the linear velocity vector is extracted from the dynamic parameter set. The linear velocity vector represents the component values ​​of the end effector's motion velocity in the X-axis, Y-axis, and Z-axis directions. Then, the future trajectory of the end effector is deduced based on the linear velocity vector. The deduction process is achieved through the linear interpolation method. The formula is: ,in Represents the future trajectory coordinate point of the end effector, Represents the current end effector position coordinate point, represents the linear velocity vector, represents the time step, An integer value representing the number of steps in the future.

[0067] Perform spatial intersection detection on the future trajectory of the end effector and the obstacle motion trajectory to identify the intersection point.

[0068] It should be noted that the spatial intersection detection is performed on the deduced future trajectory coordinate points of the end effector and the obstacle trajectory coordinate points. The spatial intersection detection is performed by calculating the Euclidean distance between the end coordinate point and the obstacle coordinate point at each same future time step. The formula is: ,in Represents distance, Represents the end coordinate component, Represents the obstacle coordinate component, when Less than the safety distance threshold , it is judged as an intersection point.

[0069] In a specific embodiment of the present invention, the safety distance threshold The value includes but is not limited to 0.50 meters, which is calibrated through the industrial robot arm span size test.

[0070] The time window and position coordinates of the intersection points are marked to generate the collision risk map.

[0071] It should be noted that the time window and position coordinates of the intersection point are marked. The time window refers to the future time range corresponding to the intersection point, such as the time step N from 1 to 5, and the position coordinates are the three-dimensional spatial coordinate values ​​of the intersection point. A collision risk map is generated. The collision risk map is based on the spatial coordinate grid. The color depth of each grid unit represents the density of intersection points in the area. The higher the density, the greater the collision risk.

[0072] S6. Generate a path optimization objective function: Based on the collision risk map and the trajectory prediction input signal, adjust the path optimization weights to generate an objective function.

[0073] In a specific embodiment of the present invention, the specific process of adjusting the path optimization weight and generating an objective function is: extracting an area whose time window is less than a preset time window threshold from the collision risk map and defining it as an emergency area.

[0074] In a specific embodiment of the present invention, the preset time window threshold value includes but is not limited to 2 time steps, representing an area with a future time range of less than 0.20 seconds (the time step is fixed at 0.10 seconds), which is determined through an industrial robot emergency obstacle avoidance response time experiment.

[0075] It should be noted that the time window refers to the future time range of the intersection points marked in the collision risk map. The emergency area refers to the collection of areas in space where the collision risk density is higher than the average level and the time is urgent.

[0076] The number of the emergency regions is obtained and associated with the motion smoothness indicator in the trajectory prediction input signal.

[0077] When the motion smoothness index is lower than a preset smoothness threshold, a path smoothing weight is doubled.

[0078] It should be noted that when the motion smoothness index is lower than the preset smoothness critical value, the path smoothing weight is multiplied. The path smoothing weight refers to the parameter that controls the smoothness priority in path optimization. The default value is set to 1.0 and is adjusted to 2.0 after doubling. If the motion smoothness index is not lower than the preset smoothness critical value, the path smoothing weight maintains the default value of 1.0. The default value of 1.0 and the multiplication value of 2.0 are optimized based on multiple path smoothness tests.

[0079] The objective function is constructed by combining the number of emergency areas and the adjusted path smoothing weight.

[0080] It should be noted that the formula for constructing the objective function is: ,in represents the objective function value, Represents the safety weight, represents the path smoothing weight, Expressed as the number of emergency areas, Expressed as path length, Expressed as the path length for setting reference; safety weight Set as the number of emergency zones multiplied by the safety factor.

[0081] It should be further clarified that the path length refers to the total length of the spatial trajectory that the industrial robot traverses from its current position to its target position. This length is calculated by summing the spatial distances between all consecutive points on the path. The reference path length used here refers to the maximum possible path length within the industrial robot's operating area.

[0082] In a specific embodiment of the present invention, the safety factor may be set to 0.5, which is calibrated based on multiple simulated collision avoidance tests.

[0083] S7. Dynamic path replanning execution: Apply the objective function to execute dynamic path replanning and generate a replanned output path.

[0084] In a specific embodiment of the present invention, the specific process of generating a replanned output path is: starting from a current robot arm joint coordinate, initiating a Monte Carlo tree search.

[0085] It should be noted that the Monte Carlo tree search starts with the current robot arm joint coordinate data, which refers to the real-time six-axis joint angle values ​​obtained from the fused perception dataset. The Monte Carlo tree search process is initiated, which generates a sequence of candidate path nodes through random sampling, where each node represents a robot arm joint angle configuration.

[0086] It should be further explained that Monte Carlo tree search refers to an optimization algorithm that constructs a path decision tree through random sampling. Its characteristic attribute is a heuristic search strategy, which includes four stages: selection, expansion, simulation, and backtracking.

[0087] Under the constraints of the objective function, the collision risk cost and smoothness cost of the candidate path nodes are evaluated.

[0088] It should be noted that the collision risk cost and smoothness cost of each node extension are evaluated under the constraints of the objective function. The collision risk cost is obtained by querying the risk density value corresponding to the node end position coordinates in the collision risk graph, and the smoothness cost is obtained by calculating the Euclidean norm of the joint angle change between adjacent nodes. The formula is: in represents the smoothness cost, Represents the new node joint angle, represents the joint angle of the predecessor node, Indicates the node number, .

[0089] An optimal path is selected from the search results as the re-planning output path.

[0090] It should be noted that the collision risk cost and the smoothness cost are weighted and summed with the corresponding proportion weights to obtain the node comprehensive cost value, and the path with the lowest node comprehensive cost value is used as the re-planning output path.

[0091] In one specific embodiment of the present invention, when determining the weights corresponding to the collision risk cost and smoothness cost, the relationship and basis for determining these weights must be considered in accordance with the actual scenario requirements. If the industrial robot's working environment is densely populated with obstacles and the consequences of collisions are severe, such as in a precision instrument production workshop, then the collision risk cost should be given a higher weight, as avoiding collisions is crucial to ensuring production safety and equipment integrity. If the work scenario requires extremely high robot motion stability and efficiency, such as in high-speed sorting robot operations, then the smoothness cost may be given a higher weight to ensure the consistency and efficiency of the robot's motion. Ultimately, by appropriately setting weights, the two are weighted and summed to obtain the node's comprehensive cost value, and the path with the lowest comprehensive cost value is selected as the replanned output path.

[0092] S8. Joint control instruction conversion: convert the replanned output path to generate a set of joint control instructions.

[0093] In a specific embodiment of the present invention, the specific process of generating a set of joint control instructions is: discretizing the replanned output path into a time-space trajectory point sequence.

[0094] It should be noted that the replanned output path is discretized into a sequence of time-space trajectory points. The discretization process is achieved by dividing the path into equal time intervals. The time interval is fixed to the industrial robot control cycle of 0.10 seconds. Each trajectory point contains a time label and a spatial coordinate value.

[0095] It should be further explained that the time-space trajectory point sequence refers to a set of discrete points formed by dividing the path at fixed time intervals, and each point contains a timestamp and X-axis coordinates, Y-axis coordinates, and Z-axis coordinates.

[0096] The joint angle target value of each trajectory point in the time-space trajectory point sequence is solved by inverse kinematics.

[0097] It should be noted that the target joint angle value of each trajectory point is solved by inverse kinematics. Inverse kinematics refers to the mathematical calculation process of inversely calculating the robot joint angle based on the spatial coordinates of the end effector. The solution formula is Represents the target value of the joint angle, Represents the end space coordinates of the trajectory point, function The validity of dimensional conversion is ensured by the robot geometric model.

[0098] The joint angle target value is converted into a set of pulse width modulation instructions for driving the servo motor.

[0099] It should be noted that pulse width modulation instructions are generated to drive the servo motor to execute the trajectory, the joint angle target value is converted into a motor rotation angle instruction, and the voltage duty cycle is adjusted through pulse width modulation technology to control the motor speed.

[0100] Reference Figure 2 As shown, the second aspect of the present invention provides an AI-based industrial robot dynamic path planning system, including: a fusion perception data set generation module, a dynamic parameter set construction module, a trajectory prediction input signal generation module, an obstacle motion trajectory prediction module, a collision risk map generation module, a path optimization objective function generation module, a dynamic path replanning execution module and a joint control instruction conversion module.

[0101] The fused perception data set generation module is connected to the dynamic parameter set construction module, the dynamic parameter set construction module is connected to the trajectory prediction input signal generation module, both the trajectory prediction input signal generation module and the fused perception data set generation module are connected to the obstacle motion trajectory prediction module, both the dynamic parameter set construction module and the obstacle motion trajectory prediction module are connected to the collision risk map generation module, both the trajectory prediction input signal generation module and the collision risk map generation module are connected to the path optimization objective function generation module, the path optimization objective function generation module is connected to the dynamic path replanning execution module, and the dynamic path replanning execution module is connected to the joint control instruction conversion module.

[0102] The fused perception data set generation module obtains the real-time motion data and environmental perception data of the industrial robot and generates a fused perception data set.

[0103] The dynamic parameter set construction module constructs a motion state vector based on the fused perception data set to generate a dynamic parameter set.

[0104] The trajectory prediction input signal generating module analyzes the dynamic parameter set and generates a trajectory prediction input signal.

[0105] The obstacle motion trajectory prediction module generates an obstacle motion trajectory based on the trajectory prediction input signal and the fused perception data set.

[0106] The collision risk map generating module generates a collision risk map by combining the dynamic parameter set and the obstacle motion trajectory.

[0107] The path optimization objective function generation module adjusts the path optimization weights according to the collision risk map and the trajectory prediction input signal to generate an objective function.

[0108] The dynamic path replanning execution module applies the objective function to execute dynamic path replanning and generates a replanned output path.

[0109] The joint control instruction conversion module converts the replanned output path to generate a set of joint control instructions.

[0110] The above contents are merely examples and explanations of the concept of the present invention. Those skilled in the art may make various modifications or additions to the described specific embodiments or replace them in a similar manner. As long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, they should all fall within the scope of protection of the present invention.

Claims

1. The AI-based dynamic path planning method for industrial robots is characterized by: include: S1. Fusion perception dataset generation: Acquire the real-time motion data and environmental perception data of the industrial robot to generate a fusion perception dataset; S2. Dynamic parameter set construction: constructing a motion state vector based on the fused perception data set to generate a dynamic parameter set; S3, trajectory prediction input signal generation: parsing the dynamic parameter set to generate a trajectory prediction input signal; S4, obstacle motion trajectory prediction: generating an obstacle motion trajectory based on the trajectory prediction input signal and the fused perception data set; S5. Collision risk map generation: combining the dynamic parameter set and the obstacle motion trajectory to generate a collision risk map; S6. Generate a path optimization objective function: Adjust path optimization weights based on the collision risk map and the trajectory prediction input signal to generate an objective function; S7, dynamic path replanning execution: applying the objective function to execute dynamic path replanning and generating a replanned output path; S8. Joint control instruction conversion: convert the replanned output path to generate a set of joint control instructions.

2. The AI-based industrial robot dynamic path planning method according to claim 1, characterized in that: The specific process of generating a fusion perception data set is as follows: Collect the angle data and angle change rate data of the six-axis joints of the industrial robot through the joint encoder; Acceleration data and angular velocity data are collected through the inertial measurement unit fixed to the end of the robotic arm; Obtain obstacle point cloud coordinate data by scanning the operating area with LiDAR; A unified time source is used to assign timestamps to the angle data, the angle change rate data, the acceleration data, the angular velocity data, and the obstacle point cloud coordinate data to generate the fused perception data set.

3. The AI-based industrial robot dynamic path planning method according to claim 2, characterized in that: The specific process of generating a dynamic parameter set is as follows: extracting angle data from the fused perception dataset and determining a real-time position of an end effector through forward kinematics calculation; Calculating and generating a linear velocity vector and an end linear acceleration vector based on the difference in the real-time position of the end effector at consecutive time stamps; Integrating the joint angular velocity change rate data in the fused perception data set to generate an angular acceleration vector; The real-time position, the linear velocity vector, the terminal linear acceleration vector, and the angular acceleration vector are combined to generate the dynamic parameter set.

4. The AI-based industrial robot dynamic path planning method according to claim 3, characterized in that: The specific process of generating a trajectory prediction input signal is as follows: Extracting the terminal linear acceleration vector and the angular acceleration vector from the dynamic parameter set; Calculating the amplitudes of the terminal linear acceleration vector and the angular acceleration vector, and mapping the amplitudes into a motion smoothness index; adjusting a smoothness threshold of a trajectory prediction algorithm based on the motion smoothness indicator and against a preset smoothness threshold; The motion smoothness indicator and the adjusted smoothness threshold are combined to generate the trajectory prediction input signal.

5. The AI-based industrial robot dynamic path planning method according to claim 4, characterized in that: The specific process of generating an obstacle motion trajectory is as follows: Extracting an obstacle point cloud coordinate sequence from the fused perception dataset; Applying the smoothness threshold in the trajectory prediction input signal to perform filtering on the obstacle point cloud coordinate sequence to generate a filtered coordinate sequence; Based on the filtered coordinate sequence, the obstacle coordinate points in multiple future time steps are predicted by velocity vector interpolation to generate the obstacle motion trajectory.

6. The AI-based industrial robot dynamic path planning method according to claim 5, characterized in that: The specific process of generating a collision risk map is as follows: Deducing a future trajectory of the end effector based on the linear velocity vector in the dynamic parameter set; Performing spatial intersection detection on the future trajectory of the end effector and the motion trajectory of the obstacle to identify the intersection point; The time window and position coordinates of the intersection points are marked to generate the collision risk map.

7. The AI-based industrial robot dynamic path planning method according to claim 6, characterized in that: The specific process of adjusting the path optimization weight and generating an objective function is as follows: Extracting an area whose time window is smaller than a preset time window threshold from the collision risk map and defining it as an emergency area; Obtaining the number of the emergency regions and correlating it with the motion smoothness indicator in the trajectory prediction input signal; When the motion smoothness index is lower than a preset smoothness threshold, doubling a path smoothing weight; The objective function is constructed by combining the number of emergency areas and the adjusted path smoothing weight.

8. The AI-based industrial robot dynamic path planning method according to claim 2, characterized in that: The specific process of generating a replanning output path is as follows: Starting from the current robot arm joint coordinates, a Monte Carlo tree search is initiated; Under the constraints of the objective function, evaluating the collision risk cost and smoothness cost of the candidate path nodes; An optimal path is selected from the search results as the re-planning output path.

9. The AI-based industrial robot dynamic path planning method according to claim 8, characterized in that: The specific process of generating a set of joint control instructions is as follows: Discretizing the replanned output path into a time-space trajectory point sequence; Solving the joint angle target value of each trajectory point in the time-space trajectory point sequence by inverse kinematics; The joint angle target value is converted into a set of pulse width modulation instructions for driving the servo motor.

10. AI-based industrial robot dynamic path planning system, characterized by: include: The fusion perception data set generation module obtains the real-time motion data and environmental perception data of the industrial robot and generates a fusion perception data set; A dynamic parameter set construction module constructs a motion state vector based on the fused perception data set to generate a dynamic parameter set; a trajectory prediction input signal generating module, which parses the dynamic parameter set and generates a trajectory prediction input signal; an obstacle motion trajectory prediction module, which generates an obstacle motion trajectory based on the trajectory prediction input signal and the fused perception data set; a collision risk map generating module, combining the dynamic parameter set and the obstacle motion trajectory to generate a collision risk map; a path optimization objective function generation module, which adjusts path optimization weights based on the collision risk map and the trajectory prediction input signal to generate an objective function; a dynamic path replanning execution module, applying the objective function to execute dynamic path replanning and generating a replanned output path; The joint control instruction conversion module converts the replanned output path to generate a set of joint control instructions.

Images