Multi-target cooperative detection and obstacle avoidance planning method and system for industrial robot

Abstract

The present application relates to industrial robot multi-target cooperative detection and obstacle avoidance planning method and system, the present application detects robot and uses reinforcement learning to drive history perception quantum (RLDHQ) iterative strategy to solve the source seeking optimization model based on Gaussian function, and approaches the leakage source area; A double-index evaluation system containing leakage intensity and following distance cost is constructed, combined with Pareto sorting and dynamic weight summation, to determine the optimal following position for the processing robot; Further, three target optimization models of path length, collision risk and smoothness are constructed, and the RLDHQ strategy is used to solve the optimal virtual path; When the processing robot travels along the path, the environment is sensed in real time through the circumferential sensor, and the optimal safe direction is selected according to the collision detection to trigger obstacle avoidance.

Description

Technical Field

[0001] This invention belongs to the field of robot control and path planning technology, and in particular to a method and system for multi-target cooperative detection and obstacle avoidance planning of industrial robots. Background Technology

[0002] In industrial production and chemical industrial parks, leaks of toxic and hazardous gases and flammable and explosive liquids occur frequently. These accidents are characterized by rapid spread, high risk, and wide impact. Timely detection and treatment of leak sources are crucial to curbing leak spread and reducing safety risks. To mitigate the personal safety risks associated with manual source tracing and treatment, the use of autonomous mobile robots has become an important means of ensuring worker safety and improving task efficiency. Particle swarm optimization (PSO) is often used to solve robot source localization problems, but due to factors such as uneven field strength distribution at the leak source and multiple environmental interferences, this algorithm is prone to getting trapped in multiple local extrema, stagnating at local optima, and failing to accurately locate the core area of ​​the leak source, thus delaying treatment and exacerbating the accident. During leak source detection, accurate source location alone is insufficient to control the accident; timely treatment of the leak source is also necessary. To this end, treatment robots can be deployed to follow the detection trajectory of the source-tracing robot in real time, arriving at the leak source area simultaneously, achieving simultaneous "source tracing and treatment." Lightweight, small robots are highly maneuverable and energy-efficient, enabling them to quickly perform source tracing tasks. In contrast, leak source handling robots are bulky, lack maneuverability, and consume more energy. To reduce their operating energy consumption and shorten travel time, scientific global path planning is necessary to generate a safe, smooth, and short preset virtual path. Furthermore, complex industrial scenarios may contain dynamic obstacles such as mobile devices, and the environment is highly uncertain. Therefore, a robust real-time obstacle avoidance mechanism is required to prevent collisions and damage to the robot, secondary pollution caused by equipment malfunctions, and ultimately, more serious safety incidents.

[0003] Disadvantages of existing technology: 1. Traditional methods are difficult to adapt to two heterogeneous optimization problems, leakage source detection and global path planning, under the same architecture. On the other hand, they are prone to getting trapped in local optima during the solution process, and cannot take into account both the overall solution accuracy and convergence efficiency. 2. The collaborative following mechanism between the handling robot and the source-finding robot is imperfect and lacks corresponding following strategies, which in turn affects the timeliness of handling the leakage source; 3. The real-time obstacle avoidance module has low sensitivity to dynamic changes, making it difficult to meet the requirements for autonomous and safe robot movement in complex obstacle scenarios. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and propose a method and system for multi-target cooperative detection and obstacle avoidance planning of industrial robots. It deeply integrates quantum search and reinforcement learning, and solves the problems of source detection, path planning and dynamic obstacle avoidance in the same framework, effectively improving source finding accuracy, following synchronization, path smoothness and obstacle avoidance response capability.

[0005] The technical problem solved by this invention is achieved through the following technical solution: A method for multi-target cooperative detection and obstacle avoidance planning of industrial robots includes the following steps: Step 1: Construct a source-finding optimization model based on the superposition of Gaussian functions. The detection robot uses a reinforcement learning-driven history perception quantum iterative strategy to iteratively solve the source-finding optimization model, obtain its next optimal movement position, and gradually approach the leakage source region. Step 2: Construct a dual-index evaluation system with leakage intensity and following distance cost as the core, and based on the field strength and position information collected in real time by the detection robot, select the unique optimal following position for the processing robot by Pareto sorting combined with dynamic weight summation. Step 3: Construct a three-objective optimization model with the objectives of shortest path length, lowest collision risk, and optimal motion smoothness, and solve it using a reinforcement learning-driven history-aware quantum iterative strategy to generate a globally optimal virtual preset path for the robot from its current position to the optimal following position. Step 4: The robot moves along the virtual preset path, perceives the surrounding environment in real time, triggers obstacle avoidance behavior based on collision detection results, selects the optimal safe direction of travel through the circumferential sensor to avoid obstacles, and returns to the virtual preset path after obstacle avoidance.

[0006] Furthermore, the source-finding optimization model based on the superposition of Gaussian functions in step 1 is as follows: in, Indicates the spatial location of the leak field. Leakage intensity at the location; each Gaussian function term All are based on Radial basis functions centered at the center; Representing the The center of each basis function corresponds to the core coordinates of a signal peak in the environment; Indicates the first The width parameter of each basis function is used to control the diffusion range of the corresponding signal peak. The larger the width, the wider the coverage of the signal peak. It is the first The mixing coefficient, representing the _th ... The maximum field strength of a leakage source is determined by maximizing the objective function. The location with the maximum field strength is obtained by maximizing the objective function.

[0007] Furthermore, the dual-index evaluation system in step 2 is as follows: in, To process the robot's candidate next position, This is the current position of the processing robot. To process the robot's dual-index evaluation vector; The field strength guides the index, characterizing the leakage intensity at candidate locations. This is used to guide the processing robot towards the direction of the peak field strength. For the distance traveled, maximize This is equivalent to minimizing the physical distance between the processing robot and the detection robot; The specific implementation method of combining Pareto sort with dynamic weight summation is as follows: in, This is a comprehensive evaluation value based on two indicators. , These are the dynamic weights of the two indicators, satisfying... ; Indicates the current iteration number. Indicates the maximum number of iterations; express The initial reference value; Indicates the magnitude of the weight adjustment; This is the attenuation coefficient, used to control the rate of change of the weight.

[0008] Furthermore, the three-objective optimization model in step 3 is as follows: in, , , These represent the total path length, path collision risk, and path smoothness, respectively. , and They are , and The weights; Indicates the current iteration number. Represents the maximum number of iterations; , , It is a preset constant. With the number of iterations Linearly decreasing, and Mutual balance Fixed and unchanging.

[0009] Moreover, the dynamic obstacle avoidance in step 4 includes obstacle avoidance triggering, sensor status marking, and optimal direction selection; Among them, obstacle avoidance triggering is used to calculate the minimum net distance between the robot's current position and each obstacle within the sensor's sensing range. ,when ≤ When obstacle avoidance behavior is triggered, among which This is the robot's safe range; Sensor status labeling is based on multiple distance sensors evenly distributed around the robot's circumference. According to the obstacle's azimuth angle relative to the robot and the danger angle boundary, the sensor status in the corresponding detection direction is marked as dangerous. The optimal direction selection is used to filter the directions corresponding to all safety status sensors, calculate the distance from the candidate next position in each direction to the target endpoint, and select the direction closest to the endpoint as the optimal obstacle avoidance direction.

[0010] A planning system for a multi-objective cooperative detection and obstacle avoidance planning method for industrial robots includes a detection robot and a processing robot. The detection robot performs the task of detecting leakage sources, and the processing robot follows the detection robot and performs the task of handling leakage sources. Both the detection robot and the processing robot are equipped with a reinforcement learning-driven history-aware quantum iterative strategy solver to solve their respective multi-objective optimization problems. The reinforcement learning-driven history-aware quantum iterative strategy solver includes a quantum behavior search module and a Q-learning decision module. The quantum behavior search module is responsible for traversing the solution space and generating candidate solutions. The Q-learning decision module adaptively adjusts the compression and expansion coefficient and the historical information reuse strategy of the quantum behavior search module through state evaluation, action selection, reward feedback, and Q-value update mechanisms.

[0011] Moreover, the detection robot uses a source-finding optimization model based on the superposition of Gaussian functions to characterize the detection field strength of the leakage source and realize the collection of leakage intensity at the corresponding location.

[0012] Furthermore, the processing robot has a built-in dual-index evaluation system module and a multi-objective path planning module. The dual-index evaluation system module is used to determine the position of the target to be followed, and the multi-objective path planning module is used to generate the optimal virtual preset path.

[0013] Furthermore, the processing robot is equipped with multiple distance sensors, collision detection units, and obstacle avoidance direction decision units that are evenly distributed around the circumference to achieve dynamic obstacle avoidance.

[0014] The advantages and positive effects of this invention are: 1. The reinforcement learning-driven history-aware quantum (RLDHQ) iterative strategy proposed in this invention effectively overcomes the shortcomings of traditional particle swarm optimization algorithms, which are prone to getting trapped in local optima. Through ergodic search of quantum behavior and the adaptive decision-making capability of Q-learning, this optimization strategy can quickly locate the global optimum in complex field strength distributions. For example... Figure 2 The heatmap simulation results shown demonstrate that the detection robot quickly and accurately converges to the core region of the real leakage source from a random initial position within 30 iterations, proving that the proposed method has both high accuracy and fast convergence characteristics in the source exploration problem.

[0015] 2. For the handling robot, this invention innovatively constructs a dual-index dynamic weighted evaluation system integrating field strength guidance and position coordination. This mechanism can both guide the handling robot to approach the leakage source and effectively control the ineffective distance traveled in the early stages, reducing energy consumption. Figure 3 As shown, the processing robot's following trajectory is free of detours, and it arrives at the leak source location synchronously with the detection robot in the later stages of the iteration, ensuring that the leak source can be dealt with in a timely manner, avoiding the previous drawback of "finding but not dealing with it in time".

[0016] 3. This invention constructs an optimization model encompassing three objectives: path length, collision risk, and smoothness. It employs a dynamic weighting strategy, adaptively adjusting the optimization focus at different iteration stages (emphasizing path efficiency in the early stages and safety and stability in later stages). This results in a generated virtual preset path that is not only shorter and less energy-intensive, but also fundamentally guarantees the safety (effectively avoiding obstacles) and smoothness (reducing sharp turns) of the robot's movement. Figure 4 As shown, the purple dashed line path is smooth and avoids obstacle areas.

[0017] 4. The global obstacle avoidance mechanism designed in this invention, combined with circumferentially uniformly arranged sensors and a precise collision risk model, can achieve real-time perception and rapid response to sudden states of static obstacle boundaries and dynamic obstacles (including linear and circular motion). For example... Figure 4As shown, the robot can effectively avoid various types of dynamic obstacles while moving along a preset path, and autonomously return to its original path after safety, without any collisions occurring throughout the process. This significantly improves the robot's robustness and reliability in complex industrial scenarios.

[0018] 5. This invention integrates multiple sub-tasks with different characteristics, such as leak source detection, robot following, global path planning, and dynamic obstacle avoidance, into a unified multi-objective optimization framework centered on the RLDHQ strategy. This design avoids conflicts and fragmentation between optimization objectives of different tasks, achieves synergistic effects among various tasks, significantly improves the stability and task completion efficiency of the entire robot collaborative operation system, and meets the stringent application requirements of industrial scenarios. Attached Figure Description

[0019] Figure 1 This is a flowchart of the robot collaborative source finding and following task processing of the present invention; Figure 2 This is an iterative exploration source point heatmap according to an embodiment of the present invention; Figure 3 This invention provides a method for processing robot following trajectory diagrams in an embodiment of the invention. Figure 4 This invention provides a simulation of path planning and real-time obstacle avoidance in complex obstacle scenarios. Detailed Implementation

[0020] The present invention will be further described in detail below with reference to the accompanying drawings.

[0021] Planning systems for multi-target cooperative detection and obstacle avoidance planning methods in industrial robots, such as... Figure 1 As shown, it includes a detection robot and a processing robot. The detection robot is used to perform the task of detecting the leakage source, and the processing robot is used to follow the detection robot and perform the task of processing the leakage source. Both the detection robot and the processing robot are equipped with a reinforcement learning driven history-aware quantum (RLDHQ) iterative policy solver to solve their respective multi-objective optimization problems.

[0022] The detection robot employs a source-finding optimization model based on Gaussian function superposition to characterize the detection field strength of the leakage source, thereby acquiring the leakage intensity at the corresponding location. The processing robot incorporates a dual-index evaluation system module and a multi-objective path planning module. The dual-index evaluation system module determines the target position to follow, while the multi-objective path planning module generates the optimal virtual preset path. The processing robot is also equipped with multiple circumferentially distributed distance sensors, a collision detection unit, and an obstacle avoidance direction decision unit to achieve dynamic obstacle avoidance.

[0023] A method for multi-target cooperative detection and obstacle avoidance planning of industrial robots includes the following steps: Step 1: Initialize the probe robot and the processing robot.

[0024] Multiple lightweight detection robots are randomly distributed at their initial positions throughout the source-finding environment, enabling the robot swarm to cover a large detection range and ensuring the overall coverage of the initial search. The processing robot is placed at a preset fixed initial position, which is a safe standby area, waiting for the real-time field strength and position information of the detection robots, and preparing to follow synchronously.

[0025] Step 2: Construct a source-finding optimization model based on the superposition of Gaussian functions. The detection robot uses a reinforcement learning-driven history-aware quantum (RLDHQ) iterative strategy to solve the source-finding optimization model, obtain its next optimal movement position, and gradually approach the leakage source region.

[0026] The source-finding optimization model based on the superposition of Gaussian functions is as follows: in, Indicates the spatial location of the leak field. The leakage intensity at the location is determined by The weighted sum of Gaussian functions represents the attenuation law of the leakage source field strength spreading from the center to the surrounding areas; each Gaussian function term... All are based on Radial basis functions centered at the center; Representing the The center of each basis function corresponds to the core coordinates of a signal peak in the environment; Indicates the first The width parameter of each basis function is used to control the diffusion range of the corresponding signal peak. The larger the width, the wider the coverage of the signal peak. It is the first The mixing coefficient, representing the _th ... The maximum field strength of the leakage source is indicated by the value; a larger value indicates a higher field strength. In this leakage field, there is only one primary leakage source, while the others are secondary leakage points or interference signals. Therefore, by maximizing this objective function, the location with the maximum field strength can be determined, providing a basis for calculating the optimal next movement position for the detection robot.

[0027] Based on the leakage intensity obtained from its current location, the detection robot employs the reinforcement learning-driven history-aware quantum (RLDHQ) iterative strategy proposed in this invention to iteratively solve the source-finding optimization objective function to determine its optimal next move position. This strategy introduces a quantum behavior search mechanism, which perceives the optimization state in real time for each iteration. Through a built-in state evaluation mechanism and intelligent decision-making rules, it autonomously adjusts parameters and search strategies, effectively improving optimization accuracy and convergence speed.

[0028] The RLDHQ iterative strategy incorporates the core principles of quantum behavior imitation and Q-learning. The core principles of Q-learning include state evaluation mechanism, adaptive action design, action selection strategy, reward mechanism design, and Q-value update rules.

[0029] 2.1 Quantum Behavior Imitation.

[0030] Quantum behavior mimicry is the core principle of strategy implementation. Its main logic is to mimic the motion characteristics of quantum states, achieving traversal of the solution space without setting velocity parameters, and rapidly discovering potential optimal candidate positions within the search region. This mechanism, centered on individual position updates, guides individuals to quickly converge towards the globally optimal solution region. Its specific implementation method is as follows: 2.1.1 Obtain the individual's historical best position, the global best position, and the historical best weighted average position.

[0031] To guide individuals toward better solutions, a clear memory and update mechanism is needed. Specifically, it is necessary to acquire... Three types of key location information: individual historical best position, which is the best position experienced by a single individual in each search process; global best position of the group, which is the best solution among all the historical best positions of all individuals; and optimal weighted average position, which is the position vector obtained by weighted averaging of the historical best positions of all individuals. To facilitate subsequent iterative updates of individual positions, the specific methods for obtaining the above three types of key location information in the optimization problem are given below.

[0032] After each iteration, the current fitness of each individual is recalculated to dynamically preserve the individual's historical best position. in, Indicates the first The generation The historical best position of an individual For the first Individuals through the first Position after +1 iteration update The evaluation index represents the maximization (which can be one of the objective functions in a single-objective optimization problem, and the weighted sum of multiple objective functions in a multi-objective optimization problem).

[0033] Global optimal position We need to summarize the historical best positions of all individuals and select the optimal solution from them. , , This represents the total number of individuals.

[0034] The average optimal position is calculated by weighting the historical optimal positions of individuals. The weights are allocated based on the historical optimal evaluation values ​​of individuals, so that individuals with higher optimization performance contribute more to the average optimal position, thereby more accurately guiding individuals to converge towards the optimal region, while preserving the potential search value of individuals with lower evaluation values, taking into account both global exploration and local development. The specific weighting coefficients and the calculation of the average optimal position are as follows: in, for The generation The weighted coefficients corresponding to the historical best position of each individual. Indicates the first The average optimal position.

[0035] 2.1.2 Individual attractor calculation.

[0036] In the individual search process, the attractor is used to determine the central location of the search and is the core of guiding the individual's quantum state search. Based on the aforementioned individual historical optimal position and the group's global optimal position, a standard attractor is further constructed to connect the individual with the global optimal information. Specifically, by weighted fusion of the individual's historical optimal position and the global optimal position, a dynamic guiding point that takes into account both local development and global exploration is constructed. The detailed design method is as follows: in, For the first The standard attractor position of each individual determines the core direction of the individual search; It is a dynamic equilibrium coefficient, whose core function is to adjust the optimal position of an individual. and the global optimal position attractors The contribution weight is used to achieve a balance between local development and global exploration. To obey A uniformly distributed random disturbance factor is dynamically adjusted through random value selection. The magnitude of makes the balance coefficient random; and As The minimum and maximum values ​​are used to limit The range of values ​​is determined to avoid unconstrained and irregular values ​​that could lead to confusion in the attractor's guidance direction, ensuring that the attractor always plays a guiding role around the optimization goal.

[0037] The standard attractor The basic guide center is constructed based on the current iteration information. After integrating the Q-learning action selection mechanism, historical auxiliary attractors will be further constructed as another type of candidate guide center. Q-learning selects between standard attractors and historical auxiliary attractors based on the current search state to determine the attractor actually used during particle position update.

[0038] 2.1.3 Individual location update.

[0039] Based on the location information obtained from the aforementioned memory and update mechanism, an individual location update pattern is further constructed. This pattern, as the core execution carrier mimicking the motion characteristics of quantum states, balances the controllability of "moving towards high-quality regions" with the randomness of "preserving search diversity." The probability distribution derived from the potential well bound state forms the theoretical basis, and its construction is shown below: = in, The actual attractor position selected for the current iteration depends on the Q-learning action selection results. It is the length of the feature vector, used to control the search range around the attractor; Indicates the first In the nth iteration The current position vector of each individual; Random parameters for location sampling, through transformation Map it to a random variable that follows an exponential distribution; The compression-expansion coefficient is used to adjust the algorithm's global search and local exploitation. The larger the size, the wider the scope of exploration. The smaller the size, the more concentrated the scope of exploration, and its regulatory mechanism depends on the action selection in the intelligent decision-making rules of reinforcement learning; Indicates the number of oscillation segments. This indicates the segment number of the current iteration. It exhibits a decaying fluctuation pattern, and through the combination of segmented oscillation and iterative decay, a dynamic balance between global exploration and local convergence of the algorithm is achieved.

[0040] 2.2 Q-learning core integration principle.

[0041] Quantum search mechanisms exhibit good flexibility and unique advantages in optimization problems, and to some extent alleviate the problems of premature convergence and unstable convergence. However, they still rely on fixed search strategies, making it difficult to dynamically adapt to the differentiated needs of different optimization stages, and they cannot autonomously adjust their search behavior based on real-time iterative states. Reinforcement learning, through reward-driven mechanisms and real-time interaction with the optimization process, establishes a mapping relationship between rewards and actions, thereby providing a reliable basis for subsequent decisions and improving the adaptability of the strategy. Deeply integrating reinforcement learning with quantum search mechanisms can achieve adaptive search behavior, effectively compensating for the original shortcomings of quantum search mechanisms. Q-learning, as a typical reinforcement learning strategy, uses state, action, and reward as core elements. Through continuous interaction between the individual and the environment, it judges the merits of actions based on reward and punishment feedback, continuously updates the state-action value, and ultimately learns the optimal decision-making strategy.

[0042] 2.2.1 Status assessment mechanism.

[0043] The state evaluation mechanism is the core of achieving autonomous state perception in iterative processes. It is used to collect and quantify the characteristics of each optimization round in real time, transforming the abstract evolutionary process into quantifiable and decision-making state indicators, and providing accurate state characteristics for intelligent decision-making. This mechanism constructs a comprehensive state evaluation from three dimensions: convergence speed, population diversity, and iterative stagnation degree, as described below.

[0044] 2.2.1.1 Convergence speed evaluation.

[0045] First, an initial convergence rate is constructed based on the rate of change of the global optimal fitness value. Then, an exponential smoothing method is introduced to smooth the convergence index, weakening the evaluation interference caused by instantaneous fluctuations and ensuring the stability and accuracy of the convergence state evaluation. in, This is the original convergence rate index; This represents the convergence speed after exponential smoothing; to avoid the denominator being zero, a very small positive constant is set. This ensures the effectiveness of the convergence rate calculation; As a convergence speed smoothing coefficient, it is used to achieve a balance between sensitivity and stability. This is an indicator of the smooth convergence speed of the previous iteration.

[0046] 2.2.1.2 Population diversity assessment.

[0047] Population diversity reflects the degree of discrete distribution of individuals in the solution space. If population diversity is too low, it can easily lead to clustering of individuals, causing the algorithm to get stuck in local optima; if diversity is too high, it will reduce the convergence speed and delay the optimization process. Therefore, the normalized Euclidean distance between an individual's position and its average optimal position is used as an evaluation index to quantitatively assess population diversity. in, Indicators representing population diversity; Indicates the first t The middle generation Individual position With the t The average optimal position in the generation The Euclidean distance; This represents the maximum distance within the solution space and is used to normalize the diversity index.

[0048] 2.2.1.3 Evaluation of iterative stagnation.

[0049] This evaluation item is used to determine the risk of the algorithm getting stuck in a local optimum. State quantization is performed by counting the number of consecutive iterations in which the global optimal fitness has not been updated, combined with a preset stagnation threshold. in, For the first Stagnation quantification indicators for the next iteration; Indicates the current number of consecutive stalled iterations; The maximum tolerance threshold is used as a criterion for judging whether the previous algorithm has approached a local optimum.

[0050] By combining three types of indicators—convergence speed, population diversity, and iteration stagnation—the first method is constructed. The combined state vector of the next iteration To meet the requirements of discrete decision-making, the indicators are graded: convergence speed after smoothing. With population diversity All are divided into three levels: high, medium, and low; iterative stagnation degree The algorithms are divided into two levels: high and low. A high level indicates that the algorithm is at risk of iterative stagnation, while a low level indicates that the algorithm is in a normal optimization state.

[0051] 2.2.2 Adaptive motion design.

[0052] Based on the three types of state evaluation mentioned above, an integrated adaptive action strategy is constructed, comprising two parts: adaptive updating of compression and expansion parameters and reuse of temporal historical information. These two types of actions work together, relying on dynamic parameter adjustment to adapt to the population distribution and algorithm convergence characteristics, and leveraging historical information reuse to improve iterative stagnation, ultimately achieving autonomous control of the algorithm's optimization strategy. The specific action design is as follows: 2.2.2.1 Compression and expansion parameters Adaptive update rules.

[0053] The core function of the compression and expansion parameters is to balance global exploration and local exploitation. Their values ​​affect the optimization range and convergence efficiency, which is achieved through adaptive action design. Intelligent adjustment: in, Indicates compressibility and expansion parameters Single adjustment step size; limiting the range of parameter values. ,in , These represent the lower and upper limits of the compression and expansion parameters, respectively.

[0054] 2.2.2.2 Strategy for reusing time-series historical information.

[0055] This strategy fully utilizes historical optimization information during algorithm iteration to assist in updating particle positions. The reference value of historical information changes with the iteration sequence; recent best historical solutions better reflect the distribution characteristics of the current solution space and the optimal search direction, while the reference value of older historical solutions gradually decreases. Based on this characteristic, a differentiated weighting mechanism is adopted, assigning higher weights to recent historical information and lower weights to older historical information, thereby achieving efficient reuse of historical data.

[0056] First, build an external history archive: for the first... Individuals, building individual historical archives Used to store the individual's near The historical optimal solution for each generation; constructing a global historical archive for the entire population. Used to store population near The global optimal solution of the generation.

[0057] Secondly, a time-decreasing weight allocation rule is designed. According to the iteration time sequence from most recent to oldest, the historical solutions in the archive are sequentially assigned decreasing weights to highlight the role of recent historical information. The weight calculation method is as follows: in, The time sequence number representing the historical solution ( (This represents the index of the most recent historical solution in the current iteration); The weights corresponding to historical solutions; This is the weight adjustment parameter. This rule ensures that the more recent the historical solution, the greater the weight assigned.

[0058] Based on the aforementioned weighting rules, historical data are weighted and calculated. The optimal solutions in the individual and global historical archives are then summed using weighted methods to obtain the individual historical memory item and the population global memory item. Further integrating individual historical memory with population-wide memory, and taking into account both individual local optimization experience and population-wide optimization experience, we obtain the final historical auxiliary attractor: This historical auxiliary attractor Introducing multi-generational historical information broadens the search space for solutions and reduces the likelihood of particles getting trapped in local optima. History-assisted attractor. With formula Standard attractor as defined in Together, they create two search characteristics for the individual particle. When using the history-assisted attractor, the particle is in an exploratory state, tending to traverse the search space extensively to discover potentially better regions. When using the standard attractor, the particle is in a convergent state, tending to quickly move towards the globally optimal individual, accelerating local development. The selection of the two attractors is adaptively switched by Q-learning based on intelligent decisions made according to the current state evaluation index, thus balancing global optimization capability and local development efficiency.

[0059] 2.2.3 Action selection strategy.

[0060] Action selection strategy is the core execution link of reinforcement learning decision rules. Based on the current optimization state and the value information stored in the Q-value table, it adopts... - A greedy strategy enables intelligent action decision-making. First, based on the current state... Query the Q-value table for all available actions in this state, and construct an adaptive exploration probability that decays with the number of iterations. As shown below: in, and These represent the upper and lower bounds of the exploration probability, respectively. (The last part, "in terms of probability," appears to be a Select the action combination that maximizes the Q-value in the current state, and execute the optimal decision using learned experience; with probability Randomly select an action from all possible action combinations in the current state to break the limitations of existing strategies and discover better adjustment schemes. As the iteration process progresses, the probability is explored. Linear decay ensures sufficient exploration capabilities in the early stages, while focusing on the learned optimal strategy in the later stages.

[0061] 2.2.4 Reward Mechanism Design.

[0062] The reward mechanism is the core driver of reinforcement learning decision-making. It is used to quantitatively evaluate the merits of actions performed in the current state, guiding the population to continuously learn the optimal action strategy. This mechanism uses the change in the overall fitness of the population before and after the action is performed as the evaluation criterion, constructing a relative improvement in fitness value. The specific reward calculation is as follows: in, The reward value at the current iteration moment can be amplified by exponential transformation to increase the difference in fitness brought about by different actions, strengthen the positive incentive effect of high-quality actions, and improve the overall optimization efficiency.

[0063] 2.2.5 Q-value update rules.

[0064] Instant rewards output in conjunction with the reward mechanism The formula is updated iteratively using Q-learning, based on the current state. Selected action and the next iteration state Iteratively refine the state-action value to continuously optimize the action selection strategy. in, State in table Q With action The corresponding value estimate is used to assess the state. Next action The expected cumulative return that can be obtained; The learning rate controls the step size for updating the Q-value; The discount factor balances the weight of immediate rewards versus future accumulated rewards. The larger the value, the more emphasis is placed on long-term returns; For the next state The maximum Q value corresponding to all action combinations is used as the expected reward for future calculations.

[0065] 2.3 Cooperative Iterative Mechanism of Quantum Behavior Imitation and Q-learning.

[0066] 2.3.1 Initialization phase.

[0067] Randomly generate initial individuals within the feasible region and calculate the fitness of each individual. Initialize the individual's historical best position Initialize the global optimal position of the population. , and according to the formula Japanese style Calculate the average optimal position Initialize the state evaluation metric, i.e., the smooth convergence speed. (Mode Japanese style Population diversity (Mode and the degree of iteration stagnation (Mode ), forming the initial synthesis state vector Initialize the Q-value table A Q-value table is constructed based on the total number of possible state combinations in the state space and the total number of action combinations in the action space, and the value of all state-action pairs is initialized to zero.

[0068] 2.3.2 Iterative update phase.

[0069] Each iteration executes the following steps sequentially: State assessment: Based on current population information, the state assessment mechanism performs the assessment according to the formula... Obtain the comprehensive state vector of the current iteration. ; Q-learning action decision-making: based on the current state vector Query the Q-value table to obtain the Q-values ​​for all available actions in the current state, according to the formula. Get the exploration probability of the current iteration ,use - Greedy strategy for choosing actions This action combination includes the coefficient of compressibility and expansion. The two control commands are adjustment and attractor type switching; Individual position update: Update the expansion / contraction coefficient based on the control instructions output by Q-learning. And select the attractor type, according to the formula Complete the position iteration of all individuals to generate a new generation of population; Fitness assessment and historical information update: Calculate the fitness value of the new location and compare it with the historical best location, according to the formula. Update individual historical best position At the same time, update the global optimal position of the group. According to the formula Japanese style Update the average optimal position and simultaneously refresh the individual historical archives. and global history archive ; Reward calculation and Q-value update: Based on the change in the overall fitness of the population before and after the action is executed, according to the formula... Calculate instant rewards The effectiveness of the selected action combination is then quantitatively evaluated. Subsequently, according to the formula... Update the status in the Q value table With action Corresponding state-action value This enables continuous optimization of decision-making strategies.

[0070] Step 3: Construct a dual-index evaluation system with leakage intensity and following distance cost as the core, and based on the field strength and position information collected in real time by the detection robot, select the unique optimal following position for the processing robot by Pareto sorting combined with dynamic weight summation.

[0071] The dual-indicator evaluation system is as follows: in, To process the robot's candidate next position (current source-finding robot position); This indicates the current position of the processing robot; To process the robot's dual-index evaluation vector; As a field strength guiding index, it represents the field strength value at the candidate location. This is used to guide the processing robot towards the direction of the peak field strength. As a metric for travel distance, by maximizing This is equivalent to minimizing the physical distance between the processing robot and the detection robot. The core purpose is to reduce the unnecessary travel distance of the processing robot, reduce operating energy consumption, and avoid blindly following the location of the maximum field strength to increase unnecessary travel distance when the location of the leak source is uncertain in the early stage.

[0072] For the aforementioned dual-index optimization problem, a Pareto ranking combined with dynamic weight summation is used to select the unique optimal following position. First, all candidate positions are Pareto ranked to eliminate invalid candidate positions with "low field strength and long distance," resulting in a set of Pareto optimal solutions. This set of solutions is then transformed into a single comprehensive evaluation function using dynamic weight summation to calculate the unique optimal solution. in, This is a comprehensive evaluation value based on two indicators; , These are the dynamic weights of the two indicators, satisfying... ; Indicates the current iteration number. Indicates the maximum number of iterations; express The initial reference value; Indicates the magnitude of the weight adjustment; This is the attenuation coefficient, used to control the rate of change of the weights. This dynamic adjustment logic adapts to the robot's task flow. As it gradually approaches the core area of ​​the leak, the requirement for path length needs to be reduced, prioritizing field strength guidance to quickly reach the area to be treated, so that the leak source can be dealt with in a timely manner.

[0073] The processing robot makes comprehensive decisions based on the dual-index evaluation system of field strength guidance and position coordination of the source-finding robot, selects a unique solution, and determines its next target position to follow, thereby achieving precise synchronous movement with the detection robot.

[0074] Step 4: Construct an optimization model that includes three objectives: path length, collision risk, and motion smoothness. Use the RLDHQ iterative strategy to optimize the model and generate a globally optimal virtual preset path for the robot from its current position to the optimal following position.

[0075] For scenarios with known static obstacles, after the robot determines its next target position using a dual-index evaluation system, it needs to generate a globally optimal virtual preset path to ensure a safe, smooth, and low-energy arrival at that position. This path starts from the robot's current actual position and ends at the determined target position. To simplify calculations, this scenario simplifies the robot and all static obstacles into circular geometric models. Based on this geometric model, a three-dimensional multi-objective optimization problem is constructed, focusing on minimizing path length, minimizing collision risk, and optimizing motion smoothness. The specific objective function is defined as follows: in, Represents the set of all waypoints along the entire path. To process the robot's current position, It is the target destination. This represents the total number of path points. , , Let represent the total path length, path collision risk, and path smoothness, respectively. All three are optimization objectives that need to be minimized. Their specific expressions and physical meanings are as follows: in, Indicates the first Path points To the next waypoint The Euclidean distance of the path is calculated by summing the Euclidean distances of all adjacent path points. Minimizing this objective can achieve the "shortest" path, reducing the energy consumption and time cost of robot motion. Collision risk subfunction The expression is: in, This indicates that the path connects two consecutive points. and The Line segment; It is a parameter representing the area of ​​influence of the obstacle, which controls the decay rate of the risk function; A positive integer, determining the effective range of influence of the obstacle; where This is the robot's safe range; Represents path segment The minimum net distance to the surface of all static obstacles is calculated using the following expression: in, Represents path segment To the A static obstacle center The distance; Indicates the equivalent radius of the processing robot; For the first The equivalent radius of a static obstacle; This represents the total number of static obstacles. When the minimum net distance between the path segment and the obstacles exceeds the safe distance threshold... At that time, it was determined that there was no risk of collision on that section of road. When the minimum net distance is less than or equal to the safe distance threshold As the minimum net distance decreases, the collision risk increases exponentially. This design can accurately quantify the collision risk at different distances, ensuring that the path actively avoids dangerous areas and improving the safety of robot operation.

[0076] deflection angle The calculation method is as follows: in, It consists of three consecutive path points , and The resulting deflection angle, of which It is the dot product of two adjacent path vectors; It is the product of the magnitudes of the two path vectors; the total turning angle is obtained by summing the deflection angles corresponding to all path points. Minimizing this objective can significantly reduce the sharp turning angles in the path and improve the stability and controllability of the robot's motion.

[0077] To address the aforementioned three-objective optimization problem, a linear weighted summation method is used to transform it into a single-objective optimization problem for solution. This approach balances solution efficiency and optimization effectiveness, ensuring the rapid generation of optimal paths suitable for complex obstacle scenarios. Furthermore, by designing dynamic weighting factors, adaptive coordination between different optimization objectives is achieved, adapting to the path planning needs of different iteration stages. The specific formula is as follows: The update rule for the dynamic weights is as follows: in, , and These are the path length targets. Path collision risk target and path smoothness The weights; , , It is a preset constant. With the number of iterations Linearly decreasing, and Mutual balance Keeping it fixed means pursuing short paths in the early stages, that is, quickly finding roughly feasible routes to improve convergence speed; in the later stages, the requirement for length is reduced, and "safety" and "smoothness" are improved to make the path safer and more suitable for the actual movement of the robot.

[0078] To solve this multi-objective optimization problem, the aforementioned RLDHQ iterative strategy is employed for iterative optimization to obtain the optimal virtual path for the robot from its current position to its next target position. RLDHQ is essentially a general solution framework for objective optimization problems, which can be directly transferred to virtual path generation tasks. Individual positions are represented as a sequence of path points, and the iterative optimization process can be transformed into searching for the optimal set of path points in the feasible solution space. Therefore, the RLDHQ strategy, which has already demonstrated high accuracy and fast convergence in the leakage source optimization task, is also suitable for handling multi-objective path planning problems for robots, ensuring that the generated virtual path is safe, smooth, and efficient.

[0079] Step 5: The robot moves along the virtual preset path, perceives the surrounding environment in real time, triggers obstacle avoidance behavior based on collision detection results, selects the optimal safe direction of travel through the circumferential sensor to avoid obstacles, and returns to the virtual preset path after obstacle avoidance.

[0080] After obtaining the optimal virtual path, the processing robot uses this path as a reference to perform path-following movement. During movement, the sensor-equipped robot perceives the surrounding environment in real time. Based on real-time collision detection and obstacle avoidance control mechanisms, it returns to the virtual path after successfully avoiding obstacles to continue moving, thus ensuring that the entire movement process is safe, smooth, and does not deviate from the target trajectory. The specific mechanism is set as follows: 5.1 Obstacle avoidance trigger condition settings.

[0081] The core of collision detection is to calculate and process in real time the minimum net distance between the robot's current position and all obstacles within the sensor's detection range. And compare this distance with a preset safety threshold. Compare and determine whether obstacle avoidance behavior is triggered. Let the current position of the processing robot be... The robot's radius is The total number of obstacles in the global scene is , No. The center position of each obstacle is , The corresponding obstacle radius is Assume there are 1 / 2 obstacles within the sensor's detection range. , No. ( The positions and radii of the obstacles are respectively , The center-to-center distance between the robot and a single effective obstacle is calculated using Euclidean distance, and then the radii of the robot and the obstacle are subtracted to obtain the actual net distance. Traversing the sensor range For each valid obstacle, construct the corresponding net distance set. Take the minimum value of the distance set. As a basis for collision risk assessment. Construct obstacle avoidance trigger rules: when... When a collision risk is detected within the sensor's detection range, obstacle avoidance action is immediately triggered; when If the robot determines that it is in a safe state, it will continue to travel along the virtual path.

[0082] 5.2 Obstacle avoidance actions are executed.

[0083] 5.2.1 Sensor detection configuration.

[0084] Obstacle avoidance decisions are based on detection information collected by uniformly distributed distance sensors mounted on the robot, balancing obstacle avoidance safety and motion efficiency to select the optimal safe direction of travel. The sensors are evenly distributed around the robot's circumference, providing full coverage of the surrounding detection range, eliminating blind spots, and capturing obstacle position information in real time from all directions within the effective range of the sensors. To balance detection accuracy and computational complexity, a [missing information - likely a specific method or approach] is employed. Circumferential sensor, thus generating A uniformly distributed detection direction, with the angle between two adjacent sensors being [value missing]. , No. Road sensor ( The corresponding detection angle range is It iterates through each sensor, detecting whether there are obstacles within the effective coverage area of ​​its corresponding detection direction, and marks the safety status of that direction. Definition For the first The safety status indicator of the road sensor, if This indicates that there are no obstacles in the direction of detection; if This indicates the presence of an obstacle in that direction, posing a collision risk. Taking any obstacle within the sensor's detection range as an example, the center position of the obstacle is... The obstacle's azimuth angle relative to the robot is The dangerous angle boundary formed by the obstacle and the robot is , Let be the equivalent radius of the obstacle; therefore, the danger angle range corresponding to the obstacle can be obtained as follows: Mapping this angle range to the sensor number index, the starting and ending indices are calculated as follows: , index range Set all sensor status indicators to hazard signs. After traversing all obstacles within the sensor's detection range and completing the aforementioned marking, the safety state vectors for all detection directions can be obtained.

[0085] 5.2.2 Selection of the optimal obstacle avoidance direction.

[0086] Based on the sensor safety state vector, state identifiers are selected. The system detects all corresponding safe detection directions. For each safe direction, it obtains the robot's next candidate position and calculates the Euclidean distance from each candidate position to the target endpoint. The sensor direction closest to the endpoint is determined as the optimal obstacle avoidance direction. After the robot completes a single obstacle avoidance movement, collision monitoring and safe distance determination are performed again. If the obstacle avoidance condition is triggered again, the sensor status recognition and optimal travel direction selection process is repeated until the collision risk within the detection range is completely eliminated, at which point the robot resumes traveling along the preset virtual path.

[0087] As the processing robot safely travels along the virtual path to its current target location, the detection robot simultaneously completes a new round of field strength acquisition and optimal position calculation, continuously approaching the core area of ​​the leak source. The system repeatedly executes the above source tracing, following, path generation, and obstacle avoidance processes, enabling the processing robot to always follow the detection robot to achieve coordinated movement. Ultimately, the two robots arrive at the leak source location simultaneously, completing the coordinated detection and processing task.

[0088] Based on the above-mentioned multi-target cooperative detection and obstacle avoidance planning method and system for industrial robots, a complex simulation environment containing leakage sources and mixed dynamic and static obstacles is established, and unified parameter configuration is completed to verify the effectiveness of the proposed strategy in the construction problem.

[0089] Leakage source environmental parameter settings: The center coordinates of the three leakage sources are as follows: The width parameter is The leakage source strengths are respectively Based on the intensity of the leak sources, the third leak source has the highest intensity and is the primary search target. Its core leak location is... The simulation environment boundary constraints are lateral ranges. Longitudinal range .

[0090] Source tracing iteration parameter settings: Number of probe robots is The maximum number of iterations is .

[0091] Follow-up process parameter settings: The relevant parameters for balancing path length and leakage field strength are as follows , , .

[0092] Virtual path generation parameter settings: Number of individuals is The maximum number of iterations is Multi-objective correlation coefficient , , , , , .

[0093] Dynamic obstacle avoidance parameter settings: To simplify the motion model, the robot and all obstacles are treated as equivalent circular rigid bodies, and the robot body radius is processed accordingly. The total number of obstacles in the scene. The number of static obstacles The center coordinates are as follows: The corresponding radius is The dynamic obstacles are divided into two categories: linear motion obstacles and circular motion obstacles. There are a total of linear motion obstacles... Initial coordinates of the obstacle center set , radius is The speed of movement is The direction angle of motion is (The unit of measurement for direction angle is degrees, used to characterize the angle between the direction of motion and the horizontal axis); Circular motion obstacles total... Initial coordinates of the obstacle center set obstacle radius The fixed center position of circular motion Circular motion around the radius angular velocity of circular motion .

[0094] like Figure 2 As shown, , , , Two-dimensional heatmaps at four iteration points visually illustrate the search process of the probe robot in the target localization task. The horizontal axis in the figure (…) ) and ordinate ( The numbers () represent the horizontal and vertical coordinates of the robot in the exploration space, respectively. The orange area indicates a higher leakage field strength, while the black dots represent the positional distribution of individual robots. As the number of iterations increases, the individual robots quickly converge towards the true source point, clearly demonstrating the efficient convergence capability and reliable localization performance of the RLDHQ strategy in the source point exploration problem.

[0095] like Figure 3 As shown in the figure, the red line segment represents the trajectory of the processing robot following the source-finding robot. The hollow circles on the trajectory mark the positions of the processing robot at different iteration stages, while the solid blue circle represents the core location of the leak source finally located by the source-finding robot. The figure demonstrates that in the later stages of iteration, the processing robot can synchronously reach the leak source area following the source-finding robot, following a simple trajectory without unnecessary detours, enabling timely handling of the leak source.

[0096] like Figure 4As shown, the robot's motion states at six typical moments are captured. In the figure, the yellow solid square represents the robot's starting point, and the green solid pentagram represents the target endpoint; the orange solid circle represents static obstacles, the red solid circle represents dynamic obstacles moving in circles, and the blue solid circle represents dynamic obstacles moving in straight lines; the purple dashed line is the globally preset path generated based on the multi-objective optimization model, and the blue solid line is the robot's actual trajectory. As can be seen from the figure, during the robot's movement along the preset path, it can monitor the collision risk of static and dynamic obstacles in real time, promptly execute obstacle avoidance actions, adjust its movement direction to avoid obstacles, and then return to the preset path to continue moving towards the endpoint. The entire process is smooth and collision-free, verifying the effectiveness of the proposed path planning model and real-time obstacle avoidance mechanism. It can balance path efficiency and motion safety in complex obstacle scenarios, achieving the coordinated completion of path planning and dynamic obstacle avoidance.

[0097] It should be emphasized that the embodiments described in this invention are illustrative and not limiting. Therefore, this invention includes, but is not limited to, the embodiments described in the specific implementation. Any other implementation methods derived by those skilled in the art based on the technical solutions of this invention also fall within the scope of protection of this invention.

Claims

1. A multi-target cooperative detection and obstacle avoidance planning method for industrial robots, characterized in that: Includes the following steps: Step 1: Construct a source-finding optimization model based on the superposition of Gaussian functions. The detection robot uses a reinforcement learning-driven history perception quantum iterative strategy to iteratively solve the source-finding optimization model, obtain its next optimal movement position, and gradually approach the leakage source region. Step 2: Construct a dual-index evaluation system with leakage intensity and following distance cost as the core, and based on the field strength and position information collected in real time by the detection robot, select the unique optimal following position for the processing robot by Pareto sorting combined with dynamic weight summation. Step 3: Construct a three-objective optimization model with the objectives of shortest path length, lowest collision risk, and optimal motion smoothness, and solve it using a reinforcement learning-driven history-aware quantum iterative strategy to generate a globally optimal virtual preset path for the robot from its current position to the optimal following position. Step 4: The robot moves along the virtual preset path, perceives the surrounding environment in real time, triggers obstacle avoidance behavior based on collision detection results, selects the optimal safe direction of travel through the circumferential sensor to avoid obstacles, and returns to the virtual preset path after obstacle avoidance.

2. The multi-target cooperative detection and obstacle avoidance planning method for industrial robots according to claim 1, characterized in that: The source-finding optimization model based on the superposition of Gaussian functions in step 1 is as follows: in, Indicates the spatial location of the leak field. Leakage intensity at the location; each Gaussian function term All are based on Radial basis functions centered at the center; Representing the The center of each basis function corresponds to the core coordinates of a signal peak in the environment; Indicates the first The width parameter of each basis function is used to control the diffusion range of the corresponding signal peak. The larger the width, the wider the coverage of the signal peak. It is the first The mixing coefficient, representing the _th ... The maximum field strength of a leakage source is determined by maximizing the objective function. The location with the maximum field strength is obtained by maximizing the objective function.

3. The multi-target cooperative detection and obstacle avoidance planning method for industrial robots according to claim 1, characterized in that: The dual-index evaluation system in step 2 is as follows: in, To process the robot's candidate next position, This is the current position of the processing robot. To process the robot's dual-index evaluation vector; The field strength guides the index, characterizing the leakage intensity at candidate locations. This is used to guide the processing robot towards the direction of the peak field strength. For the distance traveled, maximize This is equivalent to minimizing the physical distance between the processing robot and the detection robot; The specific implementation method of combining Pareto sort with dynamic weight summation is as follows: in, This is a comprehensive evaluation value based on two indicators. , These are the dynamic weights of the two indicators, satisfying... ; Indicates the current iteration number. Indicates the maximum number of iterations; express The initial reference value; Indicates the magnitude of the weight adjustment; This is the attenuation coefficient, used to control the rate of change of the weight.

4. The multi-target cooperative detection and obstacle avoidance planning method for industrial robots according to claim 1, characterized in that: The three-objective optimization model in step 3 is as follows: in, , , These represent the total path length, path collision risk, and path smoothness, respectively. , and They are , and The weights; Indicates the current iteration number. Represents the maximum number of iterations; , , It is a preset constant. With the number of iterations Linearly decreasing, and Mutual balance Fixed and unchanging.

5. The multi-target cooperative detection and obstacle avoidance planning method for industrial robots according to claim 1, characterized in that: The dynamic obstacle avoidance in step 4 includes obstacle avoidance triggering, sensor status marking, and optimal direction selection; Among them, obstacle avoidance triggering is used to calculate the minimum net distance between the robot's current position and each obstacle within the sensor's sensing range. ,when ≤ When obstacle avoidance behavior is triggered, among which This is the robot's safe range; Sensor status labeling is based on multiple distance sensors evenly distributed around the robot's circumference. According to the obstacle's azimuth angle relative to the robot and the danger angle boundary, the sensor status in the corresponding detection direction is marked as dangerous. The optimal direction selection is used to filter the directions corresponding to all safety status sensors, calculate the distance from the candidate next position in each direction to the target endpoint, and select the direction closest to the endpoint as the optimal obstacle avoidance direction.

6. A planning system for the application of the multi-target cooperative detection and obstacle avoidance planning method for industrial robots as described in any one of claims 1 to 5, characterized in that: It includes a detection robot and a processing robot. The detection robot is used to perform the task of detecting the leakage source, and the processing robot is used to follow the detection robot and perform the task of processing the leakage source. Both the detection robot and the processing robot are equipped with a reinforcement learning-driven history-aware quantum iterative policy solver to solve their respective multi-objective optimization problems. The reinforcement learning-driven history-aware quantum iterative policy solver includes a quantum behavior search module and a Q-learning decision module. The quantum behavior search module is responsible for traversing the solution space and generating candidate solutions. The Q-learning decision module adaptively adjusts the compression and expansion coefficient and the historical information reuse strategy of the quantum behavior search module through state evaluation, action selection, reward feedback and Q-value update mechanism.

7. The planning system for the application of the multi-target cooperative detection and obstacle avoidance planning method for industrial robots according to claim 6, characterized in that: The detection robot uses a source-finding optimization model based on the superposition of Gaussian functions to characterize the detection field strength of the leakage source and realize the collection of leakage intensity at the corresponding location.

8. The planning system for the application of the multi-target cooperative detection and obstacle avoidance planning method for industrial robots according to claim 6, characterized in that: The processing robot has a built-in dual-index evaluation system module and a multi-objective path planning module. The dual-index evaluation system module is used to determine the position of the target to be followed, and the multi-objective path planning module is used to generate the optimal virtual preset path.

9. The planning system for the application of the multi-target cooperative detection and obstacle avoidance planning method for industrial robots according to claim 6, characterized in that: The processing robot is also equipped with multiple distance sensors, a collision detection unit, and an obstacle avoidance direction decision unit that are evenly distributed around the circumference to achieve dynamic obstacle avoidance.

Images