A control system and method for a chemical emergency manipulator based on intelligent sensing

Abstract

This invention relates to the field of robotic arm control technology, specifically a control system and method for a chemical emergency robotic arm based on intelligent perception. The system includes: a data acquisition unit for collecting environmental data from a chemical scenario and obtaining an initial depth image of the target area of ​​the chemical equipment to be operated; performing multi-sensor fusion processing on the environmental data and the initial depth image to obtain corresponding gas concentration information, obstacle location information, and image data of the target area; and a hazard assessment unit for inputting the gas concentration information, obstacle location information, and the current posture information of the robotic arm into a hazard assessment model to obtain the corresponding operational hazard level and the distance to the nearest obstacle. This invention uses a target recognition model to accurately obtain the location information of the target area and combines this with data such as the operational hazard level and obstacle location information to generate obstacle avoidance paths and operational instructions. This enables the robotic arm to intelligently avoid obstacles in complex environments while ensuring smooth operation.

Description

Technical Field

[0001] This invention relates to the field of robotic arm control technology, specifically to a control system and method for a chemical emergency robotic arm based on intelligent sensing. Background Technology

[0002] Currently, traditional methods typically rely on manual operation or fixed sensors to monitor the chemical environment, which cannot dynamically assess the level of danger in the environment in real time. This can easily lead to the omission of potential safety risks and increase the uncertainty in the operation process. Moreover, traditional robotic arm systems usually use simple path planning algorithms, which are difficult to effectively avoid obstacles in complex environments. When faced with dynamic obstacles or changing working conditions, this method is prone to collisions or misoperations, thereby affecting work efficiency and safety.

[0003] In addition, the force control systems in traditional methods are often relatively simple and cannot predict or adjust the contact force in real time. This means that during the grasping operation, the contact force may be too large or too small, which may cause the robot arm to damage the emergency object excessively or fail to grasp it effectively. Moreover, traditional grasping operations usually rely on manual or preset grasping modes and cannot automatically adjust the grasping strategy according to the target position and real-time force data. This can easily lead to low grasping accuracy, poor stability, or even misgrabbing or object slippage. Summary of the Invention

[0004] To achieve the above objectives, the present invention provides the following technical solution: a chemical emergency manipulator control system based on intelligent sensing, comprising:

[0005] The data acquisition unit is used to collect environmental data of the chemical scene and obtain the initial depth image of the target area of ​​the chemical equipment to be operated; the environmental data and the initial depth image are processed by multi-sensor fusion to obtain the corresponding gas concentration information, obstacle location information and image data of the target area.

[0006] The hazard assessment unit is used to input gas concentration information, obstacle location information and the current posture information of the robot into the hazard assessment model to obtain the corresponding operation hazard level and the distance to the nearest obstacle, and input image data into the pre-trained target recognition model to obtain the target location information corresponding to the target area;

[0007] The coordinate calculation unit is used to generate corresponding obstacle avoidance paths and operation execution instructions based on the operation hazard level, the distance to the nearest obstacle, and the preset safe speed model, and to calculate the position deviation based on the target position information and the current position coordinates;

[0008] The motion decision unit is used to input the operation hazard level, the distance to the nearest obstacle, the obstacle avoidance path, the operation execution command, the position deviation and the target position information into the motion decision model to determine the control parameters of the robot arm.

[0009] The force control prediction unit is used to collect contact force data of the end effector of the chemical emergency manipulator, input the contact force data into the pre-trained force control prediction model to obtain the predicted contact force of the end effector, and adjust the movement trajectory and contact force of the end effector according to the manipulator control parameters, position deviation and predicted contact force.

[0010] The end effector control unit is used to compare the position deviation with a preset deviation threshold in response to the end effector moving to a preset range of the target position; compare the contact force data with a preset force threshold in response to the position deviation being less than or equal to the deviation threshold; and generate a final grasping command by combining the target position information and the contact force data in response to the contact force data being less than or equal to the force threshold, and control the end effector to grasp the emergency operation object in the target area according to the final grasping command.

[0011] Preferably, multi-sensor fusion processing is performed on environmental data and initial depth images to obtain corresponding gas concentration information, obstacle location information, and image data of the target area, including:

[0012] The initial environmental data of the chemical scene is collected by the environmental perception module to obtain the initial gas concentration and initial obstacle coordinates corresponding to the initial environmental data. The background is removed from the initial depth image to obtain the image data corresponding to the target area.

[0013] Construct the corresponding concentration state vector and concentration state covariance matrix based on the initial gas concentration, determine the fusion gain matrix based on the concentration state vector and concentration state covariance matrix, and obtain the fused concentration vector.

[0014] Based on the initial obstacle coordinates, construct the corresponding position state vector and position state covariance matrix. Based on the position state vector and position state covariance matrix, determine the fusion gain matrix to obtain the fused position vector.

[0015] Data alignment is performed on the fused concentration vector and fused location vector to obtain gas concentration information and obstacle location information.

[0016] Preferably, gas concentration information, obstacle location information, and the current posture information of the robotic arm are input into the hazard assessment model to obtain the corresponding operational hazard level and the distance to the nearest obstacle. Image data is input into a pre-trained target recognition model to obtain the target location information corresponding to the target area, including:

[0017] Based on gas concentration information and obstacle location information, construct corresponding environmental feature vectors and environmental covariance matrices, and determine the hazard assessment gain matrix based on the environmental feature vectors and environmental covariance matrix;

[0018] The state update vector and state update covariance matrix are obtained based on the hazard assessment gain matrix. The hazard level of the operation is determined based on the state update vector, and the distance to the nearest obstacle is determined based on the state update covariance matrix.

[0019] Image data is input into a pre-trained target recognition model, and the contour features of the image data are extracted by the feature extraction layer in the target recognition model and input into the classification layer.

[0020] The contour features are classified by a classification layer to obtain the category label of the target region. The target location information of the target region is obtained by querying the preset coordinate mapping table based on the category label.

[0021] Preferably, based on the operation hazard level, the distance to the nearest obstacle, and a preset safe speed model, a corresponding obstacle avoidance path and operation execution instructions are generated. The position deviation is calculated based on the target location information and the current position coordinates, including:

[0022] The current position coordinates of the end effector collected by the position sensor of the chemical emergency manipulator are obtained, and the position deviation is calculated based on the current position coordinates and the target position information.

[0023] Determine whether the hazard level of the operation exceeds the preset hazard threshold. If the hazard level exceeds the preset hazard threshold, determine whether to activate emergency obstacle avoidance or allow the operation to continue based on the distance to the nearest obstacle.

[0024] If the nearest obstacle is greater than the preset safe distance, a command to continue operation is generated to control the robot to maintain its current posture. If the nearest obstacle is less than the preset safe distance, an obstacle avoidance path is generated based on the obstacle's position information to calculate the avoidance vector.

[0025] The operation will continue or the obstacle avoidance path will be used as the operation execution instruction.

[0026] Preferably, the operation hazard level, distance to the nearest obstacle, obstacle avoidance path, operation execution command, position deviation, and target position information are input into the motion decision model to determine the robot control parameters, including:

[0027] Determine whether the hazard level of the operation exceeds the preset hazard threshold. If the hazard level of the operation exceeds the preset hazard threshold and the distance to the nearest obstacle is less than the preset safe distance, use the obstacle avoidance path as the target trajectory to control the robot arm to perform obstacle avoidance actions.

[0028] In response to the fact that the hazard level of the operation does not exceed the preset hazard threshold, a movement correction amount is generated based on the position deviation and target position information, and a joint angle adjustment amount is generated in combination with the operation execution command;

[0029] The control parameters of the robot are determined based on the movement correction amount and the joint angle adjustment amount. The control parameters of the robot include the motor speed of the movement mechanism and the output pressure of the force control mechanism.

[0030] Preferably, contact force data of the end effector of the chemical emergency manipulator is collected, and the contact force data is input into a pre-trained force control prediction model to obtain the predicted contact force corresponding to the end effector, including:

[0031] Collect several contact force samples from the end effector at historical moments, filter these contact force samples, and obtain the contact force data;

[0032] The contact force data is input into a pre-trained force control prediction model, and the temporal features of the contact force data are extracted through the force control prediction model.

[0033] Based on the time-domain characteristics, the contact force at the next moment is predicted, and the predicted contact force is output. The predicted contact force is used to determine whether the robot arm has made contact with the emergency operation object.

[0034] Preferably, the movement trajectory and contact force of the end effector are adjusted according to the robot control parameters, position deviation, and predicted contact force, including:

[0035] The movement mechanism is adjusted according to the motor speed in the robot's control parameters, driving the end effector to move towards the target position, and the movement direction is corrected in real time according to the position deviation.

[0036] In response to the predicted contact force display showing a sharp increase in contact force, the downforce of the end effector is reduced in conjunction with the output pressure in the robot control parameters.

[0037] Preferably, in response to the end effector moving to a target position within a preset range, the position deviation is compared with a preset deviation threshold; in response to the position deviation being less than or equal to the deviation threshold, the contact force data is compared with a preset force threshold, including:

[0038] Obtain the real-time position coordinates of the end effector and calculate the distance between the real-time position coordinates and the target position information as the position deviation;

[0039] The absolute value of the position deviation is compared with the deviation threshold to obtain the comparison result;

[0040] If the position deviation is greater than the deviation threshold, the position deviation is recalculated and the movement continues. If the position deviation is less than or equal to the deviation threshold, the force control fine adjustment stage is entered.

[0041] In response to the position deviation entering the allowable range, the magnitude of the contact force data is monitored in real time;

[0042] The contact force data is compared with a preset force threshold. If the contact force data is greater than the force threshold, it is determined that the contact is too tight.

[0043] In response to the contact force data being greater than a preset force threshold, the difference between the contact force data and the force threshold is calculated, and a force reduction command is generated based on the difference to control the force control mechanism to reduce the output pressure.

[0044] Preferably, in response to contact force data being less than or equal to a force threshold, a final grasping command is generated by combining the target location information and the contact force data. Based on the final grasping command, the end effector is controlled to grasp the emergency operation object in the target area, including:

[0045] After confirming that the position deviation is less than or equal to the deviation threshold and the contact force data is less than or equal to the force threshold, the final gripping instruction is generated. The final gripping instruction includes the gripping angle and gripping force.

[0046] The final gripping command is sent to the gripping drive module of the end effector, which controls the gripping drive module to close the clamping mechanism;

[0047] During the closing process, contact force data is used for closed-loop feedback, which, together with the grasping force, stably grasps the emergency operation object and completes the chemical emergency operation.

[0048] A control method for a chemical emergency manipulator based on intelligent sensing, applicable to the aforementioned control system for a chemical emergency manipulator based on intelligent sensing, includes:

[0049] Collect environmental data from the chemical scene to obtain an initial depth image of the target area of ​​the chemical equipment to be operated; perform multi-sensor fusion processing on the environmental data and the initial depth image to obtain the corresponding gas concentration information, obstacle location information and image data of the target area.

[0050] Gas concentration information, obstacle location information, and current posture information of the robot are input into the hazard assessment model to obtain the corresponding operation hazard level and the distance to the nearest obstacle. Image data is input into the pre-trained target recognition model to obtain the target location information corresponding to the target area.

[0051] Based on the job hazard level, the distance to the nearest obstacle, and the preset safe speed model, the corresponding obstacle avoidance path and job execution instructions are generated, and the position deviation is calculated based on the target position information and the current position coordinates.

[0052] Input the operation hazard level, distance to the nearest obstacle, obstacle avoidance path, operation execution command, position deviation and target position information into the motion decision model to determine the control parameters of the robot arm;

[0053] The contact force data of the end effector of the chemical emergency manipulator is collected and input into a pre-trained force control prediction model to obtain the predicted contact force of the end effector. The movement trajectory and contact force of the end effector are adjusted according to the manipulator control parameters, position deviation and predicted contact force.

[0054] In response to the end effector moving to the target position within a preset range, the position deviation is compared with a preset deviation threshold; in response to the position deviation being less than or equal to the deviation threshold, the contact force data is compared with a preset force threshold; in response to the contact force data being less than or equal to the force threshold, a final grasping command is generated by combining the target position information and the contact force data, and the end effector is controlled to grasp the emergency operation object in the target area according to the final grasping command.

[0055] Compared with the prior art, the beneficial effects of the present invention are:

[0056] (1) Through multi-sensor fusion technology, the system can acquire and analyze the degree of danger in the chemical scene in real time and assess the level of danger of the operation in a timely manner. This intelligent perception can greatly improve the safety of operation and avoid risks caused by the complexity of the environment. Moreover, the target recognition model is used to accurately acquire the location information of the target area and combine the data such as the level of danger of the operation and the location information of obstacles to generate obstacle avoidance path and operation instructions. This enables the robot to intelligently avoid obstacles in complex environments and ensure the smooth operation.

[0057] (2) The present invention predicts and adjusts the contact force through the force control prediction unit. The system can adjust the movement trajectory and force of the manipulator according to the real-time contact force data to ensure that the end effector contacts the emergency operation object with appropriate force, avoiding misoperation due to excessive or insufficient force. Moreover, when the end effector reaches the target position and performs grasping, the system automatically generates a grasping command according to the target position information and contact force data. This not only improves the grasping accuracy, but also realizes closed-loop feedback control, ensuring stable grasping of the emergency operation object and reducing the possibility of human error. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the overall system architecture in one embodiment of the present invention;

[0059] Figure 2 This is a schematic flowchart of the overall method in one embodiment of the present invention.

[0060] In the diagram: 1. Data acquisition unit; 2. Hazard assessment unit; 3. Coordinate calculation unit; 4. Action decision unit; 5. Force control prediction unit; 6. End control unit. Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] Example 1, please refer to Figure 1 This invention provides a technical solution: a chemical emergency manipulator control system based on intelligent sensing, comprising:

[0063] Data acquisition unit 1 is used to collect environmental data of the chemical scene and obtain the initial depth image of the target area of ​​the chemical equipment to be operated; multi-sensor fusion processing is performed on the environmental data and the initial depth image to obtain the corresponding gas concentration information, obstacle location information and image data of the target area.

[0064] Hazard assessment unit 2 is used to input gas concentration information, obstacle location information and current posture information of the robot into the hazard assessment model to obtain the corresponding operation hazard level and the distance to the nearest obstacle, and input image data into the pre-trained target recognition model to obtain the target location information corresponding to the target area;

[0065] The coordinate calculation unit 3 is used to generate corresponding obstacle avoidance paths and operation execution instructions based on the operation hazard level, the distance to the nearest obstacle and the preset safe speed model, and to calculate the position deviation based on the target position information and the current position coordinates;

[0066] Motion decision unit 4 is used to input the operation hazard level, distance to the nearest obstacle, obstacle avoidance path, operation execution command, position deviation and target position information into the motion decision model to determine the control parameters of the robot arm;

[0067] Force control prediction unit 5 is used to collect contact force data of the end effector of the chemical emergency manipulator, input the contact force data into the pre-trained force control prediction model to obtain the predicted contact force of the end effector, and adjust the movement trajectory and contact force of the end effector according to the manipulator control parameters, position deviation and predicted contact force.

[0068] The end control unit 6 is used to compare the position deviation with a preset deviation threshold in response to the end actuator moving to a preset range of the target position; compare the contact force data with a preset force threshold in response to the position deviation being less than or equal to the deviation threshold; and generate a final grasping command by combining the target position information and the contact force data in response to the contact force data being less than or equal to the force threshold, and control the end actuator to grasp the emergency operation object in the target area according to the final grasping command.

[0069] It should be noted that environmental data collected in chemical scenarios; for example, if a chemical leak occurs, sensors will monitor gas concentrations, and depth cameras will acquire initial image data of the target area; this data includes the type and concentration of the gas, the location of obstacles in the scene, and images of the target area; this information is processed through multi-sensor fusion technology to obtain more accurate and comprehensive environmental information; for example, when a leak occurs in a piece of equipment in a chemical plant, the sensors collect the concentration of the leaked gas (such as toxic gas), and the depth camera captures images of obstacles (such as pipes and equipment) in the surrounding environment and the target area;

[0070] Gas concentration, obstacle information, and the current posture of the robotic arm are input into the hazard assessment model. The model will assess the hazard level of the current task (e.g., high, medium, or low risk). At the same time, image data is fed into a pre-trained target recognition model to identify the specific location or equipment that needs to be operated in the target area. For example, based on the gas concentration model, the system identifies the current gas leak hazard level as high. Meanwhile, the target recognition model determines that the robotic arm needs to operate the valve position of the equipment.

[0071] Based on the hazard level of the task, the distance to the nearest obstacle, and a safe speed model, the system generates an obstacle avoidance path. This means the robotic arm will avoid collisions with obstacles and perform the task at a safe speed. In addition, the system calculates the deviation between the current position and the target position to guide the robotic arm to move more precisely. For example, if the nearest obstacle is close to the robotic arm (e.g., 1 meter), the system will calculate a path to avoid the obstacle while instructing the robotic arm to slow down to ensure safe operation.

[0072] Information such as the hazard level of the operation, the distance to obstacles, the obstacle avoidance path, the operation execution instructions, and the positional deviation are input into the motion decision model. The model determines the control parameters of the robot (such as speed and direction) based on these inputs to ensure that the robot makes appropriate motion decisions in complex environments. For example, suppose the robot's task is to grab a valve in a leaking area. The system will decide whether the robot needs to move slower or whether the robot's motion strategy needs to be adjusted based on the hazard of the operation (such as the presence of toxic gases).

[0073] When the robotic arm comes into contact with an object in the target area, the system monitors the contact force of the end effector (such as a gripper) in real time. Through a force control prediction model, the system can predict the contact force that the end effector needs to apply. This helps to adjust the movement trajectory and force of the robotic arm's end effector to ensure the precise execution of the task. For example, when the robotic arm's end effector comes into contact with a valve that needs to be closed, the system predicts the required contact force to ensure that the robotic arm does not apply too much pressure (to avoid damaging the valve) or too little pressure (to prevent the valve from closing).

[0074] When the end effector of the robotic arm contacts the target object, the system compares the positional deviation with a set deviation threshold. If the deviation is less than or equal to the threshold, the system further compares the contact force data with a preset force threshold. If the contact force also meets the requirements, a final grasping command is generated, instructing the robotic arm to complete the grasping operation. For example, when the robotic arm contacts a valve, if the positional deviation is less than the set safety range and the contact force meets the requirements, the system issues a grasping command, and the robotic arm successfully closes the leaking valve, completing the emergency operation.

[0075] In an optional embodiment, multi-sensor fusion processing is performed on environmental data and an initial depth image to obtain corresponding gas concentration information, obstacle location information, and image data of the target area, including:

[0076] The initial environmental data of the chemical scene is collected by the environmental perception module to obtain the initial gas concentration and initial obstacle coordinates corresponding to the initial environmental data. The background is removed from the initial depth image to obtain the image data corresponding to the target area.

[0077] Construct the corresponding concentration state vector and concentration state covariance matrix based on the initial gas concentration, determine the fusion gain matrix based on the concentration state vector and concentration state covariance matrix, and obtain the fused concentration vector.

[0078] Based on the initial obstacle coordinates, construct the corresponding position state vector and position state covariance matrix. Based on the position state vector and position state covariance matrix, determine the fusion gain matrix to obtain the fused position vector.

[0079] Data alignment is performed on the fused concentration vector and fused location vector to obtain gas concentration information and obstacle location information.

[0080] It should be noted that the initial data collected in the chemical environment includes gas concentrations, obstacle locations, and depth images of the target area (i.e., images that reflect the depth information of the target area). To obtain accurate images of the target area, the depth images undergo background removal processing, filtering out irrelevant parts and leaving a clear image of the target area. For example, suppose a toxic gas leak occurs in a chemical plant. The environmental perception module collects gas concentration data through sensors and acquires images of the leak area through a depth camera. After background removal, only the location of the leak source remains in the image, while other irrelevant parts are ignored.

[0081] Based on the collected gas concentration information, a concentration state vector is constructed to represent the changes in gas concentration. Simultaneously, a covariance matrix of the concentration state is constructed to represent the uncertainty or error of this concentration data. Then, these two sets of data are used to calculate a fusion gain matrix to obtain a more accurate concentration estimate when fusing data from multiple sensors. For example, suppose the chlorine concentration detected by the sensors varies at different locations. The concentration state vector might include the gas concentration at each location, while the covariance matrix represents the uncertainty of this concentration data. For instance, the concentration data from sensor A might be more accurate, while the concentration data from sensor B might have a larger error; the covariance matrix helps the system understand these differences.

[0082] Similar to gas concentration, obstacle positions also need to be modeled. The system constructs a position state vector using the obstacle's position coordinates and calculates the covariance matrix of the position state. This data is also used to determine the fusion gain matrix, enabling more accurate estimation of obstacle positions when fusing sensor information. For example, suppose a robotic arm is operating a complex chemical plant, and sensors collect obstacle data (such as machinery, pipes, etc.) at different locations. Each sensor has a different error, so the covariance matrix reflects the accuracy of this position data. The system uses this data to adjust the robotic arm's path planning to avoid collisions with obstacles.

[0083] After processing the gas concentration and obstacle location information, the next step is to fuse these two sets of data. This involves aligning the concentration vector and the location vector to ensure they synchronously reflect the real-time state of the chemical environment. Ultimately, the system obtains the fused gas concentration and obstacle location information. For example, in a chemical scenario, gas concentration data and obstacle location data must be aligned at the same point in time to ensure the operating system can simultaneously monitor hazardous gas leaks and obstacles that may affect the robot's operation. For instance, if a region has a high gas concentration and also contains important equipment, the system will use the fused information to determine how to operate while ensuring safety.

[0084] In an optional embodiment, gas concentration information, obstacle location information, and the current posture information of the robotic arm are input into a hazard assessment model to obtain the corresponding operational hazard level and the distance to the nearest obstacle. Image data is input into a pre-trained target recognition model to obtain target location information corresponding to the target area, including:

[0085] Based on gas concentration information and obstacle location information, construct corresponding environmental feature vectors and environmental covariance matrices, and determine the hazard assessment gain matrix based on the environmental feature vectors and environmental covariance matrix;

[0086] The state update vector and state update covariance matrix are obtained based on the hazard assessment gain matrix. The hazard level of the operation is determined based on the state update vector, and the distance to the nearest obstacle is determined based on the state update covariance matrix.

[0087] Image data is input into a pre-trained target recognition model, and the contour features of the image data are extracted by the feature extraction layer in the target recognition model and input into the classification layer.

[0088] The contour features are classified by a classification layer to obtain the category label of the target region. The target location information of the target region is obtained by querying the preset coordinate mapping table based on the category label.

[0089] It should be noted that gas concentration information, obstacle location information, and the robot's current posture information are input into a hazard assessment model. This data helps the system assess the hazard level of the current working environment and determine the distance between the robot and obstacles. For example, suppose the robot is performing a chemical operation, the gas concentration sensor detects a high concentration of toxic gas in a certain area, the obstacle sensor detects a large piece of machinery nearby, and the robot's posture sensor provides feedback on the robot's current position and angle. The system uses this information as input for the next step of hazard assessment.

[0090] Based on gas concentration and obstacle location information, the system constructs an "environmental feature vector" to describe the safety of the current environment; the environmental covariance matrix represents the uncertainty of these feature data; based on this information, the system calculates a "hazard assessment gain matrix" for subsequent hazard assessment; for example, suppose a gas concentration sensor reports a gas concentration of 100 ppm in a certain area, and an obstacle location sensor reports that a device is 2 meters away from the robotic arm; by constructing the environmental feature vector, the system combines the gas concentration and obstacle location information to form a feature vector describing the current environment; the covariance matrix represents the reliability of these data, which may vary due to errors from different sensors;

[0091] The state is updated based on the hazard assessment gain matrix, resulting in a new state vector and covariance matrix. This updated data helps the system determine the current hazard level of the operation and the distance between the robot and the nearest obstacle. For example, suppose the system determines that the hazard assessment result is "high hazard" and the nearest obstacle is only 0.5 meters away through the state update. With this information, the system may issue a warning and adjust the robot's motion strategy to avoid collisions with obstacles, while also alerting the operator to the risk of gas leaks.

[0092] Image data of the target area is input into a pre-trained target recognition model. The model extracts contour features from the image data and performs category inference through a classification layer to identify the category of the target object. For example, suppose the camera of the robotic arm captures an image of the working area, which contains multiple objects. Through the target recognition model, the system extracts the contour features of the objects in the image and analyzes these features through a classification layer to determine whether the objects in the image are dangerous items or objects that the target task needs to process.

[0093] After identifying the object category, the system will look up the object's location information in a preset coordinate mapping table based on the object's category label. This helps the system accurately locate the target object and perform corresponding operations. For example, suppose the object recognition model determines that the object in the image is a valve and classifies it as a "valve" category. The system will then look up the location corresponding to the "valve" in the preset coordinate mapping table and obtain the accurate coordinate information of the valve in the working area. The robotic arm can then move precisely to the valve location to perform operations based on this location information.

[0094] In an optional embodiment, based on the job hazard level, the distance to the nearest obstacle, and a preset safe speed model, a corresponding obstacle avoidance path and job execution instructions are generated. The position deviation is calculated based on the target location information and the current position coordinates, including:

[0095] The current position coordinates of the end effector collected by the position sensor of the chemical emergency manipulator are obtained, and the position deviation is calculated based on the current position coordinates and the target position information.

[0096] Determine whether the hazard level of the operation exceeds the preset hazard threshold. If the hazard level exceeds the preset hazard threshold, determine whether to activate emergency obstacle avoidance or allow the operation to continue based on the distance to the nearest obstacle.

[0097] If the nearest obstacle is greater than the preset safe distance, a command to continue operation is generated to control the robot to maintain its current posture. If the nearest obstacle is less than the preset safe distance, an obstacle avoidance path is generated based on the obstacle's position information to calculate the avoidance vector.

[0098] The operation will continue or the obstacle avoidance path will be used as the operation execution instruction.

[0099] It should be noted that the current position coordinates of the end effector (the end of the robot) are obtained through the sensors of the chemical emergency robot, and the current position is compared with the target position to calculate the position deviation. The position deviation is the difference between the current position of the robot and the target position, which helps the system determine whether the robot needs to adjust its posture or path. For example, suppose the robot's goal is to close a leaking valve; the sensors show that the current position coordinates of the robot are (5,10), while the coordinates of the target valve are (7,12); the system will calculate the deviation between the current position and the target position and determine the direction and distance to be moved.

[0100] The system determines whether the current hazard level of the operation exceeds a preset hazard threshold. If the hazard level is too high, the system will consider whether to continue the operation or whether to activate emergency avoidance measures. For example, if the gas concentration is too high and the system detects that the concentration of toxic gas exceeds the preset safety threshold, the hazard level of the operation is determined to be "high". In this case, the system will decide whether to suspend the operation based on the hazard level and assess whether there are sufficient safety measures to avoid an accident.

[0101] The system determines whether to activate emergency obstacle avoidance based on the distance to the nearest obstacle. If the nearest obstacle is less than the preset safe distance, the system will calculate an obstacle avoidance path. If the obstacle is far away, the system can continue operating. For example, suppose the system detects that the nearest obstacle is 0.8 meters away, while the preset safe distance is 1 meter. Because the obstacle is too close, the system will choose to activate the obstacle avoidance path and adjust the movement trajectory of the robotic arm to avoid a collision.

[0102] When the nearest obstacle is less than a preset safe distance, the system calculates an avoidance vector based on the obstacle's position information and generates an obstacle avoidance path. The obstacle avoidance path allows the robot to bypass the obstacle and continue working safely. For example, suppose the robot encounters a device blocking its path while performing a task. Using the obstacle's position information from the sensors, the system calculates a path to bypass the device and instructs the robot to move along the new trajectory to avoid a collision with the device.

[0103] Based on the distance to obstacles and environmental conditions, the system will generate corresponding operation instructions. If the obstacle is far enough away, the robotic arm can continue operating. If the obstacle is too close, the system will generate new operation instructions based on the calculated obstacle avoidance path, guiding the robotic arm to avoid the obstacle. For example, if the system detects that the distance between the obstacle and the robotic arm is 1.5 meters, exceeding the preset safety distance, the robotic arm can continue performing its task. The system will issue a "continue operation" instruction, allowing the robotic arm to maintain its current posture and continue moving towards the target position. If the obstacle is less than the preset safety distance (e.g., 0.5 meters), the system will calculate an avoidance path, such as slightly deflecting the robotic arm or taking an alternative path to avoid colliding with the obstacle. The system will issue an "obstacle avoidance path" instruction, guiding the robotic arm to continue operating along the new path.

[0104] In an optional embodiment, the operation hazard level, distance to the nearest obstacle, obstacle avoidance path, operation execution command, position deviation, and target position information are input into the motion decision model to determine the robot control parameters, including:

[0105] Determine whether the hazard level of the operation exceeds the preset hazard threshold. If the hazard level of the operation exceeds the preset hazard threshold and the distance to the nearest obstacle is less than the preset safe distance, use the obstacle avoidance path as the target trajectory to control the robot arm to perform obstacle avoidance actions.

[0106] In response to the fact that the hazard level of the operation does not exceed the preset hazard threshold, a movement correction amount is generated based on the position deviation and target position information, and a joint angle adjustment amount is generated in combination with the operation execution command;

[0107] The control parameters of the robot are determined based on the movement correction amount and the joint angle adjustment amount. The control parameters of the robot include the motor speed of the movement mechanism and the output pressure of the force control mechanism.

[0108] It should be noted that the system checks whether the hazard level of the operation exceeds a preset hazard threshold. If the hazard level is too high, the system will determine whether obstacle avoidance is necessary based on the distance to obstacles. If the hazard level exceeds the threshold and the nearest obstacle is less than the safe distance, the system will generate an obstacle avoidance path to control the robotic arm to avoid the obstacle. For example, in a chemical plant, the robotic arm needs to enter an area where there may be a gas leak. The gas concentration sensor reports that the concentration is too high, and the system determines that the hazard level of the operation is "high," exceeding the safety threshold. The system also detects that there is an obstacle only 0.8 meters away from the robotic arm, while the preset safe distance is 1 meter. Since these two conditions (hazard level too high, obstacle too close) are met, the system decides to have the robotic arm avoid the obstacle and perform an obstacle avoidance maneuver.

[0109] When the system decides to perform an obstacle avoidance maneuver, it calculates an obstacle avoidance path based on the obstacle's location information and uses this path as the target trajectory. The robotic arm adjusts its movement path according to the target trajectory, bypasses the obstacle, and continues to perform its task. For example, in a chemical plant scenario, the robotic arm obtains the location of the obstacle through sensors and calculates a trajectory to bypass the obstacle. The system sends this trajectory to the robotic arm and instructs the robotic arm to continue working according to the new trajectory.

[0110] If the hazard level of the operation does not exceed the preset hazard threshold, the system will generate a correction amount based on the deviation between the current position and the target position (i.e., position deviation). The correction amount represents the displacement adjustment that the robot needs to make to correct the deviation of the current position from the target position. Based on the position deviation, the system will calculate the amount of motion that the robot should adjust. For example, suppose the target position of the robot is (7,12) and the current coordinates are (6,10). The system calculates the deviation between the current position and the target position, resulting in a horizontal offset of 2 meters and a vertical offset of 2 meters. Based on this information, the system generates a correction amount to instruct the robot to move to the target position.

[0111] Position corrections can be converted into joint angle adjustments; this is because the end effector of a robotic arm is controlled by joints (such as multiple joints in a robotic arm); the system calculates the angle that each joint should be adjusted so that the robotic arm can accurately reach the target position; for example, assuming the end effector of a robotic arm needs to be adjusted, the system calculates the angle adjustment of each joint (such as the shoulder, elbow, and wrist) based on the deviation between the target position and the current position, so that the entire robotic arm can move smoothly to the target position;

[0112] Based on the movement correction and joint angle adjustment, the system will ultimately determine the control parameters of the robot. These control parameters include: the motor speed of the moving mechanism: the motor speed of the robot controls the movement speed of the robot; the system calculates the required motor speed based on the movement correction so that the robot can move along a predetermined trajectory.

[0113] The force control mechanism controls the gripping force or contact force between the robotic arm and the object by adjusting the output pressure. The system adjusts the output pressure of the robotic arm according to the task requirements and the current working environment to ensure stability and safety during operation. For example, with a movement correction of 2 meters, the system calculates that the robotic arm needs a certain motor speed to cover this distance. If the goal is to complete the task as quickly as possible, the system may set a higher motor speed. If the robotic arm needs to grasp a fragile object (such as a glass bottle), the system will adjust the output pressure of the force control mechanism to ensure that the robotic arm does not exert excessive force to avoid damaging the object.

[0114] In an optional embodiment, contact force data of the end effector of a chemical emergency manipulator is collected, and the contact force data is input into a pre-trained force control prediction model to obtain the predicted contact force corresponding to the end effector, including:

[0115] Collect several contact force samples from the end effector at historical moments, filter these contact force samples, and obtain the contact force data;

[0116] The contact force data is input into a pre-trained force control prediction model, and the temporal features of the contact force data are extracted through the force control prediction model.

[0117] Based on the time-domain characteristics, the contact force at the next moment is predicted, and the predicted contact force is output. The predicted contact force is used to determine whether the robot arm has made contact with the emergency operation object.

[0118] It should be noted that when the end effector of the robotic arm (such as a gripper or tool head) performs a task, it comes into contact with the object, generating contact forces. The system collects data samples of these contact forces at multiple historical moments. These sampled data can come from force sensors, recording the specific force magnitude when the robotic arm contacts the target object. For example, suppose the robotic arm is handling a chemical reagent container. The force sensor records the force data when the robotic arm's end effector contacts the container at different times, specifically: 0.5N of force applied at the first moment, 0.7N of force applied at the second moment, 0.6N of force applied at the third moment, and so on. In this way, we obtain multiple sample data of contact forces.

[0119] Because the collected data may be affected by noise (such as sensor errors, environmental changes, etc.), it is necessary to filter the data to remove these unnecessary interferences. The purpose of filtering is to smooth the data, making the contact force signal more accurate and reliable. For example, suppose that when collecting contact force data, due to occasional sensor jitter, a certain data point (e.g., a force value of 1N) may be an error value. By filtering (such as using a moving average method), such abrupt changes can be removed, making the contact force data more stable. For example, after filtering, the original 1N becomes 0.6N, and such data is more representative of the actual contact situation of the robot arm.

[0120] Once the filtered contact force data is obtained, the system inputs this data into a pre-trained force control prediction model. This prediction model learns the patterns in historical contact force data, extracts the temporal features of the data (such as trends and periodicity), and predicts the contact force based on these features. For example, in historically collected data, the force changes of the robot arm contacting the container show certain patterns; for instance, the force gradually increases over a period of time, indicating that the robot arm is making closer contact with the target. By inputting this data into the force control prediction model, the model can identify the pattern of contact force changes over time.

[0121] Temporal features refer to the characteristics of contact force data over time, including trends, oscillation periods, and increase / decrease patterns. Force control prediction models analyze these temporal features to better understand the patterns of contact force changes and predict contact forces in the future. For example, when handling a relatively hard container, a force control prediction model might detect a gradual increase in contact force, indicating that the robotic arm is applying more pressure to the container. For softer objects, the force might stabilize after reaching a certain threshold. The prediction model analyzes these patterns and extracts useful temporal features, such as the trend of gradually increasing contact force.

[0122] Once the temporal features are extracted, the predictive model uses these features to predict the contact force at future moments. In this way, the system can know in advance the force that the robotic arm may apply at the next moment. For example, suppose the force control model discovers from historical data that the contact force reaches 1N after a certain period of contact and gradually stabilizes. In subsequent predictions, the model may predict that the contact force at the next moment will be 1N. This predicted value can help the system determine whether it has made contact with the target.

[0123] The system determines whether the robotic arm has made contact with the target object based on the predicted contact force value. This determination usually relies on a preset contact force threshold. If the predicted contact force is greater than a certain threshold, it means that the robotic arm has made contact with the object and may need to perform further operations (such as emergency operations). For example, suppose that in an emergency operation, the robotic arm needs to grab a container. If the contact force reaches 1N, it means that the robotic arm has made correct contact with the container. If the model predicts that the contact force will reach 1N, the system will determine that the robotic arm has made contact with the target object and then perform relevant follow-up operations, such as continuing the operation or initiating an emergency procedure.

[0124] In an optional embodiment, adjusting the movement trajectory and contact force of the end effector based on robot control parameters, positional deviation, and predicted contact force includes:

[0125] The movement mechanism is adjusted according to the motor speed in the robot's control parameters, driving the end effector to move towards the target position, and the movement direction is corrected in real time according to the position deviation.

[0126] In response to the predicted contact force display showing a sharp increase in contact force, the downforce of the end effector is reduced in conjunction with the output pressure in the robot control parameters.

[0127] It should be noted that the end effector of the robotic arm achieves position control through a motor-driven moving mechanism. The motor speed determines the movement speed of the end effector. Based on the motor speed in the control parameters, the system can adjust the movement trajectory of the robotic arm's end effector to accurately guide it to the target position. For example, suppose the robotic arm is moving an object from one place to another. In the control parameters, the motor speed is set to a moderate speed so that the end effector can move smoothly to the target position. If the target position is far away, the motor speed may need to be adjusted to a higher speed to ensure efficient task completion; if the target position is close, the motor speed will be reduced to prevent errors caused by excessively fast movement.

[0128] During the movement of the robotic arm, a positional deviation will occur between the current position and the target position of the end effector. To ensure that the end effector accurately reaches the target position, the system will detect and correct the positional deviation in real time. If the end effector is found to have deviated from the target position, the system will adjust the direction of movement to guide it back to the correct direction. For example, if the end effector of the robotic arm is grasping an object, but due to external interference (such as vibration, friction, etc.), the positional deviation occurs, and the end effector may shift to the left or right of the target position. In this case, the system will automatically correct the direction of movement by detecting the positional deviation in real time (for example, if it detects a deviation of 2cm from the target position), guiding the end effector of the robotic arm back to the correct trajectory.

[0129] Predictive contact force models can anticipate the force values ​​when a robotic arm's end effector contacts an object. If the predicted contact force increases sharply (indicating that the contact force may be too large), the system needs to respond to the contact force of the robotic arm to avoid excessive pressure that could damage the target object. For example, suppose the robotic arm's end effector is grasping a fragile glass container. During contact, the force control prediction model predicts that the contact force may increase rapidly, exceeding the set safety range. To avoid damaging the container, the system will automatically react by slowing down the increase in downward pressure, preventing excessive force.

[0130] Based on the output pressure in the control parameters of the robotic arm (usually referring to the maximum or current pressure that the motor or hydraulic system can output), if the predicted contact force indicates excessive pressure, the system will reduce the downward pressure of the end effector in a timely manner, thereby reducing the pressure applied by the robotic arm to the target object. For example, suppose the end effector of the robotic arm is grasping a chemical reagent bottle, and the output pressure in the control parameters is set to a high value (e.g., 500N). If the force control model predicts a sharp increase in contact force, meaning that the robotic arm is applying excessive pressure, the system will automatically reduce the output pressure based on the prediction result, adjust the motor speed and hydraulic system settings to reduce the pressure applied by the end effector, and ensure that the bottle is not crushed.

[0131] In an optional embodiment, in response to the end effector moving to a target position within a preset range, the position deviation is compared with a preset deviation threshold; in response to the position deviation being less than or equal to the deviation threshold, the contact force data is compared with a preset force threshold, including:

[0132] Obtain the real-time position coordinates of the end effector and calculate the distance between the real-time position coordinates and the target position information as the position deviation;

[0133] The absolute value of the position deviation is compared with the deviation threshold to obtain the comparison result;

[0134] If the position deviation is greater than the deviation threshold, the position deviation is recalculated and the movement continues. If the position deviation is less than or equal to the deviation threshold, the force control fine adjustment stage is entered.

[0135] In response to the position deviation entering the allowable range, the magnitude of the contact force data is monitored in real time;

[0136] The contact force data is compared with a preset force threshold. If the contact force data is greater than the force threshold, it is determined that the contact is too tight.

[0137] In response to the contact force data being greater than a preset force threshold, the difference between the contact force data and the force threshold is calculated, and a force reduction command is generated based on the difference to control the force control mechanism to reduce the output pressure.

[0138] It should be noted that the system tracks the position of the end effector in real time and compares its current actual position coordinates with the target position. By calculating the distance between the two, a position deviation value is obtained, which represents the distance between the end effector and the target position. For example, suppose the robot's end effector is placing an object in a precise location. The system obtains the current position of the end effector in real time. For instance, if the current coordinates of the end effector are (5,5) and the target position is (5,7), the distance between the two is 2 units. Therefore, the position deviation is 2 units.

[0139] A preset deviation threshold is set, which defines the maximum allowable deviation range. If the position deviation is less than or equal to this threshold, the end effector is considered to be close to the target position; otherwise, the position needs to be adjusted further. For example, if the deviation threshold set by the system is 1 unit and the current position deviation is 2 units (as described above), then the position deviation is greater than the threshold, and the system needs to continue adjusting the position of the end effector until the deviation is less than or equal to 1 unit.

[0140] When the position deviation is less than or equal to a preset threshold, it indicates that the end effector is very close to the target position. At this time, the system will enter the contact force control stage, and ensure that the target object is not excessively squeezed by finely adjusting the applied force. For example, suppose that when the position deviation is less than or equal to 1 unit, the end effector has made precise contact with the target object (e.g., placing a bottle). At this time, the system no longer focuses on the change in position, but switches to the force control stage to ensure precise adjustment of the contact force.

[0141] Once the end effector contacts the target object, the system monitors the contact force data in real time. The contact force represents the pressure applied between the robot and the target object. The system compares this contact force with a preset force threshold to determine whether the applied force is too large. For example, suppose the robot's end effector is grasping a precision electronic component, and the system sets the force threshold to 10N. The system detects a contact force of 12N in real time, which exceeds the set threshold.

[0142] If the contact force exceeds the preset force threshold, the system will calculate the difference between the contact force and the force threshold, and generate an instruction to reduce the applied force to avoid damage to the target object. The system will adjust the output pressure of the end effector to ensure that the force control fine adjustment is within a safe range. For example, if the contact force is 12N and the preset force threshold is 10N, the difference is 2N. The system will calculate the pressure that needs to be reduced based on this difference, and reduce the output pressure by adjusting the control instruction, such as reducing the output of the motor or slowing down the force of the hydraulic system, until the contact force drops below 10N.

[0143] In an optional embodiment, in response to contact force data being less than or equal to a force threshold, a final grasping command is generated by combining target location information and contact force data. Based on the final grasping command, the end effector is controlled to grasp the emergency operation object in the target area, including:

[0144] After confirming that the position deviation is less than or equal to the deviation threshold and the contact force data is less than or equal to the force threshold, the final gripping instruction is generated. The final gripping instruction includes the gripping angle and gripping force.

[0145] The final gripping command is sent to the gripping drive module of the end effector, which controls the gripping drive module to close the clamping mechanism;

[0146] During the closing process, contact force data is used for closed-loop feedback, which, together with the grasping force, stably grasps the emergency operation object and completes the chemical emergency operation.

[0147] It should be noted that before performing a gripping operation, two conditions are checked first: whether the positional deviation of the end effector is within the allowable range; and whether the contact force of the end effector meets the safety standard (i.e., less than or equal to the force threshold). For example, suppose the robotic arm's end effector is gripping a chemical bottle, with the goal of safely moving the bottle from a table to a storage location. The system first ensures that the end effector is close to the target position of the bottle and confirms that the contact force is not too large to avoid damaging the bottle. If the positional deviation is 0.8 units and the contact force is 3N, while the preset deviation threshold is 1 unit and the force threshold is 5N, then the gripping conditions are met.

[0148] After the above two conditions are met, the system will generate the final gripping instruction based on the current position and contact force. This instruction includes: the gripping angle, which is the angle at which the robotic arm gripper needs to be positioned to ensure that it can correctly grasp the target object; and the gripping force, which is the force that the robotic arm needs to apply, which cannot be too large to damage the object, nor too small to ensure a stable grip. For example, assuming the target bottle is round, the system determines the gripping angle to be 45 degrees and the gripping force to be 4N. The final gripping instruction contains these parameters to ensure that the robotic arm does not apply excessive force during the gripping process and crush the bottle.

[0149] After generating the gripping command, the system transmits it to the gripping drive module of the end effector; the gripping drive module controls the movement of the clamping mechanism, closes the clamp according to the command, and begins the gripping process; for example: the system controls the clamp to close through the command, the clamp is correctly aligned with the bottle at a 45-degree angle, and gradually closes with a force of 4N; the clamp begins to consciously and slowly contact the surface of the bottle.

[0150] During the clamp closing process, the system monitors contact force data in real time. This data serves as feedback information, and the clamp's movements are adjusted through closed-loop control to ensure appropriate gripping force and a stable gripping process, preventing excessively tight or loose gripping. For example, when the clamp begins to contact the bottle, the system monitors the contact force in real time. If the force sensor detects that the contact force reaches the preset 4N, the system confirms that the gripping force is stable. If the contact force exceeds 4N, the system automatically adjusts the force, reducing the output to prevent damage to the bottle. If the contact force is too small, the system increases the force to ensure that the clamp firmly grips the bottle.

[0151] Through this precise grasping control process, the end effector can stably grasp the target object and complete the chemical emergency operation. The system avoids dangers caused by excessive positional deviation or excessive contact force through fine control, ensuring the safe and smooth operation. For example, in a chemical emergency operation, after the end effector successfully grasps the bottle, it can safely move the bottle to the designated location, avoiding the risk of chemical leakage or breakage. The entire process ensures the accuracy and stability of the operation through real-time feedback and control.

[0152] Example 2, please refer to Figure 2 This invention provides a technical solution: a control method for a chemical emergency manipulator based on intelligent sensing, applicable to the aforementioned control system for a chemical emergency manipulator based on intelligent sensing, comprising:

[0153] S1. Collect environmental data of the chemical scene and obtain the initial depth image of the target area of ​​the chemical equipment to be operated; perform multi-sensor fusion processing on the environmental data and the initial depth image to obtain the corresponding gas concentration information, obstacle location information and image data of the target area.

[0154] S2. Input the gas concentration information, obstacle location information and the current posture information of the robot into the hazard assessment model to obtain the corresponding operation hazard level and the distance to the nearest obstacle. Input the image data into the pre-trained target recognition model to obtain the target location information corresponding to the target area.

[0155] S3. Based on the job hazard level, the distance to the nearest obstacle, and the preset safe speed model, generate the corresponding obstacle avoidance path and job execution instructions, and calculate the position deviation based on the target position information and the current position coordinates;

[0156] S4. Input the operation hazard level, distance to the nearest obstacle, obstacle avoidance path, operation execution command, position deviation and target position information into the motion decision model to determine the control parameters of the robot arm;

[0157] S5. Collect contact force data of the end effector of the chemical emergency manipulator, input the contact force data into the pre-trained force control prediction model to obtain the predicted contact force of the end effector; adjust the movement trajectory and contact force of the end effector according to the manipulator control parameters, position deviation and predicted contact force;

[0158] S6. In response to the end effector moving to the target position preset range, compare the position deviation with the preset deviation threshold; in response to the position deviation being less than or equal to the deviation threshold, compare the contact force data with the preset force threshold; in response to the contact force data being less than or equal to the force threshold, combine the target position information and contact force data to generate a final grasping command, and control the end effector to grasp the emergency operation object in the target area according to the final grasping command.

[0159] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A smart sensing-based control system for an emergency mechanical hand in a chemical industry, characterized in that, include: The data acquisition unit is used to collect environmental data from the chemical scene and obtain the initial depth image of the target area of ​​the chemical equipment to be operated. Multi-sensor fusion processing is performed on environmental data and initial depth images to obtain corresponding gas concentration information, obstacle location information, and image data of the target area; The hazard assessment unit is used to input gas concentration information, obstacle location information and the current posture information of the robot into the hazard assessment model to obtain the corresponding operation hazard level and the distance to the nearest obstacle, and input image data into the pre-trained target recognition model to obtain the target location information corresponding to the target area; The coordinate calculation unit is used to generate corresponding obstacle avoidance paths and operation execution instructions based on the operation hazard level, the distance to the nearest obstacle, and the preset safe speed model, and to calculate the position deviation based on the target position information and the current position coordinates; The motion decision unit is used to input the operation hazard level, the distance to the nearest obstacle, the obstacle avoidance path, the operation execution command, the position deviation and the target position information into the motion decision model to determine the control parameters of the robot arm. The force control prediction unit is used to collect contact force data of the end effector of the chemical emergency manipulator, input the contact force data into the pre-trained force control prediction model to obtain the predicted contact force of the end effector, and adjust the movement trajectory and contact force of the end effector according to the manipulator control parameters, position deviation and predicted contact force. The end control unit is used to compare the position deviation with a preset deviation threshold in response to the end actuator moving to a preset range of the target position; In response to a position deviation being less than or equal to a deviation threshold, the contact force data is compared with a preset force threshold; In response to the contact force data being less than or equal to the force threshold, a final grasping command is generated by combining the target location information and the contact force data. Based on the final grasping command, the end effector is controlled to grasp the emergency operation object in the target area. This involves multi-sensor fusion processing of environmental data and initial depth images to obtain corresponding gas concentration information, obstacle location information, and image data of the target area, including: The initial environmental data of the chemical scene is collected by the environmental perception module to obtain the initial gas concentration and initial obstacle coordinates corresponding to the initial environmental data. The background is removed from the initial depth image to obtain the image data corresponding to the target area. Construct the corresponding concentration state vector and concentration state covariance matrix based on the initial gas concentration, determine the fusion gain matrix based on the concentration state vector and concentration state covariance matrix, and obtain the fused concentration vector. Based on the initial obstacle coordinates, construct the corresponding position state vector and position state covariance matrix. Based on the position state vector and position state covariance matrix, determine the fusion gain matrix to obtain the fused position vector. Data alignment is performed on the fused concentration vector and fused location vector to obtain gas concentration information and obstacle location information.

2. The chemical emergency manipulator control system based on intelligent sensing according to claim 1, characterized in that, Gas concentration information, obstacle location information, and the current posture information of the robotic arm are input into the hazard assessment model to obtain the corresponding operational hazard level and the distance to the nearest obstacle. Image data is input into a pre-trained target recognition model to obtain the target location information corresponding to the target area, including: Based on gas concentration information and obstacle location information, construct corresponding environmental feature vectors and environmental covariance matrices, and determine the hazard assessment gain matrix based on the environmental feature vectors and environmental covariance matrix; The state update vector and state update covariance matrix are obtained based on the hazard assessment gain matrix. The hazard level of the operation is determined based on the state update vector, and the distance to the nearest obstacle is determined based on the state update covariance matrix. Image data is input into a pre-trained target recognition model, and the contour features of the image data are extracted by the feature extraction layer in the target recognition model and input into the classification layer. The contour features are classified by a classification layer to obtain the category label of the target region. The target location information of the target region is obtained by querying the preset coordinate mapping table based on the category label.

3. The chemical emergency manipulator control system based on intelligent sensing according to claim 2, characterized in that, Based on the job hazard level, the distance to the nearest obstacle, and a preset safe speed model, a corresponding obstacle avoidance path and job execution instructions are generated. Position deviation is calculated based on the target location information and the current location coordinates, including: The current position coordinates of the end effector collected by the position sensor of the chemical emergency manipulator are obtained, and the position deviation is calculated based on the current position coordinates and the target position information. Determine whether the hazard level of the operation exceeds the preset hazard threshold. If the hazard level exceeds the preset hazard threshold, determine whether to initiate emergency obstacle avoidance or allow the operation to continue based on the distance to the nearest obstacle. If the nearest obstacle is greater than the preset safe distance, a command to continue operation is generated to control the robot to maintain its current posture. If the nearest obstacle is less than the preset safe distance, an obstacle avoidance path is generated based on the obstacle's position information to calculate the avoidance vector. The operation will continue or the obstacle avoidance path will be used as the operation execution instruction.

4. The chemical emergency manipulator control system based on intelligent sensing according to claim 3, characterized in that, The operation hazard level, distance to the nearest obstacle, obstacle avoidance path, operation execution command, position deviation, and target position information are input into the motion decision model to determine the robot's control parameters, including: Determine whether the hazard level of the operation exceeds the preset hazard threshold. If the hazard level of the operation exceeds the preset hazard threshold and the distance to the nearest obstacle is less than the preset safe distance, use the obstacle avoidance path as the target trajectory to control the robot arm to perform obstacle avoidance actions. In response to the fact that the hazard level of the operation does not exceed the preset hazard threshold, a movement correction amount is generated based on the position deviation and target position information, and a joint angle adjustment amount is generated in combination with the operation execution command; The control parameters of the robot are determined based on the movement correction amount and the joint angle adjustment amount. The control parameters of the robot include the motor speed of the movement mechanism and the output pressure of the force control mechanism.

5. The chemical emergency manipulator control system based on intelligent sensing according to claim 4, characterized in that, Contact force data of the end effector of a chemical emergency manipulator is collected and input into a pre-trained force control prediction model to obtain the predicted contact force corresponding to the end effector, including: Collect several contact force samples from the end effector at historical moments, filter these contact force samples, and obtain the contact force data; The contact force data is input into a pre-trained force control prediction model, and the temporal features of the contact force data are extracted through the force control prediction model. Based on the time-domain characteristics, the contact force at the next moment is predicted, and the predicted contact force is output. The predicted contact force is used to determine whether the robot arm has made contact with the emergency operation object.

6. The chemical emergency manipulator control system based on intelligent sensing according to claim 5, characterized in that, Adjusting the movement trajectory and contact force of the end effector based on the robot's control parameters, positional deviation, and predicted contact force, including: The movement mechanism is adjusted according to the motor speed in the robot's control parameters, driving the end effector to move towards the target position, and the movement direction is corrected in real time according to the position deviation. In response to the predicted contact force display showing a sharp increase in contact force, the downforce of the end effector is reduced in conjunction with the output pressure in the robot control parameters.

7. A chemical emergency manipulator control system based on intelligent sensing according to claim 6, characterized in that, In response to the end effector moving to a target position within a preset range, the position deviation is compared with a preset deviation threshold; In response to a position deviation being less than or equal to a deviation threshold, the contact force data is compared with a preset force threshold, including: Obtain the real-time position coordinates of the end effector and calculate the distance between the real-time position coordinates and the target position information as the position deviation; The absolute value of the position deviation is compared with the deviation threshold to obtain the comparison result; If the position deviation is greater than the deviation threshold, the position deviation is recalculated and the movement continues. If the position deviation is less than or equal to the deviation threshold, the force control fine adjustment stage is entered. In response to the position deviation entering the allowable range, the magnitude of the contact force data is monitored in real time; The contact force data is compared with a preset force threshold. If the contact force data is greater than the force threshold, it is determined that the contact is too tight. In response to the contact force data being greater than a preset force threshold, the difference between the contact force data and the force threshold is calculated, and a force reduction command is generated based on the difference to control the force control mechanism to reduce the output pressure.

8. The chemical emergency manipulator control system based on intelligent sensing according to claim 7, characterized in that, In response to contact force data being less than or equal to a force threshold, a final grasping command is generated by combining target location information and contact force data. Based on this command, the end effector is controlled to grasp the emergency operation object in the target area, including: After confirming that the position deviation is less than or equal to the deviation threshold and the contact force data is less than or equal to the force threshold, the final gripping instruction is generated. The final gripping instruction includes the gripping angle and gripping force. The final gripping command is sent to the gripping drive module of the end effector, which controls the gripping drive module to close the clamping mechanism; During the closing process, contact force data is used for closed-loop feedback, which, together with the grasping force, stably grasps the emergency operation object and completes the chemical emergency operation.

9. A control method for a chemical emergency manipulator based on intelligent sensing, applicable to the chemical emergency manipulator control system based on intelligent sensing as described in any one of claims 1-8, characterized in that, include: Collect environmental data from the chemical industry scenario to obtain an initial depth image of the target area of ​​the chemical equipment to be operated; Multi-sensor fusion processing is performed on environmental data and initial depth images to obtain corresponding gas concentration information, obstacle location information, and image data of the target area; Gas concentration information, obstacle location information, and current posture information of the robot are input into the hazard assessment model to obtain the corresponding operation hazard level and the distance to the nearest obstacle. Image data is input into the pre-trained target recognition model to obtain the target location information corresponding to the target area. Based on the job hazard level, the distance to the nearest obstacle, and the preset safe speed model, the corresponding obstacle avoidance path and job execution instructions are generated, and the position deviation is calculated based on the target position information and the current position coordinates. Input the operation hazard level, distance to the nearest obstacle, obstacle avoidance path, operation execution command, position deviation and target position information into the motion decision model to determine the control parameters of the robot arm; The contact force data of the end effector of the chemical emergency manipulator is collected and input into a pre-trained force control prediction model to obtain the predicted contact force of the end effector. The movement trajectory and contact force of the end effector are adjusted according to the manipulator control parameters, position deviation and predicted contact force. In response to the end effector moving to a target position within a preset range, the position deviation is compared with a preset deviation threshold; In response to a position deviation being less than or equal to a deviation threshold, the contact force data is compared with a preset force threshold; In response to contact force data being less than or equal to a force threshold, a final grasping command is generated by combining the target location information and contact force data. Based on the final grasping command, the end effector is controlled to grasp the emergency operation object in the target area.

Images