Control methods for hydrogen refueling robots

Abstract

This disclosure relates to a control method for a hydrogen refueling robot, including acquiring a 2D planar image and a 3D point cloud image of the vehicle's hydrogen refueling station; using deep learning-based 2D recognition technology to coarsely match the 2D planar image to obtain a coarse matching position of the hydrogen refueling station; cropping the 3D point cloud image based on the coarse matching position to obtain an original point cloud image of the hydrogen refueling station; denoising the original point cloud image to obtain a denoised point cloud image; precisely matching the denoised point cloud image based on a preset hydrogen refueling port template to obtain a hydrogen refueling port image of the hydrogen refueling station; further precisely matching the denoised point cloud hydrogen refueling port image based on a preset model template to obtain the hydrogen refueling port of the hydrogen refueling station and determining the position of the hydrogen refueling port; establishing a force-position hybrid control model for the hydrogen refueling robot; and controlling the hydrogen refueling robot to insert the hydrogen refueling nozzle into the hydrogen refueling port of the hydrogen refueling station based on the hydrogen refueling port position and the force-position hybrid control model. This technical solution can improve the working efficiency of the hydrogen refueling robot.

Description

Technical Field

[0001] This disclosure relates to the field of robot control technology, and more specifically, to a control method for a hydrogen refueling robot. Background Technology

[0002] Hydrogen-powered vehicles use hydrogen as their energy source, converting the chemical energy from the hydrogen reaction into mechanical energy to propel the vehicle. Hydrogen refueling stations, serving as the infrastructure for providing hydrogen to these vehicles, are becoming increasingly numerous. Refueling at these stations typically utilizes refueling robots. The hydrogen-powered vehicle is equipped with a refueling base with a refueling port. The refueling robot's refueling nozzle connects to this port, eliminating the need for manual operation, thus improving efficiency and saving labor costs. However, the location of the refueling port varies between different hydrogen-powered vehicles, and it also changes depending on the vehicle's parking position. When the port location changes, the refueling robot often needs to adjust the refueling nozzle multiple times to align it correctly, impacting its overall efficiency. Summary of the Invention

[0003] The purpose of this disclosure is to provide a control method for a hydrogen refueling robot to solve the problem in related technologies where the hydrogen refueling robot typically needs to adjust the position of the hydrogen refueling gun multiple times after the position of the hydrogen refueling port changes, resulting in low working efficiency of the hydrogen refueling robot.

[0004] To achieve the above objectives, this disclosure provides a control method for a hydrogen refueling robot, comprising the following steps:

[0005] Acquire 2D planar images and 3D point cloud images of the vehicle's hydrogen refueling station;

[0006] The coarse matching position of the hydrogenation base is obtained by coarsely matching the 2D planar image using deep learning-based 2D recognition technology.

[0007] Based on the coarse matching position, the 3D point cloud image is cropped to obtain the original point cloud image of the hydrogen refueling base;

[0008] The original point cloud image is denoised to obtain a denoised point cloud image;

[0009] The hydrogen filling port image of the hydrogen filling base is obtained by accurately matching the denoised point cloud image according to the preset hydrogen filling port template.

[0010] The hydrogen filling port of the hydrogen filling base is obtained by further precise matching of the hydrogen filling port image according to the preset model template, and the position of the hydrogen filling port is determined.

[0011] A force-position hybrid control model for the hydrogen refueling robot is established. Based on the hydrogen refueling port position and force-position hybrid control model, the hydrogen refueling robot is controlled to insert the hydrogen refueling gun into the hydrogen refueling port of the hydrogen refueling base.

[0012] Optionally, the 2D planar image can be acquired using the 2D mode of a 3D camera.

[0013] Optionally, the coarse matching specifically includes:

[0014] The 2D planar image is input into a convolutional neural network to extract features;

[0015] The extracted features are input into the second layer of the Transformer Encoder-Decoder to obtain the coarse matching position.

[0016] Optionally, the noise reduction process specifically includes:

[0017] Based on the coarse matching position, the original point cloud image is filtered and denoised in both the x and y directions to obtain the region of interest;

[0018] Perform boundary segmentation on the point cloud of the region of interest;

[0019] The segmented region of interest is further filtered and denoised to obtain the denoised point cloud image, which contains the point cloud data of the hydrogen refueling base.

[0020] Optionally, the filtering and noise reduction includes at least one of radius denoising and statistical denoising.

[0021] Optionally, the hydrogen filling port of the hydrogen filling base is obtained by performing small template precise matching on the denoised point cloud image according to the preset model template, and the position of the hydrogen filling port is determined.

[0022] Optionally, the specific process of establishing the force-position hybrid control model is as follows:

[0023] The rigid plane is taken as the xoy plane, and the direction perpendicular to the rigid plane and outward is taken as the z-axis direction;

[0024] When the end effector of the hydrogen refueling robot moves toward the rigid plane but has not yet contacted the rigid plane, the hydrogen refueling robot is in position control in the X-axis, Y-axis and Z-axis directions;

[0025] When the end effector of the hydrogen refueling robot contacts the rigid plane, the hydrogen refueling robot is under position control in the X and Y axes and under force control in the Z axis; it is still under position control in the X and Y axes, but not under force control.

[0026] Optionally, when not in contact with a rigid plane, the force-position hybrid control model satisfies natural constraints; when in contact with a rigid plane, the force-position hybrid control model satisfies both natural and artificial constraints.

[0027] Natural constraints are constraint relationships determined based on the geometry of the task, and are expressed as:

[0028]

[0029] Human-defined constraints are the desired movements set according to the requirements of the task, expressed as:

[0030]

[0031] Among them, f x f represents the force along the x-axis. y f represents the force along the y-axis. Z f represents the force along the z-axis. dz G represents the desired force along the z-axis. x The torque g represents the force along the x-axis. y The torque g represents the force along the y-axis. Z The torque along the z-axis, v x v represents the translational velocity along the x-axis. y v represents the translational velocity along the y-axis. Z w represents the translational velocity along the z-axis. x w represents the rotational speed along the x-axis. y w represents the rotational speed along the y-axis. z The rotational velocity along the z-axis, v dx v represents the desired velocity along the x-axis. dy v represents the desired velocity along the y-axis. dz This represents the desired velocity along the z-axis.

[0032] Optionally, the position control process is as follows: the desired displacement of the joints of the hydrogen refueling robot along the x-axis, y-axis and z-axis is used as the set value, the position sensor detection value of each joint of the hydrogen refueling robot is used as the displacement feedback value, and the difference between the desired displacement and the displacement feedback value is used as the input value of the hydrogen refueling robot controller for control.

[0033] The force control process is as follows: the desired torque of the joints of the hydrogen refueling robot along the x-axis, y-axis and z-axis is used as the set value; the torque sensor detection value of each joint of the hydrogen refueling robot is used as the torque feedback value; and the difference between the desired torque and the torque feedback value is used as the input value of the hydrogen refueling robot controller for control.

[0034] Optionally, during the force control process, the joint torque τ required by the end joint of the hydrogen refueling robot is:

[0035]

[0036] Where H is the joint space inertia matrix, q is the joint angle vector, C is the matrix for calculating Coriolis force and centripetal force, G is the end-effector load gravity, and τ is the joint angle vector. f F is the frictional force experienced by the end joint. load Environmental forces.

[0037] The above technical solution first acquires 2D planar images and 3D point cloud images of the hydrogen refueling base. Coarse matching is then performed to obtain the coarse matching position of the hydrogen refueling base. Next, based on the coarse matching position, the 3D point cloud image is cropped to obtain the original point cloud image of the hydrogen refueling base. After denoising, small template matching is performed using a model template to obtain the accurate position of the hydrogen refueling port. This ensures the accuracy of the hydrogen refueling robot's position control, allowing for better alignment of the hydrogen refueling nozzle with the refueling port and improving work efficiency. Simultaneously, a force-position hybrid control model for the hydrogen refueling robot is established. This model can correct error signals introduced by visual positioning of the refueling port. The hydrogen refueling robot can adjust its torque based on these error signals, making it adaptable and compliant during use. This compliant operation improves the adaptability and flexibility of the hydrogen refueling process, enabling the robot to adjust based on real-time force feedback for more precise and reliable hydrogen refueling operations.

[0038] Other features and advantages of this disclosure will be described in detail in the following detailed description section. Attached Figure Description

[0039] The accompanying drawings are provided to further illustrate the present disclosure and form part of the specification. They are used together with the following detailed description to explain the present disclosure, but do not constitute a limitation thereof. In the drawings:

[0040] Figure 1 This is a flowchart of a control method for a hydrogen refueling robot provided in an exemplary embodiment of this disclosure;

[0041] Figure 2 This is a flowchart illustrating the coarse matching process provided in an exemplary embodiment of this disclosure;

[0042] Figure 3 This is a schematic diagram of a force-position hybrid control model provided in an exemplary embodiment of this disclosure;

[0043] Figure 4 This is a schematic diagram of a torque mode implementation force-position hybrid control model provided in an exemplary embodiment of this disclosure;

[0044] Figure 5 This is a schematic diagram of a position mode implementation force-position hybrid control model provided by an exemplary embodiment of this disclosure. Detailed Implementation

[0045] The specific embodiments of this disclosure will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit this disclosure.

[0046] In this disclosure, unless otherwise stated, directional terms generally refer to the orientation of the relevant component in its actual use. "Inner" and "outer" can refer to the inner and outer contours of the corresponding component, or its location within or outside its environment, depending on the specific context. Furthermore, when the following description relates to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The terms "first," "second," etc., used in this disclosure are for distinguishing one element from another and do not indicate sequence or importance.

[0047] This invention is based on an explosion-proof robot, an explosion-proof 3D camera, and an explosion-proof force control sensor. These components can all be existing explosion-proof robots, 3D cameras, and force control sensors. The explosion-proof robot is a multi-joint robot equipped with an explosion-proof force control sensor and a position sensor, enabling it to sense and adjust the force and position during the hydrogen refueling process. In other words, the control method of this invention can be applied to existing hydrogen refueling robots.

[0048] like Figures 1 to 5 As shown, this disclosure provides a control method for a hydrogen refueling robot, including the following steps:

[0049] In step S1: 2D planar image and 3D point cloud image of the vehicle's hydrogen refueling station are acquired. Both images can be acquired simultaneously using a 3D camera to reduce image processing time and improve efficiency. The acquired 2D planar image is shown below. Figure 3 As shown; the 2D planar image and 3D point cloud image include the hydrogen refueling base, but the shooting range is relatively large, so the position of the hydrogen refueling base needs to be accurately located;

[0050] In step S2: a deep learning-based 2D recognition technology is used to perform coarse matching on the 2D planar image to obtain the coarse matching position of the hydrogen refueling base. The coarse matching is fast and can quickly determine the approximate position of the hydrogen refueling base in the 2D planar image.

[0051] In step S3: Based on the coarse matching position, the 3D point cloud image is cropped to obtain the original point cloud image of the hydrogenation base. The original point cloud image only includes the hydrogenation base and a small surrounding area, which can greatly reduce the amount of data to be processed and improve the efficiency of subsequent processing; however, the point cloud image at this time is unprocessed and the background is noisy.

[0052] In step S4: the original point cloud image is denoised to obtain a denoised point cloud image;

[0053] In step S5: The hydrogen filling port image of the hydrogen filling base is obtained by accurately matching the denoised point cloud image according to the preset hydrogen filling port template. Here, the hydrogen filling port template can be the outline of the hydrogen filling port. The preset hydrogen filling port template is matched with the denoised point cloud image. The part of the denoised point cloud image that can match the hydrogen filling port template is the hydrogen filling port. At this time, there is still a lot of noise. Denoising can be performed again or multiple times to reduce the impact of noise, thereby improving accuracy and subsequent accurate matching speed.

[0054] In step S6: The hydrogen filling port of the hydrogen filling base is obtained by further precise matching of the denoised point cloud image according to the preset model template, and the position of the hydrogen filling port is determined. At this time, the preset model template can be the outer contour of the hydrogen filling port, that is, the line template. The preset model template is matched with the point cloud image of the hydrogen filling port to finally determine the hydrogen filling port and its position. The accuracy of obtaining the position of the hydrogen filling port at this time is very high.

[0055] In step S7: Establish a force-position hybrid control model for the hydrogen refueling robot, and control the hydrogen refueling robot to insert the hydrogen refueling gun into the hydrogen refueling port of the hydrogen refueling base according to the hydrogen refueling port position and force-position hybrid control model.

[0056] The above technical solution first acquires 2D planar images and 3D point cloud images of the hydrogen refueling base. Coarse matching is then performed to obtain the coarse matching position of the hydrogen refueling base. Next, based on the coarse matching position, the 3D point cloud image is cropped to obtain the original point cloud image of the hydrogen refueling base. After denoising, small template matching is performed using a model template to obtain the accurate position of the hydrogen refueling port. This ensures the accuracy of the hydrogen refueling robot's position control, allowing for better alignment of the hydrogen refueling nozzle with the refueling port and improving work efficiency. Simultaneously, a force-position hybrid control model for the hydrogen refueling robot is established. This model can correct error signals introduced by visual positioning of the refueling port. The hydrogen refueling robot can adjust its torque based on these error signals, making it adaptable and compliant during use. This compliant operation improves the adaptability and flexibility of the hydrogen refueling process, enabling the robot to adjust based on real-time force feedback for more precise and reliable hydrogen refueling operations.

[0057] In some implementations, 2D planar images can be acquired using the 2D mode of a 3D camera. A 3D camera can not only capture two-dimensional images of a scene, but also obtain the positional relationships between objects. After further processing, a 3D point cloud image can be obtained. Simultaneously, a 3D camera has a wider field of view and can better observe and capture the surrounding environment near the hydrogen refueling port.

[0058] Coarse matching can be performed as follows: A planar image is input into a convolutional neural network (CNN) to extract features. These extracted features are then input into the second layer, the Transformer Encoder-Decoder, to obtain the coarse matching position. The Transformer Encoder helps the network learn global features more effectively. Then, the Transformer Decoder and ObjectQuery are used to learn the object to be detected from the features. The ObjectQuery result and the ground truth are then subjected to bipartite graph matching (Set-to-Set Loss). Finally, the classification loss and location regression loss are calculated based on the matching results. The Set-to-Set Loss process adds a bipartite graph matching step before calculating the network loss, ensuring that the final prediction result is only compared with the ground truth value for loss calculation, as shown in the following formula:

[0059]

[0060] The 3D point cloud images captured by the 3D camera are unprocessed and have noisy backgrounds. Therefore, denoising processing is usually required for the 3D point cloud images of the hydrogen refueling base. Specifically, the denoising process involves: filtering and reducing noise in both the x and y directions of the original point cloud image based on the coarse matching position to obtain the region of interest (ROI); then segmenting the point cloud within the ROI; and further filtering and reducing noise in the segmented ROI to obtain a denoised point cloud image containing only the point cloud data of the hydrogen refueling base. The filtering and noise reduction can include at least one of radius denoising and statistical denoising.

[0061] Precise matching can be specifically described as follows: Based on a preset model template, a small-template precise matching is performed on the denoised point cloud image to obtain the hydrogen refueling port of the hydrogen refueling station, and the location of the hydrogen refueling port is determined. The preset model template is created based on hydrogen refueling stations of various vehicles to meet the hydrogen refueling needs of various vehicles. The closest model template is matched based on the denoised point cloud image, and then the closest model template is selected.

[0062] Furthermore, the preset model templates, including the hydrogen refueling port template mentioned above, can be multiple, and the corresponding template can be selected according to the perspective of the 3D point cloud image.

[0063] The specific process of establishing the force-position hybrid control model can be as follows: Using a rigid plane as the xoy plane and the direction perpendicular to the rigid plane outwards as the z-axis; here, the rigid plane can be a contact surface on the vehicle, and the end effector of the hydrogen refueling robot can refer to the hydrogen refueling nozzle. When the end effector of the hydrogen refueling robot moves towards the rigid plane but has not yet contacted it, the hydrogen refueling robot performs position control in the X, Y, and Z axes; when the end effector of the hydrogen refueling robot contacts the rigid plane, the hydrogen refueling robot performs position control in the X and Y axes, and force control in the Z axis; in the X and Y axes, it remains position control, without force control. When the end effector of the hydrogen refueling robot moves towards the rigid plane but has not yet made contact with it, there is no contact with the rigid plane in any direction, and it is all position control, which can increase the moving speed of the hydrogen refueling gun. When the end effector of the hydrogen refueling robot makes contact with the rigid plane, the movement along the z-axis is hindered by the rigid plane, so position control in the z-axis direction is no longer possible, and only force control is possible. In the x-axis and y-axis directions, it is still position control, but there is no force control, to avoid damage to the hydrogen refueling gun when it comes into contact with the rigid plane.

[0064] When not in contact with a rigid plane, the force-position hybrid control model satisfies natural constraints; when in contact with a rigid plane, the force-position hybrid control model satisfies both natural and artificial constraints.

[0065] Natural constraints are constraints determined by the geometry of the task. After the robot's end effector contacts the rigid plane, its position along the z-axis is restricted by the plane's resistance. It is also assumed that, neglecting friction, the friction between the end effector and the rigid plane in the xoy plane is zero, and there are no opposing torques on the end effector's rotation about the xyz axes. Thus, natural constraints can be expressed as:

[0066]

[0067] Human-defined constraints are the desired movements set according to the requirements of the task, and can be expressed as:

[0068]

[0069] Among them, f x f represents the force along the x-axis. y f represents the force along the y-axis. Z f represents the force along the z-axis. dz G represents the desired force along the z-axis. x The torque g represents the force along the x-axis. y The torque g represents the force along the y-axis. Z The torque along the z-axis, v x v represents the translational velocity along the x-axis.y v represents the translational velocity along the y-axis. Z w represents the translational velocity along the z-axis. x w represents the rotational speed along the x-axis. y w represents the rotational speed along the y-axis. z The rotational velocity along the z-axis, v dx v represents the desired velocity along the x-axis. dy v represents the desired velocity along the y-axis. dz This represents the desired velocity along the z-axis.

[0070] The position control process is as follows: the desired displacement of the joints of the hydrogen refueling robot along the x-axis, y-axis and z-axis is used as the set value, the position sensor detection value of each joint of the hydrogen refueling robot is used as the displacement feedback value, and the difference between the desired displacement and the displacement feedback value is used as the input value of the hydrogen refueling robot controller for control.

[0071] The force control process is as follows: the desired torque of the joints of the hydrogen refueling robot along the x-axis, y-axis and z-axis is used as the set value, the torque sensor value of each joint of the hydrogen refueling robot is used as the torque feedback value, and the difference between the desired torque and the torque feedback value is used as the input value of the hydrogen refueling robot controller for control.

[0072] exist Figures 3 to 5 In this matrix, S is the selection matrix, a diagonal matrix used to control which dimensions are used for force control and which for position control. I is the identity matrix. PCL represents the position control law, FCL represents the force control law, and CT represents the coordinate transformation. a x is the location feedback value. d Enter the position value, x e f is the position difference. a f is the torque feedback value. d f is the torque input value. e This represents the torque difference.

[0073] Depending on the control method of the servo motors at the robot's joints, a control method can be designed. Where J represents the robot's Jacobian matrix. The estimated environmental stiffness is represented by PCLT, the position controller in the compliant coordinate system is represented by PCLJ, the position controller in the joint space is represented by FCL, and the force controller is represented by FCL.

[0074] In general, robot dynamics models are obtained without considering the robot's load. However, in actual tasks, the robot's load is constantly changing (such as the end effector gripper). These changes in end effector load will obviously cause changes in the robot's dynamic characteristics, so the traditional dynamics model needs to be modified.

[0075] In robot statics, neglecting friction between joints, and using the principle of virtual work, we can obtain the following formula relating the output force F (force and torque) of the robot's end effector joint to the required joint torque τ.

[0076] τ=J T F

[0077] Based on this characteristic, when the inertia of the end-effector load during motion is ignored, it can be approximated that the robot moves against the gravity of the end-effector load. That is, the gravity of the end-effector load is considered as the force Fload (including force and torque) exerted by the external environment on the robot's end effector. Thus, by decoupling this environmental force Fload to each joint using the above formula, the dynamic equations of the loaded robot can be constructed as shown in the following formula.

[0078]

[0079] Thus, if the end-effector load changes frequently, compensation can be made by only considering the torque required to overcome the gravity of the end-effector load, while ignoring the rotational characteristics of the load. If friction is to be considered, a friction term can be added to the formula, yielding the joint torque τ required by the end-effector of the hydrogen refueling robot during force control:

[0080]

[0081] Where H is the joint space inertia matrix, q is the joint angle vector, C is the matrix for calculating Coriolis force and centripetal force, G is the end-effector load gravity, and τ is the joint angle vector. f F is the frictional force experienced by the end joint. load Environmental forces.

[0082] Combining a loaded dynamic model with joint torque sensors can provide a foundation for subsequent force / position hybrid control.

[0083] While the hydrogen refueling nozzle doesn't have a six-dimensional force sensor, it does have torque sensors at each joint, providing real-time feedback on the joint torques between the links. Therefore, the relationship between the dynamic model and the joint torques can be used to estimate the external forces acting on the hydrogen refueling nozzle. This allows us to obtain the external forces acting on the robot's end effector through the torque sensors at each joint, providing a foundation for subsequent force-position hybrid control.

[0084] Impedance control equates a robot to a mass-damped-spring system, describing its dynamic properties using inertia, damping, and stiffness. Similar to adding a spring to the robot's end effector for passive compliant control, we allow the robot to exhibit greater flexibility (low stiffness) in directions requiring force control and greater rigidity (high stiffness) in directions requiring position control.

[0085] Assuming the hydrogen refueling robot's end effector is subjected to an external force Fe due to an external obstacle, and experiences a slight deflection, a corresponding restoring force F should be generated, pointing it to the desired position (the position of the end effector when no external force is applied). This can be mathematically described by the following formula:

[0086]

[0087] Where e = xd - x, is the tiny offset caused by the external force, i.e., the deviation between the desired pose and the actual pose of the robot's end effector. Md is the desired inertia of the robot, Bd is the desired damping of the robot, and Kd is the desired stiffness of the robot. These three characteristic matrices are all diagonal matrices, and the values ​​on the diagonal represent the characteristics in the corresponding dimension.

[0088] Impedance control only describes the dynamic relationship between external force and offset, but not the relationship with the desired force. For force control tracking in force / position hybrid control tasks, the formula needs to be modified to introduce the desired external force Fd, as shown in the figure. This allows force tracking and position tracking to be combined.

[0089]

[0090] Thus, in the force-position hybrid control model, impedance control can be used in the direction requiring force control, while position control remains the same in the direction requiring position control. The role of impedance control is to modify the trajectory in the force control direction, ensuring that the robot's movement meets the control requirements of compliant tasks.

[0091] When the robot's end effector is not in contact with the plane, position-mode control is used in all directions. When the end effector contacts the plane, it experiences a force Fe exerted by the plane on the robot. The position of the end effector at this point is taken as the reference position zr. Based on the calculation results of the formula, the small offset of the end effector in the z-direction is obtained as:

[0092]

[0093] Where zd represents the desired position in the z-direction, i.e., the modified desired position. That is, the desired position obtained by modifying the formula is:

[0094]

[0095] A torque sensor is placed between the end of the hydrogen refueling robot and the refueling nozzle to sense force information in real time during the refueling process. By feeding the force signal back to the hydrogen refueling robot controller, appropriate force commands can be calculated and applied to the robot to apply proper force to the refueling nozzle, thus achieving a smooth refueling process. When visual positioning is inaccurate, an intelligent floating algorithm can adjust the movement of the hydrogen refueling robot based on the force value from the torque sensor, ensuring that the refueling nozzle is accurately and concentrically inserted into the refueling port. The robot controller can adjust the torque based on error signals, making the robot adaptable and compliant, thereby correcting errors introduced by visual positioning. Flexible refueling improves the adaptability and flexibility of the refueling process, enabling the hydrogen refueling robot to adjust based on real-time force feedback for more precise and reliable refueling operations.

[0096] The preferred embodiments of this disclosure have been described in detail above with reference to the accompanying drawings. However, this disclosure is not limited to the specific details of the above embodiments. Within the scope of the technical concept of this disclosure, various simple modifications can be made to the technical solutions of this disclosure, and these simple modifications all fall within the protection scope of this disclosure.

[0097] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, this disclosure will not describe the various possible combinations separately.

[0098] Furthermore, various different embodiments of this disclosure can be combined in any way, as long as they do not violate the spirit of this disclosure, they should also be regarded as the content disclosed in this disclosure.

Claims

1. A control method for a hydrogen refueling robot, characterized in that, Includes the following steps: Acquire 2D planar images and 3D point cloud images of the vehicle's hydrogen refueling station; The coarse matching position of the hydrogenation base is obtained by coarsely matching the 2D planar image using deep learning-based 2D recognition technology. Based on the coarse matching position, the 3D point cloud image is cropped to obtain the original point cloud image of the hydrogen refueling base; The original point cloud image is denoised to obtain a denoised point cloud image; The hydrogen filling port image of the hydrogen filling base is obtained by accurately matching the denoised point cloud image according to the preset hydrogen filling port template. The hydrogen filling port of the hydrogen filling base is obtained by further precise matching of the denoised point cloud image according to the preset model template, and the position of the hydrogen filling port is determined. A force-position hybrid control model for the hydrogen refueling robot is established. Based on the hydrogen refueling port position and force-position hybrid control model, the hydrogen refueling robot is controlled to insert the hydrogen refueling nozzle into the hydrogen refueling port of the hydrogen refueling base. The noise reduction process specifically includes: Based on the coarse matching position, the original point cloud image is filtered and denoised in both the x and y directions to obtain the region of interest; Perform boundary segmentation on the point cloud of the region of interest; The segmented region of interest is further filtered and denoised to obtain the denoised point cloud image, which contains the point cloud data of the hydrogen refueling base.

2. The control method for a hydrogen refueling robot according to claim 1, characterized in that, The 2D planar image is acquired using the 2D mode of a 3D camera.

3. The control method for a hydrogen refueling robot according to claim 2, characterized in that, The coarse matching specifically refers to: The 2D planar image is input into a convolutional neural network to extract features; The extracted features are input into the second layer of the Transformer Encoder-Decoder to obtain the coarse matching position.

4. The control method for a hydrogen refueling robot according to claim 1, characterized in that, The filtering and noise reduction includes at least one of radius denoising and statistical denoising.

5. The control method for a hydrogen refueling robot according to claim 1, characterized in that, The denoised point cloud image is precisely matched with a small template according to a preset model template to obtain the hydrogen filling port of the hydrogen filling base and determine the position of the hydrogen filling port.

6. The control method for a hydrogen refueling robot according to claim 1, characterized in that, The specific process of establishing the force-position hybrid control model is as follows: take the rigid plane as the xoy plane, and take the direction perpendicular to the rigid plane and outward as the z-axis direction; When the end effector of the hydrogen refueling robot moves toward the rigid plane but has not yet contacted the rigid plane, the hydrogen refueling robot is in position control in the X-axis, Y-axis and Z-axis directions; When the end effector of the hydrogen refueling robot contacts the rigid plane, the hydrogen refueling robot is under position control in the X and Y axes and under force control in the Z axis; it is still under position control in the X and Y axes, but not under force control.

7. The control method for a hydrogen refueling robot according to claim 1, characterized in that, When not in contact with a rigid plane, the force-position hybrid control model satisfies natural constraints; when in contact with a rigid plane, the force-position hybrid control model satisfies both natural and artificial constraints. Natural constraints are constraint relationships determined based on the geometry of the task, and are expressed as: ； Human-defined constraints are the desired movements set according to the requirements of the task, expressed as: ； in, This represents the force along the x-axis. This represents the force along the y-axis. This represents the force along the z-axis. This represents the desired force along the z-axis. This represents the torque along the x-axis. This represents the torque along the y-axis. This represents the torque along the z-axis. This represents the translational velocity along the x-axis. This represents the translational velocity along the y-axis. This represents the translational velocity along the z-axis. This represents the rotational speed along the x-axis. This represents the rotational speed along the y-axis. This represents the rotational speed along the z-axis. This represents the desired velocity along the x-axis. This represents the desired velocity along the y-axis. This represents the desired velocity along the z-axis.

8. The control method for a hydrogen refueling robot according to claim 6, characterized in that, The position control process is as follows: the desired displacement of the joints of the hydrogen refueling robot along the x-axis, y-axis and z-axis is used as the set value, the position sensor detection value of each joint of the hydrogen refueling robot is used as the displacement feedback value, and the difference between the desired displacement and the displacement feedback value is used as the input value of the hydrogen refueling robot controller for control. The force control process is as follows: the desired torque of the joints of the hydrogen refueling robot along the x-axis, y-axis and z-axis is used as the set value; the torque sensor detection value of each joint of the hydrogen refueling robot is used as the torque feedback value; and the difference between the desired torque and the torque feedback value is used as the input value of the hydrogen refueling robot controller for control.

9. The control method for a hydrogen refueling robot according to claim 6, characterized in that, During the force control process, the joint torque τ required by the joint at the end of the hydrogenation robot is: ； Where H is the joint space inertia matrix, q is the joint angle vector, C is the Coriolis force and centripetal force calculation matrix, G is the end-effector load gravity, and τ is the joint angle vector. f F is the frictional force experienced by the end joint. load Environmental forces.

Images