Six-dimensional posture and force sensing intelligent joint and mechanical arm

Abstract

The invention provides a six-dimensional posture and force sensing intelligent joint and a mechanical arm. A contact pressure head is mounted on one of two connecting bodies through a connecting body connecting piece, and a first elastic connecting piece supporting part is arranged on the connecting body connecting piece; the illumination module, the image acquisition module and the touch sensing module are arranged on the other one of the two connectors; a second elastic connecting piece supporting part is arranged on the touch sensing module, and the two ends of the elastic connecting piece are connected with the first elastic connecting piece supporting part and the second elastic connecting piece supporting part respectively; the contact pressure head can press the gel flexible layer on the touch sensing module when the joint moves, and the gel flexible layer can deform when pressed by the contact pressure head. The structure is compact, the cost is low, and real-time and high-precision control and interaction of the robot on the joints in a complex, dynamic and unstructured environment can be achieved.

Description

Technical Field

[0001] This invention relates to the field of high-precision sensing technology for robots, specifically to a six-dimensional posture and force sensing intelligent joint and robotic arm, and more particularly to a high-precision six-dimensional posture and multi-dimensional force sensing intelligent joint based on camera tactile sensing. Background Technology

[0002] With the rapid development of collaborative robots, humanoid robots, and dexterous hands, higher demands are being placed on robots' ability to perceive their own state (including joint posture and external forces). Achieving high-precision six-dimensional perception, that is, simultaneously sensing the six degrees of freedom (3 degrees of freedom of pose + 3 degrees of freedom of force / torque) of a robotic arm joint, has become an important research direction in this field. This is directly related to the robotic arm's ability to achieve autonomous exploration, precise operation, and adaptive control. Currently, while mainstream six-dimensional force sensors have high accuracy, their structures are complex and expensive, and they are difficult to embed in applications such as biomimetic snake robots and humanoid wrists. Traditional sensors typically suffer from problems such as high cost, bulkiness, and inability to achieve multimodal perception. To perceive the pose and force of a robotic arm's wrist / joints, it is necessary to develop a low-cost, lightweight, and high-precision multimodal sensor.

[0003] In recent years, force and attitude perception algorithms based on artificial intelligence methods such as neural networks have been introduced. These algorithms infer position and force state from multiple dimensions of information, including visual and deformation data, through multimodal fusion perception models. These algorithms typically require a large amount of highly labeled force and attitude data as training samples to achieve robust perception models. However, traditional data acquisition methods rely on manual demonstrations of robotic arms to acquire corresponding operational state and force information data through high-precision sensors. This data acquisition process is complex, time-consuming, costly, and its efficiency and accuracy are difficult to guarantee.

[0004] Therefore, how to design a high-precision force and attitude sensor that is simple in structure, low in cost, highly adaptable to the environment, and can be well compatible with neural perception algorithms to achieve fusion perception has become a core problem that urgently needs to be solved in the field of robot perception.

[0005] Currently, the sensing of robotic arm joints mainly relies on traditional sensing elements such as force / torque sensors, encoders, IMUs (inertial measurement units), and strain gauges to measure the force and posture of the joints. These technologies have been widely used in robot control and manipulation tasks, but in some application scenarios, especially in mechanical structures such as humanoid wrists and biomimetic snake robots, there are still many shortcomings.

[0006] While current sensing technologies for robotic arm joints / wrists have made some progress in force and posture sensing, they still have many shortcomings in high-precision six-dimensional sensing, multimodal fusion, system integration, and environmental adaptability. These shortcomings are mainly reflected in the following aspects: 1) Limitations of force sensing technology: a) High cost and bulky structure: Although commercially available wrist force / torque sensors have high accuracy, they are generally expensive and bulky, making them difficult to apply to small, lightweight mechanical structures; b) Dependence on external devices and integration difficulties: Magnetic-driven flexible catheter robots attempt to integrate FBG (fiber Bragg grating) force sensors at the end of the arm to achieve force feedback. While this improves safety and flexibility in medical procedures, the structure remains complex and costly, and it is difficult to directly apply to the force sensing needs of multiple joints in continuous robotic arms; furthermore, this type of method is difficult to achieve direct sensing of forces within joints; 2) Limitations of posture perception: a) Separation of joint angle perception and overall shape reconstruction: Most current mainstream joint posture perception methods use embedded Hall angle encoders to construct the overall state by measuring the angle changes of each joint. Although simple and effective, it is difficult to fully reflect the nonlinear deformation and overall posture of the robotic arm; b) Weak ability to perceive contact force and environmental interaction: Although fiber optic sensing technology can be used for overall shape reconstruction, it cannot simultaneously perceive the internal forces of the joints and the contact state with the environment, resulting in a lack of real-time feedback information when the system performs autonomous exploration or fine operation tasks. 3) Challenges of tactile sensors: a) Vulnerability and installation limitations: Camera-based tactile sensors (such as GelSight and iFEM) use gel deformation for force recognition, resulting in high information density and sensitivity. However, they rely on delicate circuit structures and imaging systems, making the devices structurally fragile and difficult to install in frequently moving joints or bending areas; b) Modeling difficulties and high data dependence: There is a lack of effective analytical models between gel deformation and force. Force estimation usually relies on a large number of manually collected training sets, which limits the system's generalization ability and real-time performance in practical applications; c) Poor structural compatibility: Since the flexible gel layer needs to maintain a certain thickness to achieve effective deformation observation, it is difficult to withstand the continuous pressure during joint movement, making it difficult to directly apply to the joints of robotic arms, especially in robotic arms with high dynamics or continuous deformation.

[0007] In summary, existing sensing technologies for robotic arm joints generally suffer from the following problems: 1) Force and posture sensing capabilities are separated or missing, making it impossible to achieve multimodal fusion; 2) Sensors are large in size and difficult to integrate, making them unsuitable for high-degree-of-freedom and highly flexible robotic arms; 3) They have poor adaptability to dynamic environments and complex deformations; 4) They are costly and easily damaged, making it difficult to achieve large-scale deployment and long-term stable application.

[0008] Patent document CN117437383A discloses a visual-tactile endoscopic sensing device and its control method. This visual-tactile endoscopic sensing device includes: a miniature support frame; a gel module with multiple distributed marker points; the gel module is used to obtain displacement information of the target marker points upon receiving external pressure; an image acquisition device, the non-acquisition end of which is embedded in a cavity and tightly fitted to the inner wall of the cavity, and the acquisition end of the image acquisition device is aligned with the center of the inner wall of the gel module; the image acquisition module is used to extract tactile information from the displacement information corresponding to the marker point images, and the tactile information is used to represent the pressure sensing capability of the gel module. However, this patent document has the drawback of weak load-bearing capacity, making it unsuitable for direct assembly into the joints of a robotic arm.

[0009] Therefore, there is an urgent need for a novel sensing method that requires no additional devices, has a compact structure, strong compatibility, and can achieve force and posture fusion sensing to improve the autonomy and robustness of robotic arms in complex tasks. This is precisely the problem that this invention aims to solve. Summary of the Invention

[0010] To address the shortcomings of existing technologies, the purpose of this invention is to provide a six-dimensional posture and force sensing intelligent joint and robotic arm.

[0011] According to the present invention, a six-dimensional posture and force sensing intelligent joint is provided for connecting two connectors, including: an elastic connector, a contact pressure head, an illumination module, an image acquisition module, and a tactile sensing module; The contact pressure head is mounted on one of the two connectors via a connector, and the connector is provided with a first elastic connector support portion. The lighting module, the image acquisition module, and the tactile sensing module are disposed on the other of the two connectors; The tactile sensing module is provided with a second elastic connector support portion, and the two ends of the elastic connector are respectively connected to the first elastic connector support portion and the second elastic connector support portion. The contact pressure head can apply pressure to the flexible gel layer on the tactile sensing module during joint movement, and the flexible gel layer can deform when it is pressed by the contact pressure head. The image acquisition module can acquire images of the deformation of the deformation area on the flexible gel layer; the illumination module can illuminate the deformation area on the flexible gel layer to enhance the visibility of the indentation structure in the image.

[0012] Preferably, the tactile sensing module includes: a gel support structure, an acrylic support structure, an elastic connector support portion, a gel flexible layer, a gel support plate, and a tactile sensing module shell. The tactile sensing module housing is disposed on the connector; the gel support structure and the acrylic support structure are disposed on the tactile sensing module housing; The flexible gel layer is connected to the gel support plate; the flexible gel layer is located within the gel support structure, and the gel support plate is located within the acrylic support structure. The degree of deformation of the gel flexible layer under pressure can reflect the contact position and contact force.

[0013] Preferably, the lighting module includes: an LED light strip and a lighting module housing; The lighting module housing is disposed on the connector and located below the tactile sensing module housing of the tactile sensing module; The LED light strip is disposed on the housing of the lighting module; the LED light strip is used for combined lighting to create contrast between different indentation features in the image, thereby improving the accuracy and robustness of image recognition.

[0014] Preferably, the image acquisition module includes: an industrial camera, a camera acquisition card, a camera support frame, and an image acquisition module housing; The image acquisition module housing is mounted on the connector and located below the lighting module housing; the camera support bracket is mounted on the image acquisition module housing. The camera acquisition card is mounted on the camera support frame, and the industrial camera is mounted on the camera acquisition card.

[0015] Preferably, the contact pressure head includes: a contact pressure head body, a contact pressure head plane, and a silicone pressure head array; The contact pressure head body is mounted on the connector, the contact pressure head plane is disposed on the contact pressure head body, and the silicone pressure head array is disposed on the contact pressure head plane; The silicone compression head array includes multiple silicone compression heads arranged in an array.

[0016] Preferably, the connecting body connector includes: a connecting base, an elastic connecting support portion, and a bottom mating groove for the contact pressure head; The connecting base is disposed on the connecting body, and the elastic connecting member support part and the bottom mating groove of the contact pressure head are disposed on the connecting base; The contact head body is provided with a contact head bottom mating pin, which can be connected with the contact head bottom mating groove.

[0017] Preferably, the vertical distance from the top of the silicone pressure head array to the contact pressure head plane is defined as the contact pressure head height. The diagonal length of the silicone compression head array is defined as... The contact pressure head plane is a circular surface, and the diameter of the contact pressure head plane is defined as... Through the height of the contact head Distance from the contact head Together, they determine the maximum measurable bending angle of the contact indenter. ; For the gel flexible layer with a thickness of N mm, the maximum pressing depth is Initial pressing depth The maximum measurable angle of the contact indenter is constrained by both the pressable depth of the gel flexible layer (94) and the diameter of the contact indenter plane. The maximum measurable angle obtained from these two constraints is denoted as follows: and ; and The minimum value between the two is the final maximum measurable bending angle. Maximum measurable bending angle Represented as:

[0018] The industrial camera has a resolution of M mm / pixel; defined in the pressed area. The maximum tangent slope within is ; Minimum angular resolution of the contact indenter The maximum change in cross-sectional diameter of the contact indenter as it penetrates the flexible gel layer. The height of the contact head descent when the resolution is equal to that of an industrial camera. The ratio between the contact head spacing D and half of the contact head spacing; , , D The relation is expressed as:

[0019] .

[0020] Preferably, the elastic connector is an elastic compression spring or a flexible soft rubber support; And / or, the image acquisition module includes: a prediction model and an attention mechanism model; The prediction model employs a convolutional neural network structure to predict key pose parameters from tactile images; the convolutional neural network structure includes 3 convolutional layers, 2 max pooling layers, and 1 fully connected layer. The attention mechanism model adopts a network structure that adds channel attention layer and spatial attention layer to the ResNet-18 structure for the prediction of three-dimensional force and torque; The image acquisition module extracts the pressing contour information and pressing depth information of the colored pressing indentation information in the image through the convolutional layer, and uses multiple fully connected layers to decode the pressing contour information and pressing depth information to obtain the 2D planar rotation information of the contact indenter. Based on the pressing depth information and pressing contact situation, the joint pose and force direction are calculated.

[0021] Preferably, the neural network model mounted on the image acquisition module can predict the force on the joint by analyzing the image data of the indentation formed by the contact indenter pressing the flexible gel layer; When the input indentation image data is the internal force indentation data transmitted from the deformation of the elastic connector to the gel flexible layer during joint movement, the neural network model outputs the joint internal force parameters; When the input indentation image data is external contact indentation data generated by external load acting on the joint end, the neural network model outputs external contact force parameters; And / or, by replacing elastic connectors with different stiffness specifications, the force measurement range of the joint can be adjusted; The smaller the stiffness of the elastic connector, the greater the deformation of the elastic connector under a unit force, which in turn makes the force resolution of the joint higher. The greater the stiffness of the elastic connector, the smaller the deformation of the elastic connector under a unit force, and the larger the corresponding force measurement range.

[0022] The present invention also provides a robotic arm, including the aforementioned six-dimensional posture and force sensing intelligent joint.

[0023] Preferably, the robotic arm includes: multiple six-dimensional posture and force sensing intelligent joints and multiple connecting bodies; the connecting body is an arm body; the multiple arm bodies are connected in series, and two adjacent arm bodies are connected through one of the six-dimensional posture and force sensing intelligent joints; Alternatively, the robotic arm includes: a six-dimensional posture and force sensing intelligent joint and two connectors; one connector is the arm body, and the other connector is the hand body; the arm body and the hand body are connected through the six-dimensional posture and force sensing intelligent joint.

[0024] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention utilizes a camera-assisted tactile sensing structure within an intelligent joint. An industrial camera is installed inside the intelligent joint, along with a specific light-transmitting protective cover and light source components, to achieve multi-directional acquisition of tactile information from the joint's contact surface. This solves the problem of incomplete tactile sensing caused by the limited field of view of a single camera, providing rich raw data for subsequent 6D posture and force perception.

[0025] 2. This invention adopts a multi-intelligent joint collaborative perception and control method, which transmits the camera tactile sensing data, 6-dimensional posture and force sensing data of each intelligent joint to the central controller of the robotic arm. The controller analyzes the overall motion state of the robotic arm based on the perception information of multiple joints, formulates a collaborative control strategy, coordinates the motion and force output of each joint, and realizes stable and efficient operation of the robotic arm in scenarios such as precision assembly and material handling.

[0026] 3. This invention realizes an intelligent joint adaptive adjustment system based on 6-dimensional posture feedback. It protects the 6-dimensional posture data calculated by camera tactile sensing, compares the target posture in real time, generates adjustment signals for the joint drive motor, controls the motor to adjust the joint angle and position, realizes adaptive posture correction, and combines force sensing data to avoid joint damage due to excessive force during adjustment. Attached Figure Description

[0027] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 A schematic diagram of a three-dimensional structure connecting two arms with a six-dimensional posture and force sensing intelligent joint; Figure 2 for Figure 1 Schematic diagram of the cross-sectional structure along line AA; Figure 3 A cross-sectional schematic diagram of the six-dimensional posture and force sensing intelligent joint connecting two arms; Figure 4 for Figure 3 Schematic diagram of the cross-sectional structure along line BB; Figure 5 A schematic diagram of the three-dimensional structure of the tactile sensing module; Figure 6 This is a schematic diagram of the exploded cross-sectional structure of the tactile sensing module; Figure 7 This is a three-dimensional structural diagram of the lighting module; Figure 8 This is a bottom view of the lighting module; Figure 9 This is a schematic diagram of the three-dimensional structure of the image acquisition module; Figure 10 This is a top view of the image acquisition module; Figure 11 This is a schematic diagram of the contact pressure head. Figure 12 Schematic diagram of an explosion involving a contact pressure head Figure 1 ; Figure 13 Schematic diagram of an explosion involving a contact pressure head Figure 2 ; Figure 14 A schematic diagram of a structure connecting multiple arm bodies to multiple six-dimensional posture and force sensing intelligent joints; Figure 15 This is a schematic diagram of a flexible connector and a flexible soft rubber support. Figure 1 ; Figure 16 This is a schematic diagram of a flexible connector and a flexible soft rubber support. Figure 2 ; Figure 17 A schematic diagram of a three-dimensional structure connecting an arm and a hand to a six-dimensional posture and force sensing intelligent joint; Figure 18 A schematic diagram illustrating the workflow of software and hardware working together; Figure 19 This is a schematic diagram illustrating the principle of image preprocessing strategies. Figure 20 This is a schematic diagram of the structure when the flexible gel layer is spherical. Figure 21 This is a schematic diagram of the structure when the flexible gel layer is irregularly shaped. Figure 22 This is a schematic diagram of the structure when the contact pressure head is irregularly shaped; Figure 23 A schematic diagram illustrating the evolution from individual smart joints to smart joint and arm-body assemblies; Figure 24 A schematic diagram of a smart joint installed in a dexterous finger to measure the internal force of the finger joint.

[0028] The diagram shows: Detailed Implementation

[0029] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0030] Example 1 like Figures 1 to 4As shown, this embodiment provides a six-dimensional posture and force sensing intelligent joint for connecting two connectors, including: an elastic connector, a contact pressure head 4, an illumination module 7, an image acquisition module 8, and a tactile sensing module 9; the contact pressure head 4 is mounted on one of the two connectors via a connector 6, and the connector 6 is provided with a first elastic connector support portion 62; the illumination module 7, the image acquisition module 8, and the tactile sensing module 9 are disposed on the other connector; the tactile sensing module 9 is provided with a second elastic connector support portion 93, and the two ends of the elastic connector are respectively It is connected to the first elastic connector support part 62 and the second elastic connector support part 93; the contact pressure head 4 can press the gel flexible layer 94 on the tactile sensing module 9 when the joint moves. When the gel flexible layer 94 is pressed by the contact pressure head 4, it can deform. Then, the built-in vision sensor can capture this deformation and accurately calculate the multi-axis force and angle information of the joint; the image acquisition module 8 can acquire images of the deformation area on the gel flexible layer 94; the illumination module 7 can illuminate the deformation area on the gel flexible layer 94 to enhance the visibility of the indentation structure in the image.

[0031] like Figure 5 and Figure 6 As shown, the tactile sensing module 9 includes: a gel support structure 91, an acrylic support structure 92, an elastic connector support portion 93, a gel flexible layer 94, a gel support plate 95, and a tactile sensing module housing 96; the tactile sensing module housing 96 is disposed on the connector; the gel support structure 91 and the acrylic support structure 92 are disposed on the tactile sensing module housing 96; the gel flexible layer 94 is connected to the gel support plate 95; the gel flexible layer 94 is located inside the gel support structure 91, and the gel support plate 95 is located inside the acrylic support structure 92; the degree of deformation of the gel flexible layer 94 under pressure can reflect the contact position and contact force.

[0032] like Figure 7 and Figure 8 As shown, the lighting module 7 includes an LED light strip 71 and a lighting module housing 72. The lighting module housing 72 is disposed on the connector and located below the tactile sensing module housing 96 of the tactile sensing module 9. The LED light strip 71 is disposed on the lighting module housing 72. The LED light strip 71 is used for combined lighting to create contrast between different indentation features in the image, thereby improving the accuracy and robustness of image recognition. The LED light strip 71 includes a multi-color light source. The LED light strip 71 is an annular light strip, and the gel flexible layer 94 has a frustum structure. The LED light strip 71 and the gel flexible layer 94 are arranged concentrically.

[0033] Illumination with light sources of different wavelengths allows for a better understanding of the subtle characteristics of different indentations. When the indenter presses against the gel surface, the pressed area transforms from a flat plane into a downward-convex curved surface. This curved surface reflects light from different light sources in the LED strip, resulting in different colors on the pressed surface in different directions. The deeper the press, the greater the color difference and the greater the light intensity. Compared to a single-color light source, this multi-color light source combination lighting design solves the problem of single-color light easily producing reflections or shadows on certain materials or textures, leading to a loss of detail. Furthermore, color information can serve as an additional perceptual dimension, enhancing the ability to judge the contact state.

[0034] like Figure 9 and Figure 10 As shown, the image acquisition module 8 includes: an industrial camera, a camera acquisition card 81, a camera support frame 83, and an image acquisition module housing 84; the image acquisition module housing 84 is disposed on the connector and located below the lighting module housing 72 of the lighting module 7; the camera support frame 83 is disposed on the image acquisition module housing 84; the camera acquisition card 81 is disposed on the camera support frame 83, and the industrial camera is disposed on the camera acquisition card 81.

[0035] The tactile sensing module housing 96, the lighting module housing 72, and the image acquisition module housing 84 constitute the external support module 3.

[0036] like Figures 11 to 13 As shown, the contact pressure head 4 includes: a contact pressure head body, a contact pressure head plane 41, and a silicone pressure head array 42; the contact pressure head body is mounted on the connecting body connector 6, the contact pressure head plane 41 is disposed on the contact pressure head body, and the silicone pressure head array 42 is disposed on the contact pressure head plane 41; the silicone pressure head array 42 includes a plurality of silicone pressure heads arranged in an array. The connecting body connector 6 includes: a connecting base, an elastic connecting member support portion 62, and a contact pressure head bottom mating groove 63; the connecting base is disposed on the connecting body, and the elastic connecting member support portion 62 and the contact pressure head bottom mating groove 63 are disposed on the connecting base; the contact pressure head body is provided with a contact pressure head bottom mating pin 43, which can be engaged with the contact pressure head bottom mating groove 63 for connection.

[0037] On one hand, this design improves the ability to distinguish minute contacts and interactions. With two silicone pressure heads facing each other, compared to using only a single large contact head, the pressing angle and position can be characterized by the difference in the diameter of the relative contact heads when pressure is applied, amplifying the deformation effect caused by pressure. On the other hand, this design can accurately identify the location, pressure magnitude, and distribution pattern of contact points. Each silicone pressure head responds independently to local pressure, and the array structure divides the contact area into multiple "sensing units." The deformation of each pressure head can be measured using vision (such as a camera) or optical sensors (such as optical fibers or capacitors). Higher array density results in higher spatial resolution, allowing for the reconstruction of more detailed pressure maps.

[0038] The vertical distance from the top of the silicone pressure head array 43 to the contact pressure head plane 41 is defined as the contact pressure head height. The diagonal length of the silicone compression head array 43 is defined as... The contact pressure head plane 41 is a circular surface, and its diameter is defined as follows: ; by contact head height Distance between contact head Together they determine the maximum measurable bending angle of the contact indenter 4. For a gel flexible layer 94 with a thickness of N mm, the maximum pressing depth is Initial pressing depth The maximum measurable bending angle is expressed as:

[0039] The maximum measurable angle is constrained by both the compressible depth of the gel flexible layer (94) and the diameter of the base. The maximum measurable angle obtained from these two constraints is denoted as follows: and ; and The minimum value between the two is the final maximum measurable angle; Factors affecting angular resolution include the minimum pixel size of the industrial camera in image acquisition module 8. Contact head spacing Maximum slope of the pressing area The industrial camera has a resolution of M mm / pixel, while the area change rate of the contact pressure head is higher than the industrial camera's resolution M; defined in the pressed area. The maximum tangent slope within is The minimum rate of change of the profile width When the angle changes hour, Solve Greater than Mmm / pixel.

[0040] When a pressing action occurs along the direction of one of the contact indenters in the contact indenter array, the contact indenter presses downwards against the gel flexible layer, increasing the maximum cross-sectional diameter of the portion of the contact indenter embedded in the gel flexible layer. The minimum angular resolution of the contact indenter refers to the ratio between the height the contact indenter descends and half the contact indenter spacing D, when the change in the maximum cross-sectional diameter of the portion of the contact indenter embedded in the gel flexible layer is equal to the resolution of an industrial camera. The maximum ratio of the descending height of the contact indenter to the maximum change in the maximum cross-sectional diameter of the portion of the contact indenter embedded in the gel flexible layer is the maximum slope kma of the pressing area. The maximum slope kma determines the minimum value of the change in the maximum cross-sectional diameter of the portion of the contact indenter embedded in the gel flexible layer when the contact indenter descends by a height Δh. When this value equals the resolution of the industrial camera, the minimum descent height Δh of the contact indenter can be calculated. Then, using the formula... Solve for the minimum angular resolution Δθ.

[0041] By preset the maximum measurable bending angle The target value, combined with the formula

[0042] make = This allows us to calculate the contact head spacing. Then, by setting the target value of the minimum angular resolution Δθ, combined with the contact head spacing... and formula It can calculate Then combine with the formula It is possible to calculate the maximum tangent slope as Based on the contact head spacing The maximum tangent slope is It is possible to design contact pressure heads with corresponding structures.

[0043] The image acquisition module 8 includes a prediction model and an attention mechanism model. The prediction model uses a convolutional neural network structure to predict key pose parameters from tactile images. The convolutional neural network structure includes three convolutional layers, two max-pooling layers, and one fully connected layer. The attention mechanism model uses a network structure with channel attention layers and spatial attention layers added to the ResNet-18 structure for the prediction of three-dimensional forces and moments. The image acquisition module 8 extracts the pressing contour information and pressing depth information from the color pressing indentation information in the image through convolutional layers, and uses multiple fully connected layers to decode the pressing contour information and pressing depth information to obtain the two-dimensional planar rotation information of the contact head 4. Based on the pressing depth information and pressing contact situation, the joint pose and force direction are calculated.

[0044] The image acquisition module 8 is equipped with a neural network model that includes a prediction model and an attention mechanism model. By analyzing the indentation image data formed by the contact indenter 4 pressing the flexible gel layer 94, it achieves accurate prediction of the joint force. Specifically, the image acquisition module first captures the indentation image on the flexible gel layer 94 using an industrial camera, and then transmits it to the neural network model via a camera acquisition card. The model extracts the core features from the indentation image through convolutional layers (including the shape of the indentation contour, edge clarity, gray-level gradient distribution, and other pressing contour information, as well as the depth and area of ​​the indentation, and other pressing depth information). Then, it decodes the features through fully connected layers, and combines the attention mechanism model to focus on and enhance the key indentation areas, finally outputting accurate force parameters.

[0045] The predicted type of force parameters strictly corresponds to the type of input indentation image data: When the input image data is internal force indentation data transmitted from the deformation of the elastic connector to the gel flexible layer during joint movement, the model outputs joint internal force parameters. This type of internal force indentation is caused by the joint's own movement, which triggers the elastic connector's expansion / bending deformation, causing the contact indenter to compress the gel flexible layer. Its indentation characteristics (such as distribution uniformity and deformation amplitude) are directly related to the degree of deformation of the elastic connector, thus reflecting the force state inside the joint. When the input image data is external contact indentation data generated by external loads acting on the joint end, the model outputs external contact force parameters. This type of external contact indentation is driven by the load generated by the joint contacting the external environment / object. The contact indenter is subjected to external loads and applies pressure to the gel flexible layer. The position, shape, and depth of the indentation directly correspond to the point of application and magnitude of the external contact force.

[0046] By replacing elastic connectors with different stiffness specifications (such as elastic compression spring 2 or flexible soft rubber support 11), the force measurement range of the joint can be flexibly adjusted to meet the force sensing needs of different working conditions.

[0047] The stiffness characteristics of the elastic connector are clearly related to the force measurement performance of the joint: the smaller the stiffness of the elastic connector, the greater the deformation of the elastic connector under a unit force (whether internal or external contact force). This deformation is transmitted through the support parts of the first and second elastic connectors, causing the contact indenter to exert more significant pressure on the gel flexible layer, making the indentation features (such as indentation depth and contour recognition) formed on the gel flexible layer clearer. The neural network model has higher analytical accuracy for the indentation features, thus significantly improving the force resolution of the joint and enabling precise perception of minute forces. When the stiffness of the elastic connector is greater, the deformation of the elastic connector under a unit force is smaller, its resistance to deformation is stronger, and it can withstand greater ultimate loads. Correspondingly, the pressure of the contact indenter on the gel flexible layer will not cause damage to the gel flexible layer due to excessive deformation, and the joint can stably perceive a wider range of forces, resulting in a larger force measurement range.

[0048] In this embodiment, as Figure 1 , Figure 3 , Figure 14 As shown, the elastic connector is an elastic compression spring 2. This embodiment also provides a robotic arm, which includes: multiple six-dimensional posture and force-sensing intelligent joints and multiple connecting bodies; the connecting body is an arm body 1; multiple arm bodies 1 are connected in series, and adjacent two arm bodies 1 are connected through a six-dimensional posture and force-sensing intelligent joint. The multiple arm bodies 1 are connected by steel wire ropes 5, and the movement of the contact pressure head is controlled by the steel wire ropes 5.

[0049] In other embodiments, such as Figure 15 , Figure 16 , Figure 17 As shown, the elastic connector is a flexible soft rubber support 11. A robotic arm includes: a six-dimensional posture and force sensing intelligent joint and two connectors; one connector is the arm body 1, and the other connector is the hand body 12; the arm body 1 and the hand body 12 are connected via the six-dimensional posture and force sensing intelligent joint. The arm body 1 and the hand body 12 are connected by an airbag 13, which is a sealed air chamber, and the movement of the contact pressure head is controlled by air pressure.

[0050] Example 2 Those skilled in the art can understand this embodiment as a more specific description of Embodiment 1.

[0051] This embodiment provides a high-precision 6D posture and force sensing intelligent joint based on camera tactile sensing, which mainly includes two parts: hardware system design and data-driven modeling. The hardware system includes an image acquisition module, an illumination module, a tactile sensing module, and an external support module.

[0052] I. Hardware System Design.

[0053] The hardware system design of the intelligent joint aims to achieve high-precision pose and force sensing functions. Its structure is modular and highly integrated, comprising four core components: an image acquisition module 8, an illumination module 7, a tactile sensing module 9, and an external support module 3. These modules are highly integrated and work collaboratively to achieve high-precision, high-frequency real-time sensing.

[0054] The image acquisition module mainly consists of an industrial camera, used to acquire images of the surface deformation of the intermediate gel material after pressure. The industrial camera slides into the groove of the camera support frame 83, and the two are fitted with a clearance to ensure a stable fit. The camera's USB data cable is led out from the bottom of the camera support frame 83 and passes through the camera's USB cable hole 82. By acquiring surface deformation information from the images, and combining the data-driven prediction model LeNet-6 and the improved attention mechanism ResNet-18 structure (Modified Attention ResNet-18, MA-ResNet-18), the color-coded pressure indentation information in the images is extracted through convolutional layers to obtain its pressure contour information and pressure depth information. Then, multiple fully connected layers are used to decode the above information to obtain the 2D planar rotation information of the contact indenter, pressure depth information, and pressure contact situation, which can accurately calculate the joint pose and force direction.

[0055] The lighting module 7 consists of three sets of LED light sources, each in red, green, and blue, to enhance the visibility of indentation structures in the image. Unlike traditional single-light sources, this embodiment uses multi-color LED combination lighting to create contrast between different indentation features in the image, thereby improving the accuracy and robustness of image recognition. The LED strip 71 is bonded to the lighting module housing 72 with adhesive. The LED power cable exits from the LED power cable and signal cable routing hole 73 and passes through the camera USB routing hole 82, powered by an external power supply. The LED strip is arranged concentrically with and installed adjacent to the gel flexible layer to eliminate its reflection effect on the transparent acrylic base plate, avoiding image interference caused by light reflection and improving the quality of image acquisition.

[0056] The tactile sensing module 9 consists of a three-layer structure, including an upper contact pressure head covered with a graphite-coated gel flexible layer, a middle transparent acrylic support plate, and an external pressure compensation structure. The gel support plate 95 is made of transparent acrylic. It is placed within the acrylic support structure 92, with a gap fit ensuring a stable connection. The elastic gel flexible layer 94 is an inverted frustum shape, tightly fitted to the gel support plate 95 and placed within the gel support structure 91. The gel support structure 91 and gel flexible layer 94 are then fixed together with hot melt adhesive. The upper surface of the gel flexible layer 94 is coated with silver graphite to shield against ambient light. The gel flexible layer is the core sensing element; its deformation under pressure reflects the contact position and contact force.

[0057] The external pressure compensation structure consists of a set of elastic springs 2 positioned between the sensor and the contact indenter. These springs compensate for the pressure generated by the rope compressing the gel, improving the robustness and long-term stability of the tactile perception. The contact indenter 4, elastic springs 2, and acrylic support structure 92 together form a sandwich-like structure, a core innovation of this embodiment. This design ensures that the sensor can be installed within the robot joint and withstand significant intra-joint forces. The elastic springs 2 and contact indenter 4 are connected by an elastic connecting support portion 62, the diameter of which is slightly larger than the inner diameter of the elastic springs 2, with a clearance fit. The elastic springs 2 and acrylic support structure 92 also employ the same fitting principle. Because the sensor primarily bears pressure inside the robot joint, this design ensures stability.

[0058] The external support module 3 encapsulates the entire sensor structure with an opaque material, effectively preventing image distortion or noise caused by external light reflection from the gel surface. The external support structure 3 comprises three parts: the housing 84 of the image acquisition module, the housing 72 of the illumination module, and the housing 96 of the tactile sensing module. These three parts are fitted together with a designed protruding pin and pin hole for clearance. This module ensures that the sensor as a whole has a certain degree of compressive strength, capable of withstanding external loads from the elastic spring 2 and the contact head 4. Simultaneously, this module provides excellent external environmental isolation, enabling the sensor to maintain stable operation in environments with complex lighting, electromagnetic interference, or temperature fluctuations. Furthermore, the encapsulation structure has high mechanical strength and environmental sealing, further improving the sensor's durability and adaptability.

[0059] II. Parametric design of contact head.

[0060] The contact head mainly consists of a connecting body connector 6 and a contact head 4. The contact head 4 and the connecting body connector 6 are fixed together by a bottom mating pin 43 and a bottom mating groove 63, achieving a quick-release and quick-replacement design for the contact head. To fix it, first align the bottom mating pin 43 of the contact head with the upper mating hole of the bottom mating groove 63, then rotate the contact head clockwise to complete the fixation.

[0061] Based on the sensor design, this embodiment further refines the structural parameters of the contact indenter array 42. The gel deforms when the contact indenter compresses it, and this deformation significantly impacts the sensor's measurement range and resolution. The design parameters of the contact indenter include the contact indenter spacing and the maximum indentation depth; these two parameters critically affect the system's measurement performance (maximum measurable angle and angular resolution). For the smart joint designed in this embodiment, the following two main performance indicators are set: maximum measurable angle... and angular resolution By optimizing the parameters, the contact indenter can achieve high-precision angle measurement while ensuring the system's compact structure and operational stability, thereby improving overall sensing performance and environmental adaptability.

[0062] The distance from the top of the silicone pressure head array 43 to the contact pressure head plane 41 is defined as... The diagonal length of the contact head array is The diameter of the contact head plane 41 is Contact head height Distance between contact head Together, they determine the maximum measurable bending angle of the contact indenter. For a 3 mm thick gel flexible layer 94, the maximum pressing depth is mm, initial pressure depth mm. Therefore, the maximum measurable angle can be expressed as:

[0063] Angular resolution is affected by the minimum pixel size of the camera. Contact head spacing Maximum slope of the pressing area The angular resolution is related to the rate of change of the pressed area, which is sensed by the camera. Therefore, the resolution is determined by the camera's pixel size. The camera resolution used in this embodiment is 0.026 mm / pixel. The rate of change of the contact pressure head's area should be higher than the camera's resolution; otherwise, it cannot be sensed by the camera (the change in the outline width on the image should be greater than 1 pixel). [The last sentence appears to be incomplete and possibly refers to a different topic:] Assuming the pressed area... Inside, the maximum tangent slope is Therefore, the minimum rate of change of the profile width is When the angle changes hour, The solution derived from this It should be greater than 0.026 mm / pixel, otherwise the camera will not be able to recognize the angle change.

[0064] III. Data-driven modeling.

[0065] To achieve efficient modeling of the relationship between pose features and tactile images, this embodiment employs a data-driven approach to construct a multimodal fusion prediction model. Considering the relatively simple mapping relationship between pose features and tactile images, this embodiment designs a concise convolutional neural network structure—LeNet-6. This network contains three convolutional layers, two max-pooling layers, and one fully connected layer, used to predict key pose parameters from tactile images. Furthermore, this embodiment uses a modified attention mechanism ResNet-18 structure (Modified Attention ResNet-18, MA-ResNet-18) for the prediction of three-dimensional forces and moments, such as... Figure 19 As shown. Considering that the relative contour size formed by the four contact indenters in the tactile image is a key characteristic representing force information, this embodiment designs an image preprocessing strategy to subtract the background image from the real-time acquired image to highlight the features of the deformation area of ​​the contact indenter. Compared with the traditional ResNet-18, this embodiment adds a channel attention layer and a spatial attention layer to solve the problem of possible slight offsets in the initial position of the contact indenter, enabling the network to adaptively ignore position offsets and focus on the features of the contact indenter. At the same time, considering that low-dimensional features (such as contour and color) are also related to force information, this embodiment achieves efficient reuse of low-dimensional features and multi-scale information fusion by cascading the three output feature maps of the last layer of the network. This design significantly improves the robustness and generalization ability of the model to complex tactile scenes.

[0066] The intelligent joint in this embodiment can be applied to robotic arms and humanoid wrists.

[0067] The intelligent joint in this embodiment can be fitted into a large aspect ratio continuous robotic arm, achieving low cost, lightweight design, and multimodal sensing. It can be extended to intelligent joints in collaborative robotic arms, continuous robotic arms, etc., where the joint can be installed within the joints of a continuous robotic arm to achieve pose and force sensing. In a multi-segment continuous robotic arm connected in series, the intelligent joint can be used to measure the pose and six-axis force information of each joint.

[0068] The intelligent joint in this embodiment can be fitted onto a humanoid wrist for posture and force sensing, achieving a low-cost, lightweight intelligent joint with multimodal sensing capabilities. It can be extended to intelligent joints in humanoid wrists; the rigid elastic spring mechanism can be replaced with soft rubber to achieve the same pressure relief function. Similarly, the rope can be replaced with a drive mechanism such as an airbag, adjusting air pressure to move the contact pressure head and thus achieving the same contact pressure head compression function.

[0069] Both elastic compression springs and ropes can be replaced; flexible soft rubber and airbags can achieve the same pressure resistance and contact pressure head compression effect.

[0070] This embodiment of the force-sensing smart joint can sense both internal joint forces and external contact forces. The principle behind this smart joint's force measurement is that a neural network predicts the force by analyzing images of an indenter pressing against a gel layer. This force varies depending on the dataset; if the collected data is internal joint force data, then the internal joint force can be predicted. If the collected data is external contact force, then the external contact force can be predicted. Alternatively, two networks predicting different forces can be trained and used simultaneously to achieve the effect of simultaneously predicting both internal joint forces and external contact forces.

[0071] This embodiment forms a sandwich structure of "indenter-spring-gel". By replacing springs or elastomers with different stiffnesses, the force measurement range can be freely varied. Elastomers with lower stiffness exhibit larger deformation under unit force, resulting in higher force resolution. Conversely, elastomers with lower stiffness exhibit smaller deformation under unit force, resulting in a wider force measurement range. The load-bearing capacity can be freely varied from millinewtons to kilonewtons depending on the selection of the spring or elastomer, providing a very considerable range of load-bearing capacity and force measurement range. This structure also reduces the internal stress of the rigid-flexible coupling robot by an order of magnitude, significantly improving operational dexterity and service life.

[0072] This embodiment employs a camera-assisted tactile sensing structure within an intelligent joint. An industrial camera is installed inside the intelligent joint, along with a specific light-transmitting protective cover and light source components, to achieve multi-directional acquisition of tactile information from the joint's contact surface. This solves the problem of incomplete tactile sensing caused by the limited field of view of a single camera, providing rich raw data for subsequent 6D posture and force perception.

[0073] This embodiment employs a collaborative sensing and control method using multiple intelligent joints. The camera tactile sensing data, 6D posture and force sensing data of each intelligent joint are transmitted to the central controller of the robotic arm. Based on the sensing information from multiple joints, the controller analyzes the overall motion state of the robotic arm, formulates a collaborative control strategy, and coordinates the motion and force output of each joint to achieve stable and efficient operation of the robotic arm in scenarios such as precision assembly and material handling.

[0074] This embodiment employs an intelligent joint adaptive adjustment system based on 6D posture feedback. It compares the 6D posture data calculated by the camera's tactile sensing with the target posture in real time, generates adjustment signals for the joint drive motor, controls the motor to adjust the joint angle and position, and achieves adaptive posture correction. At the same time, it combines force sensing data to avoid damage to the joint due to excessive force during the adjustment process.

[0075] This embodiment aims to solve the problems of complex, bulky and costly high-precision pose and force sensing technologies in current robot systems, and provides a compact, low-cost and high-precision six-dimensional pose and force sensing system to realize real-time, high-precision control and interaction of robot joints in complex, dynamic and unstructured environments.

[0076] Existing pose and force sensing methods often rely on costly and complex sensor configurations, making them difficult to integrate into compact or high-degree-of-freedom robot structures, severely limiting the control accuracy and application range of robot systems. To overcome these limitations, this embodiment proposes a novel high-precision sensing module (Tac6D) based on a visual-tactile sensing mechanism. Employing a lightweight sandwich structure, it observes the compression deformation of the middle layer gel material through an upper contact indenter, and combines this with a bottom-layer camera to achieve precise perception of joint pose and force state.

[0077] This embodiment significantly improves the sensor's sensitivity, resolution, and sensing range by introducing a symmetrical contact indenter and an elastic suspension design, and by jointly optimizing the contact indenter structure and viscoelastic gel parameters. It can achieve high-frequency sensing at the sub-millisecond level (less than 6 ms), with an accuracy of 0.1° pose error and 0.081 N force error, respectively, demonstrating good real-time performance and robustness.

[0078] The sensing system proposed in this embodiment can operate stably in low-light, strong magnetic field, complex lighting, and structured environments, exhibiting excellent environmental adaptability and anti-interference capabilities. Furthermore, this embodiment further improves sensing accuracy and overall system performance by constructing a multi-feature fusion neural network sensing algorithm.

[0079] This embodiment has been successfully applied to a highly redundant robotic arm, achieving real-time feedback control and high-precision operation in complex dynamic environments. It provides a universal, efficient, and low-cost high-precision sensing solution for future multimodal intelligent robot joints. Compared to traditional six-dimensional sensors, the Tac6D sensor in this embodiment has the improvements shown in the table below:

[0080] Example 3 The difference between this embodiment and embodiment 2 is that the 6-dimensional force sensing of the human fingertip based on visual-tactile perception provided in this embodiment replaces the planar gel flexible layer with a spherical gel flexible layer or an irregularly shaped gel flexible layer, and / or replaces the spherical contact indenter with an irregularly shaped contact indenter. Combined with the original data-driven modeling method, it can be placed in the robot finger to realize 6-dimensional contact force perception of the finger joint.

[0081] like Figure 20As shown in the figure, the structure on the left connects two arms. The contact head 4 is a spherical contact head, the gel flexible layer 94 is a spherical gel flexible layer, and the elastic connector is a flexible soft rubber support 11. The structure on the right connects one arm and one hand. The contact head 4 is a spherical contact head, the gel flexible layer 94 is a spherical gel flexible layer, and the elastic connector is an airbag 13. The airbag 13 forms a sealed air cavity, and the movement of the head is controlled by air pressure.

[0082] like Figure 21 As shown in the diagram, the structure on the left connects two arms. The contact head 4 is a spherical contact head, the gel flexible layer 94 is an irregularly shaped gel flexible layer, and the elastic connector is a flexible soft rubber support 11. The structure on the right connects one arm and one hand. The contact head 4 is a spherical contact head, the gel flexible layer 94 is an irregularly shaped gel flexible layer, and the elastic connector is an airbag 13. The airbag 13 forms a sealed air cavity, and the movement of the head is controlled by air pressure.

[0083] like Figure 22 As shown in the figure, the structure on the left connects two arm bodies, the contact head 4 is an irregularly shaped contact head, the gel flexible layer 94 is a planar gel flexible layer, and the elastic connector is a flexible soft rubber support 11. The structure on the right connects one arm body and one hand body, the contact head 4 is an irregularly shaped contact head, the gel flexible layer 94 is a planar gel flexible layer, and the elastic connector is an airbag 13. The airbag 13 forms a sealed air cavity, and the movement of the pressure head is controlled by air pressure.

[0084] In other embodiments, the gel flexible layer can also be of other shapes, and the contact indenter can also be of other shapes, as long as the maximum slope of the indenter shape satisfies the original formula constraint.

[0085] like Figure 23 As shown, the left side of the arrow is a structural diagram of a single intelligent joint, and the right side of the arrow is a structural diagram of the intelligent joint and the arm assembly. Figure 23 The upper part of Figure a is a schematic diagram of the structure of a single intelligent joint, and the lower part is a schematic diagram of the cross-sectional structure along line CC of the upper part. Figure 23 The upper part of Figure b is a cross-sectional view of a single intelligent joint, and the lower part is a cross-sectional view of the upper part along line DD. Figure 23 Figure c in the diagram is a structural schematic of the intelligent joint and arm assembly. Figure 23 Figure d in the diagram is a cross-sectional structural diagram of the intelligent joint and arm assembly. For example... Figure 24 The diagram shown is a schematic of a smart joint installed in a dexterous finger to measure the internal force of the finger joint.

[0086] The present invention has a compact structure and low cost, and can realize real-time, high-precision control and interaction of robot joints in complex, dynamic and unstructured environments.

[0087] In the description of this application, it should be understood that the terms "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.

[0088] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A six-dimensional posture and force sensing intelligent joint for connecting two connecting bodies, characterized in that, include: The elastic connector, the contact pressure head (4), the lighting module (7), the image acquisition module (8), and the tactile sensing module (9) are included. The contact head (4) is mounted on one of the two connectors via a connector (6), and the connector (6) is provided with a first elastic connector support part (62). The lighting module (7), the image acquisition module (8), and the tactile sensing module (9) are disposed on the other of the two connectors; The tactile sensing module (9) is provided with a second elastic connector support part (93), and the two ends of the elastic connector are respectively connected to the first elastic connector support part (62) and the second elastic connector support part (93); The contact pressure head (4) can exert pressure on the gel flexible layer (94) on the tactile sensing module (9) during joint movement, and the gel flexible layer (94) can deform when it is pressed by the contact pressure head (4); The image acquisition module (8) can acquire images of the deformation of the deformation area on the gel flexible layer (94); the illumination module (7) can illuminate the deformation area on the gel flexible layer (94) to enhance the visibility of the indentation structure in the image.

2. The six-dimensional posture and force sensing intelligent joint according to claim 1, characterized in that, The tactile sensing module (9) includes: a gel support structure (91), an acrylic support structure (92), an elastic connector support part (93), a gel flexible layer (94), a gel support plate (95), and a tactile sensing module shell (96). The tactile sensing module housing (96) is disposed on the connector; the gel support structure (91) and the acrylic support structure (92) are disposed on the tactile sensing module housing (96); The flexible gel layer (94) is connected to the gel support plate (95); the flexible gel layer (94) is located inside the gel support structure (91), and the gel support plate (95) is located inside the acrylic support structure (92); The degree of deformation of the gel flexible layer (94) under pressure can reflect the contact position and contact force.

3. The six-dimensional posture and force sensing intelligent joint according to claim 1, characterized in that, The lighting module (7) includes: an LED light strip (71) and a lighting module housing (72); The lighting module housing (72) is disposed on the connector and located below the tactile sensing module housing (96) of the tactile sensing module (9); The LED light strip (71) is disposed on the housing (72) of the lighting module; the LED light strip (71) is used for combined lighting to make different indentation features contrast in the image, thereby improving the accuracy and robustness of image recognition; And / or, the image acquisition module (8) includes: an industrial camera, a camera acquisition card (81), a camera support frame (83), and an image acquisition module housing (84). The image acquisition module housing (84) is disposed on the connector and located below the lighting module housing (72) of the lighting module (7); the camera support bracket (83) is disposed on the image acquisition module housing (84); The camera acquisition card (81) is mounted on the camera support frame (83), and the industrial camera is mounted on the camera acquisition card (81).

4. The six-dimensional posture and force sensing intelligent joint according to claim 1, characterized in that, The contact head (4) includes: a contact head body, a contact head plane (41), and a silicone pressure head array (42). The contact head body is mounted on the connector (6), the contact head plane (41) is disposed on the contact head body, and the silicone pressure head array (42) is disposed on the contact head plane (41); The silicone compression head array (42) includes a plurality of silicone compression heads arranged in an array.

5. The six-dimensional posture and force sensing intelligent joint according to claim 4, characterized in that, The connector (6) includes: a connecting base, an elastic connector support (62), and a bottom mating groove (63) of the contact head; The connecting base is disposed on the connecting body, and the elastic connecting member support part (62) and the bottom mating groove (63) of the contact pressure head are disposed on the connecting base; The contact head body is provided with a contact head bottom mating pin (43), which can be connected with the contact head bottom mating groove (63).

6. The six-dimensional posture and force sensing intelligent joint according to claim 4, characterized in that, The vertical distance from the top of the silicone pressure head array (43) to the contact pressure head plane (41) is defined as the contact pressure head height. The diagonal length of the silicone compression head array (43) is defined as... The contact pressure head plane (41) is a circular surface, and the diameter of the contact pressure head plane (41) is defined as follows: ; through the height of the contact head Distance from the contact head Together, they determine the maximum measurable bending angle of the contact indenter (4). ; For the gel flexible layer (94) with a thickness of N mm, the maximum pressing depth is Initial pressing depth The maximum measurable angle of the contact indenter is constrained by both the pressable depth of the gel flexible layer (94) and the diameter of the contact indenter plane. The maximum measurable angle obtained from these two constraints is denoted as follows: and ; and The minimum value between the two is the final maximum measurable bending angle. Maximum measurable bending angle Represented as: The industrial camera has a resolution of M mm / pixel; defined in the pressed area. The maximum tangent slope within is ; Minimum angular resolution of the contact indenter The maximum change in cross-sectional diameter of the contact indenter as it penetrates the flexible gel layer. The height of the contact head descent when the resolution is equal to that of an industrial camera. The ratio between the contact head spacing D and half of the contact head spacing; , , D The relation is expressed as: 。 7. The six-dimensional posture and force sensing intelligent joint according to claim 1, characterized in that, The elastic connector is an elastic compression spring (2) or a flexible soft rubber support (11). And / or, the image acquisition module (8) includes: a prediction model and an attention mechanism model; The prediction model employs a convolutional neural network structure to predict key pose parameters from tactile images; the convolutional neural network structure includes 3 convolutional layers, 2 max pooling layers, and 1 fully connected layer. The attention mechanism model adopts a network structure that adds channel attention layer and spatial attention layer to the ResNet-18 structure for the prediction of three-dimensional force and torque; The image acquisition module (8) extracts the pressing contour information and pressing depth information of the color pressing indentation information in the image through the convolutional layer, and uses multiple fully connected layers to decode the pressing contour information and the pressing depth information to obtain the 2D planar rotation information of the contact head (4). Based on the pressing depth information and pressing contact situation, the joint pose and force direction are calculated.

8. The six-dimensional posture and force sensing intelligent joint according to claim 1, characterized in that, The neural network model mounted on the image acquisition module (8) can predict the force on the joint by analyzing the image data of the indentation formed by the contact indenter (4) pressing the gel flexible layer (94); When the input indentation image data is the internal force indentation data of the elastic connector deformation transmitted to the gel flexible layer (94) during joint movement, the neural network model outputs the joint internal force parameters; When the input indentation image data is external contact indentation data generated by external load acting on the joint end, the neural network model outputs external contact force parameters; And / or, by replacing elastic connectors with different stiffness specifications, the force measurement range of the joint can be adjusted; The smaller the stiffness of the elastic connector, the greater the deformation of the elastic connector under a unit force, which in turn makes the force resolution of the joint higher. The greater the stiffness of the elastic connector, the smaller the deformation of the elastic connector under a unit force, and the larger the corresponding force measurement range.

9. A robotic arm, characterized in that, Including the six-dimensional posture and force sensing intelligent joint as described in any one of claims 1 to 8.

10. The robotic arm according to claim 9, characterized in that, The robotic arm includes: Multiple six-dimensional posture and force sensing intelligent joints and multiple connecting bodies; the connecting body is an arm (1); multiple arm bodies (1) are connected in series, and two adjacent arm bodies (1) are connected through a six-dimensional posture and force sensing intelligent joint; Alternatively, the robotic arm includes: a six-dimensional posture and force sensing intelligent joint and two connectors; one connector is an arm body (1) and the other connector is a hand body (12); the arm body (1) and the hand body (12) are connected by the six-dimensional posture and force sensing intelligent joint.

Images