A control method and system for a high torque density robot joint module

Abstract

The application provides a control method and system of a high-torque-density robot joint module, and relates to the technical field of robot joint module control. The method comprises the following steps: acquiring three-dimensional image information of waste to be compressed; determining a compression resistance three-dimensional distribution map of the waste to be compressed based on the three-dimensional image information of the waste to be compressed; controlling the high-torque-density robot joint module to fix and shake the waste to be compressed based on a plurality of sets of test fixing schemes; determining an initial fixing scheme and a preliminary compression depth based on a shaking video of the waste to be compressed after being fixed according to each set of test fixing scheme; and controlling the high-torque-density robot joint module to fix the waste to be compressed and control a compressor to compress based on an adjusted fixing scheme. The method can efficiently and accurately determine an optimal fixing and compression cooperative control scheme of the waste to be compressed.

Description

Technical Field

[0001] This invention relates to the field of robot joint module control technology, specifically to a control method and system for a high torque density robot joint module. Background Technology

[0002] In industrial solid waste volume reduction and material compression molding, the waste to be compressed is complex and diverse in form, with significant differences in physical properties such as density, hardness, and looseness in different areas, directly affecting the stability and efficiency of the compression operation. Traditional waste compression fixing methods mostly rely on manual planning of fixing points or a single mechanical fixing scheme, which has obvious limitations. Manually determining fixing points depends on subjective experience, making it difficult to match with the actual mechanical properties of the waste. This can easily lead to displacement and uncontrolled deformation of the waste during compression due to insecure fixing, affecting the volume reduction effect and even causing equipment jamming and failure. A single scheme cannot adapt to different types and forms of waste, and the low matching degree between compression depth and fixing force can easily lead to insufficient compression, low volume reduction rate, and uneven distribution of fixing stress, which can aggravate equipment wear and increase energy consumption, failing to meet the requirements of high efficiency and energy saving. Existing fixing schemes still have many shortcomings, lacking precise control over the three-dimensional structure and mechanical properties of waste, failing to formulate targeted fixing strategies based on the waste form and resistance distribution, and having unreasonable fixing positions and poor stability. Existing technologies lack a robust mechanism for verifying and optimizing fixed effects, relying solely on initial parameters. This makes them ill-suited to adapting to changes in waste morphology during compression, leading to fixation failures and operational instability. Furthermore, existing technologies struggle to meet the high-precision control requirements of high-torque-density robot joint modules, failing to leverage their high torque and stability advantages, resulting in insufficient control precision and low efficiency in collaborative operations.

[0003] Therefore, how to efficiently and accurately determine the optimal control scheme for fixing and compressing the waste material to be compressed is an urgent problem to be solved. Summary of the Invention

[0004] The main technical problem solved by this invention is how to efficiently and accurately determine the optimal fixation and compression coordinated control scheme for the waste material to be compressed.

[0005] According to a first aspect, the present invention provides a control method for a high torque density robot joint module, comprising: acquiring three-dimensional image information of a waste material to be compressed; determining a three-dimensional distribution map of the compressibility resistance of the waste material to be compressed based on the three-dimensional image information; determining multiple key fixing point information and multiple auxiliary fixing point information based on the three-dimensional distribution map of the compressibility resistance of the waste material to be compressed; generating multiple sets of test fixing schemes based on the three-dimensional distribution map of the compressibility resistance of the waste material to be compressed, the multiple key fixing point information, and the multiple auxiliary fixing point information; controlling the high torque density robot joint module to fix and shake the waste material to be compressed respectively based on the multiple sets of test fixing schemes, and acquiring the data for each set. The test involves recording a video of the shaking of the waste material to be compressed after it has been fixed in place using a test fixing scheme. Based on the shaking video of the waste material to be compressed after each test fixing scheme, an initial fixing scheme and a preliminary compression depth are determined. Based on the initial fixing scheme and the preliminary compression depth, a high torque density robot joint module is controlled to fix the waste material to be compressed, and an initial compression video of the compressor on the waste material to be compressed at the preliminary compression depth is obtained. Based on the initial fixing scheme and the initial compression video of the compressor on the waste material to be compressed at the preliminary compression depth, an adjusted fixing scheme is determined. Based on the adjusted fixing scheme, a high torque density robot joint module is controlled to fix the waste material to be compressed and the compressor is controlled to compress it.

[0006] In one possible implementation, generating multiple test fixing schemes based on the three-dimensional distribution map of the compressibility resistance of the waste to be compressed, the information of multiple key fixed points, and the information of multiple auxiliary fixed points includes: constructing a map of the waste to be compressed, the map including multiple nodes and edges between the nodes, the multiple nodes including multiple key fixed nodes and multiple auxiliary fixed nodes, the node features of the key fixed nodes being key fixed point information and the three-dimensional distribution map of the compressibility resistance of the waste to be compressed, and the node features of the auxiliary fixed nodes being auxiliary fixed point information; and processing the map of the waste to be compressed based on a graph neural network to generate multiple test fixing schemes.

[0007] In one possible implementation, determining the initial fixing scheme and preliminary compression depth based on the shaking video of the waste to be compressed after fixing each set of test fixing schemes includes: determining multiple reasonable fixing point information for each set of test fixing schemes using a reasonable fixing point determination model based on the shaking video of the waste to be compressed after fixing each set of test fixing schemes; clustering multiple clusters based on the multiple reasonable fixing point information for each set of test fixing schemes; and determining the initial fixing scheme and preliminary compression depth based on the multiple clusters.

[0008] In one possible implementation, the reasonable fixed point determination model is a recurrent neural network model.

[0009] According to a second aspect, the present invention provides a control system for a high torque density robot joint module, comprising: an acquisition module for acquiring three-dimensional image information of a waste material to be compressed; a compression resistance determination module for determining a three-dimensional distribution map of the compression resistance of the waste material to be compressed based on the three-dimensional image information of the waste material to be compressed; a fixing point determination module for determining multiple key fixing point information and multiple auxiliary fixing point information based on the three-dimensional distribution map of the compression resistance of the waste material to be compressed; a test scheme generation module for generating multiple test fixing schemes based on the three-dimensional distribution map of the compression resistance of the waste material to be compressed, the multiple key fixing point information, and the multiple auxiliary fixing point information; and a shaking test module for controlling the high torque density robot joint module to fix and shake the waste material to be compressed based on the multiple test fixing schemes, and obtaining... The system includes: a shaking video of the waste material to be compressed after each set of test fixing schemes; an initial scheme determination module, used to determine the initial fixing scheme and preliminary compression depth based on the shaking video of the waste material to be compressed after each set of test fixing schemes; an initial compression module, used to control a high torque density robot joint module to fix the waste material to be compressed based on the initial fixing scheme and the preliminary compression depth, and to acquire an initial compression video of the compressor on the waste material to be compressed at the preliminary compression depth; an adjustment scheme determination module, used to determine an adjustment fixing scheme based on the initial fixing scheme and the initial compression video of the compressor on the waste material to be compressed at the preliminary compression depth; and a compression execution module, used to control the high torque density robot joint module to fix the waste material to be compressed based on the adjustment fixing scheme and to control the compressor to compress it.

[0010] In one possible implementation, the test scheme generation module is further configured to: construct a graph of the waste to be compressed, the graph including multiple nodes and edges between the nodes, the multiple nodes including multiple key fixed nodes and multiple auxiliary fixed nodes, the node features of the key fixed nodes being key fixed point information and a three-dimensional distribution map of the compressibility resistance of the waste to be compressed, and the node features of the auxiliary fixed nodes being auxiliary fixed point information; and process the graph of the waste to be compressed based on a graph neural network to generate multiple test fixing schemes.

[0011] In one possible implementation, the initial scheme determination module is further configured to: determine multiple reasonable fixing point information for each set of test fixing schemes based on the shaking video of the waste to be compressed after each set of test fixing schemes is fixed, using a reasonable fixing point determination model; perform clustering based on the multiple reasonable fixing point information for each set of test fixing schemes to obtain multiple clusters; and determine the initial fixing scheme and preliminary compression depth based on the multiple clusters.

[0012] In one possible implementation, the reasonable fixed point determination model is a recurrent neural network model.

[0013] According to a third aspect, embodiments of the present invention provide an electronic device, including: a processor; a memory; and a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement the method described above, the method including: acquiring three-dimensional image information of a waste material to be compressed; determining a three-dimensional distribution map of the compressibility resistance of the waste material to be compressed based on the three-dimensional image information of the waste material to be compressed; determining multiple key fixed point information and multiple auxiliary fixed point information based on the three-dimensional distribution map of the compressibility resistance of the waste material to be compressed; generating multiple sets of test fixing schemes based on the three-dimensional distribution map of the compressibility resistance of the waste material to be compressed, the multiple key fixed point information, and the multiple auxiliary fixed point information; and controlling a high torque density machine based on the multiple sets of test fixing schemes. The robot joint module fixes and shakes the waste material to be compressed, and acquires a shaking video of the waste material after fixing with each test fixing scheme; based on the shaking video of the waste material to be compressed with each test fixing scheme, an initial fixing scheme and an initial compression depth are determined; based on the initial fixing scheme and the initial compression depth, the high torque density robot joint module is controlled to fix the waste material to be compressed, and an initial compression video of the compressor on the waste material to be compressed at the initial compression depth is acquired; based on the initial fixing scheme and the initial compression video of the compressor on the waste material to be compressed with the initial compression depth, an adjustment fixing scheme is determined; based on the adjustment fixing scheme, the high torque density robot joint module is controlled to fix the waste material to be compressed and the compressor is controlled to compress it.

[0014] According to a fourth aspect, this embodiment provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the program implements the aforementioned control method for a high torque density robot joint module. The method includes: acquiring three-dimensional image information of a waste material to be compressed; determining a three-dimensional distribution map of the compressibility resistance of the waste material based on the three-dimensional image information; determining multiple key fixed point information and multiple auxiliary fixed point information based on the three-dimensional distribution map of the compressibility resistance of the waste material; generating multiple sets of test fixing schemes based on the three-dimensional distribution map of the compressibility resistance of the waste material, the multiple key fixed point information, and the multiple auxiliary fixed point information; and controlling the high torque density robot joint module based on the multiple sets of test fixing schemes. The waste material to be compressed is fixed and shaken separately, and a shaking video of the waste material to be compressed after fixing with each test fixing scheme is acquired; based on the shaking video of the waste material to be compressed after fixing with each test fixing scheme, an initial fixing scheme and an initial compression depth are determined; based on the initial fixing scheme and the initial compression depth, a high torque density robot joint module is controlled to fix the waste material to be compressed, and an initial compression video of the waste material to be compressed by the compressor at the initial compression depth is acquired; based on the initial fixing scheme and the initial compression video of the waste material to be compressed by the compressor at the initial compression depth, an adjusted fixing scheme is determined; based on the adjusted fixing scheme, a high torque density robot joint module is controlled to fix the waste material to be compressed and the compressor is controlled to compress it.

[0015] This invention provides a control method and system for a high torque density robot joint module. The method includes acquiring three-dimensional image information of waste material to be compressed; determining a three-dimensional distribution map of the compressibility resistance of the waste material based on the three-dimensional image information; determining multiple key fixing point information and multiple auxiliary fixing point information based on the three-dimensional distribution map of the compressibility resistance of the waste material; generating multiple sets of test fixing schemes based on the three-dimensional distribution map of the compressibility resistance of the waste material, the multiple key fixing point information, and the multiple auxiliary fixing point information; controlling the high torque density robot joint module to fix and shake the waste material to be compressed according to the multiple sets of test fixing schemes, and acquiring the data of the waste material after fixing according to each set of test fixing schemes. The method involves several steps: First, a shaking video is used. Based on the shaking video of the waste material to be compressed after each set of test fixing schemes, an initial fixing scheme and preliminary compression depth are determined. Then, based on the initial fixing scheme and the preliminary compression depth, a high-torque-density robot joint module is controlled to fix the waste material to be compressed, and an initial compression video of the compressor at the preliminary compression depth is obtained. Next, an adjusted fixing scheme is determined based on the initial fixing scheme and the initial compression video of the compressor at the preliminary compression depth. Finally, based on the adjusted fixing scheme, a high-torque-density robot joint module is controlled to fix the waste material to be compressed, and the compressor is controlled to compress it. This method can efficiently and accurately determine the optimal fixing and compression coordinated control scheme for the waste material to be compressed. Attached Figure Description

[0016] Figure 1 A flowchart illustrating a control method for a high torque density robot joint module provided in an embodiment of the present invention;

[0017] Figure 2 A flowchart illustrating the generation of multiple fixed test schemes is provided in an embodiment of the present invention.

[0018] Figure 3 A schematic diagram of a high torque density robot provided in an embodiment of the present invention;

[0019] Figure 4 This is a schematic diagram of a process for determining an initial fixing scheme and a preliminary compression depth, provided in an embodiment of the present invention.

[0020] Figure 5 This is a schematic diagram of a control system for a high torque density robot joint module provided in an embodiment of the present invention. Detailed Implementation

[0021] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings. Similar elements in different embodiments are referred to by associated similar element reference numerals. In the following embodiments, many details are described to facilitate a better understanding of the invention. However, those skilled in the art will readily recognize that some features may be omitted in different situations, or may be replaced by other elements, materials, or methods. In some cases, certain operations related to the present invention are not shown or described in the specification. This is to avoid obscuring the core parts of the invention with excessive description. For those skilled in the art, detailed description of these related operations is not necessary; they can fully understand the related operations based on the description in the specification and general technical knowledge in the art.

[0022] In this embodiment of the invention, the following are provided: Figure 1 The method for controlling a high torque density robot joint module, as shown, includes steps S1 to S9:

[0023] Step S1: Obtain three-dimensional image information of the waste to be compressed.

[0024] Waste to be compressed refers to industrial irregular metal parts that remain after machining and have no uniform standard of shape, such as alloy blocks containing casting flash, welding scrap fragments, and irregular metal connector remnants.

[0025] The 3D image information of the waste to be compressed is data containing the spatial geometry of the waste obtained by scanning the waste with a 3D scanner. The 3D image information of the waste to be compressed includes the coordinate data of each point on the surface of the waste in 3D space, the color and texture data of the waste, and the overall volume and outline data of the waste.

[0026] Step S2: Determine the three-dimensional distribution map of the compressibility resistance of the waste material to be compressed based on the three-dimensional image information of the waste material to be compressed.

[0027] In some embodiments, a compressibility resistance analysis model can be used to determine the three-dimensional distribution map of the compressibility resistance of the waste material to be compressed. The compressibility resistance analysis model is a convolutional neural network. The input to the compressibility resistance analysis model is the three-dimensional image information of the waste material to be compressed, and the output of the compressibility resistance analysis model is the three-dimensional distribution map of the compressibility resistance of the waste material to be compressed.

[0028] Convolutional Neural Networks (CNNs) can be applied to image processing and feature extraction tasks. A CNN can consist of multiple layers and must contain at least one or more of the following: convolutional layers (CONV), rectified linear unit (ReLU) layers, pooling layers, or fully connected layers (FC). CNNs can extract and combine local features layer by layer from input image data, gradually uncovering deeper information about the image, thereby enabling the analysis and judgment of image targets.

[0029] The three-dimensional distribution map of the compressibility resistance of the waste to be compressed is a spatial distribution map that describes the ability of different locations inside the waste to resist compressive deformation, output by the compressibility resistance analysis model.

[0030] The three-dimensional distribution map of the compression resistance of the waste material to be compressed includes the resistance coefficient value, material hardness grade, and estimated deformation threshold of each voxel unit in the waste space.

[0031] The three-dimensional image information of the waste to be compressed provides the external geometry and texture features of the waste, and these appearance features are mapped to the material distribution and structural strength of the waste to be compressed.

[0032] Convolutional neural networks (CNNs) can extract spatial features from the 3D image information of the waste to be compressed through convolutional layers, thereby identifying the texture and shape features of different material regions such as rigid metal components and deformable plastic components. Through nonlinear transformations of multiple layers, CNNs can establish the correlation between the appearance and internal structural strength of the waste to be compressed, thereby calculating the drag value at each point in space and ultimately generating a 3D distribution map of the compressibility drag of the waste.

[0033] Step S3: Based on the three-dimensional distribution map of the compression resistance of the waste material to be compressed, determine information on multiple key fixed points and multiple auxiliary fixed points.

[0034] In some embodiments, a fixed-point determination model can be used to determine multiple key fixed-point information and multiple auxiliary fixed-point information. The fixed-point determination model is a convolutional neural network. The input to the fixed-point determination model is a three-dimensional distribution map of the compressibility resistance of the waste material to be compressed, and the output of the fixed-point determination model is multiple key fixed-point information and multiple auxiliary fixed-point information.

[0035] Critical anchor point information refers to the location data of suitable points on the waste material to be compressed, determined by the anchor point determination model, which are suitable as the main force support positions. Critical anchor point information includes the three-dimensional spatial coordinates of the critical anchor point, the maximum normal pressure value that the critical anchor point can withstand, and the friction coefficient of the surface at that point.

[0036] Auxiliary fixed point information refers to the location data of secondary force-bearing positions on the waste to be compressed, determined by the fixed point determination model, which are used to assist in balancing and restrict the degrees of freedom. Auxiliary fixed point information includes the three-dimensional spatial coordinates of the auxiliary fixed point, the surface curvature of the auxiliary fixed point, and the spatial distance of the point relative to the critical fixed point.

[0037] The three-dimensional distribution map of the compression resistance of the waste material to be compressed can accurately quantify the structural strength of each part of the waste material and clearly present the compression resistance of each region of the waste material. The difference in resistance distribution can directly affect the selection of the fixed point, providing a structural mechanics basis for the model to find a stable gripping position.

[0038] Convolutional neural networks (CNNs) utilize the receptive field mechanism to search for localized regions of high resistance concentration in the three-dimensional distribution map of the compressibility resistance of the waste material to be compressed. By analyzing the gradient changes in the resistance distribution, CNNs can identify the center of the region where the structure is most stable and less prone to collapse as a key fixed point. Based on the principle of mechanical equilibrium, CNNs then search for suitable locations for auxiliary supports around the key point, thereby determining information on multiple key fixed points and multiple auxiliary fixed points.

[0039] In some embodiments, determining multiple key fixed point information and multiple auxiliary fixed point information based on the three-dimensional distribution map of the compressibility resistance of the waste to be compressed includes steps S31 to S33:

[0040] Step S31: Based on the three-dimensional distribution map of the compressibility resistance of the waste to be compressed, determine multiple high compressibility resistance core regions, multiple medium and low compressibility resistance core regions, boundary information of resistance change regions, and pressure resistance level of each core region.

[0041] In some embodiments, a convolutional neural network can be used to determine multiple high compressive resistance core regions, multiple medium and low compressive resistance core regions, boundary information of resistance change regions, and the pressure resistance level of each core region.

[0042] Multiple high compressibility core regions are spatial volume ranges of the waste material to be compressed, determined by a convolutional neural network, that can withstand high-intensity compression and are not easily deformed.

[0043] Multiple low-to-medium compressibility core regions are spatial volume ranges in which the structure of the waste to be compressed is prone to deformation, as determined by a convolutional neural network.

[0044] The boundary information of the resistance abrupt change region is determined by a convolutional neural network, which shows the boundary locations between regions with different resistance intensities within the waste material to be compressed. This boundary information includes the coordinates of the gradient band where the resistance value changes drastically and the thickness data of the material transition zone.

[0045] The pressure resistance level of each core area is a numerical indicator that quantifies the maximum pressure that each core area can withstand, determined by a convolutional neural network.

[0046] Convolutional neural networks (CNNs) extract spatial texture features and density variation patterns by performing 3D convolution operations on the 3D distribution map of compressibility drag. They can identify pixel blocks with continuous and high drag values ​​as high-drag cores and capture the boundaries of drag value jumps through edge detection mechanisms. Utilizing its semantic segmentation capabilities, the CNN can discretize the continuous distribution map into multiple high-compressibility drag core regions, multiple medium- and low-compressibility drag core regions, and boundary information of drag abrupt change regions, and calculate the compressibility level of each core region based on the feature intensity within each region.

[0047] Step S32: Based on the multiple high compressive resistance core regions, the multiple medium and low compressive resistance core regions, the boundary information of the resistance change region, and the pressure resistance level of each core region, determine multiple adaptive representative points, the collapse resistance score of each adaptive representative point, the stress stability coefficient, and the complementary support weight value.

[0048] In some embodiments, a deep neural network model can be used to determine multiple adaptive representative points, the collapse resistance score of each adaptive representative point, the stress stability coefficient, and the complementary support weight value.

[0049] Deep neural network models include deep neural networks (DNNs), which are feedforward artificial neural networks containing multiple hidden layers. Through hierarchical nonlinear transformations, deep neural networks can simulate complex functional relationships between inputs and outputs. Each neuron in a deep neural network layer receives the output of the previous layer as input and can be computed using weight matrices and activation functions. Deep neural networks possess powerful feature learning and fitting capabilities and can handle high-dimensional input data and perform complex classification or regression predictions.

[0050] Multiple adaptive representative points are selected from the core regions through a deep neural network, representing potential optimal grab position coordinates.

[0051] The collapse resistance score for each adaptive representative point is a quantitative assessment of the structural integrity maintained by the representative point under vertical compression, obtained through a deep neural network.

[0052] The stress stability coefficient is an index determined by a deep neural network that indicates the ability of a representative point to remain non-slip when subjected to lateral shear force or torque.

[0053] The complementary support weight value is calculated by a deep neural network and is an indicator that measures whether a representative point can form a stable mechanical structure when combined with other points.

[0054] The complementary support weight value includes the contribution of the point to the overall moment balance, the spatial geometric distance score from other points, and the load distribution weight when multiple points are subjected to force together.

[0055] Deep neural networks, by inputting the geometric features and physical properties of various regions, can learn the nonlinear mapping relationship between the overall characteristics of the region and the local optimal geometric points. Simultaneously, deep neural networks can comprehensively consider the shape, size, and boundary constraints of the region, thereby inferring the most suitable coordinates for applying force. Utilizing its regression analysis capabilities, and based on the region's stress resistance level and boundary stability, deep neural networks can efficiently predict the physical response of each point after being subjected to force, thus accurately calculating multiple representative adaptive points and their corresponding collapse resistance scores, stress stability coefficients, and complementary support weight values.

[0056] Step S33: Based on the multiple adaptive representative points, the collapse resistance score of each adaptive representative point, the stress stability coefficient, and the complementary support weight value, determine multiple key fixed point information and multiple auxiliary fixed point information.

[0057] In some embodiments, a deep neural network model can be used to determine multiple key fixed point information and multiple auxiliary fixed point information.

[0058] Deep neural networks, through weighted analysis of representative input points and their multidimensional scoring indicators, can identify which points possess extremely high resistance to collapse and stability characteristics to become primary load-bearing points. By analyzing complementary support weights, the model can find combinations that spatially maximize the support for key points to maintain balance.

[0059] Step S4: Generate multiple test fixing schemes based on the three-dimensional distribution map of the compression resistance of the waste to be compressed, the information of the multiple key fixing points, and the information of the multiple auxiliary fixing points.

[0060] In some embodiments, Figure 2 This is a flowchart illustrating the generation of multiple test fixation schemes according to an embodiment of the present invention. The generation of multiple test fixation schemes includes steps S41 to S42:

[0061] Step S41: Construct a map of the waste to be compressed. The map of the waste to be compressed includes multiple nodes and edges between the multiple nodes. The multiple nodes include multiple key fixed nodes and multiple auxiliary fixed nodes. The node features of the key fixed nodes are key fixed point information and a three-dimensional distribution map of the compressibility resistance of the waste to be compressed. The node features of the auxiliary fixed nodes are auxiliary fixed point information.

[0062] The waste map to be compressed is a topological data structure that can be used to abstractly represent the spatial relationships and attribute associations between various potential fixed points on the waste. Multiple key fixed nodes in the waste map correspond to multiple key fixed points, and multiple auxiliary fixed nodes correspond to multiple auxiliary fixed points. The edges between nodes represent the associations between different fixed nodes, and the edges indicate the positional relationships between nodes.

[0063] Positional relationships include the relative orientation and spatial distance of fixed points in three-dimensional space.

[0064] Relative directions include directly above, to the left, and diagonally in front.

[0065] Spatial distance includes straight-line distance and path distance.

[0066] Step S42: Process the graph of the waste to be compressed based on the graph neural network to generate multiple test fixation schemes.

[0067] Graph Neural Networks (GNNs) are neural network models capable of processing graph-structured data. GNNs utilize the feature information of nodes and the edge information between nodes in a graph, and achieve information interaction and aggregation between nodes through message passing mechanisms, thereby learning global and local features of the graph. The input to the GNN is the graph of the waste to be compressed, and the output of the GNN is multiple sets of fixed test schemes.

[0068] Multiple test fixation schemes are combinations of different robotic arm fixation and gripping strategies for the waste to be compressed, output by a graph neural network. Each test fixation scheme includes a selected combination of key fixation points, a selected combination of auxiliary fixation points, the direction of force applied to each fixation point, and the magnitude of the force applied to each fixation point. The fixation point combinations in each test fixation scheme include all key fixation points and some auxiliary fixation points.

[0069] By constructing a graph of the waste to be compressed, the core object attributes of key and auxiliary fixed nodes in the waste fixing scenario can be clearly presented, thus providing a clear data foundation for the selection of fixed solution combinations. Using key fixed point information and the 3D distribution map of the compressibility resistance of the waste to be compressed as features of key fixed nodes, and auxiliary fixed point information as features of auxiliary fixed nodes, while considering positional relationships as edges between nodes, this fully integrates core data and related information related to waste fixing. Furthermore, it helps the graph neural network comprehensively grasp the intrinsic value, adaptability, and spatial distribution relationships of each fixed point, thereby accurately generating multiple test fixing solutions that meet actual needs. Compared to traditional fixing solution design methods, graph neural networks combined with the waste to be compressed graph can more efficiently discover potential adaptable solutions for fixed point combinations.

[0070] Graph neural networks (Graph Neural Networks) can analyze the interactions and synergies between different fixed points by performing message passing and aggregation operations on the graph of the waste to be compressed. They can also assess the overall mechanical stability of different node combinations and infer the optimal node connection pattern. Through message passing, the Graph Neural Network can capture the positional relationships between key fixed nodes and auxiliary fixed nodes. Combining information on multiple key fixed points carried by key fixed nodes, the three-dimensional distribution map of the compressibility resistance of the waste to be compressed, and information on multiple auxiliary fixed points carried by auxiliary fixed nodes, the Graph Neural Network can quantify the spatial adaptability of different node combinations and prioritize the retention of all key fixed nodes to ensure core fixation effectiveness. The Graph Neural Network can also select auxiliary fixed nodes that match the spatial distribution of key fixed nodes and can compensate for weak areas in the fixation, forming multiple combinations of all key fixed nodes plus some auxiliary fixed nodes. Then, based on the resistance values ​​corresponding to each node in the three-dimensional distribution map of the compressibility resistance of the waste to be compressed, the Graph Neural Network can calculate the required force for each fixed point. Areas with high resistance require high torque to ensure firm fixation, while areas with low resistance require moderate force to avoid waste deformation. Meanwhile, the model can determine the direction of force application by combining the positional relationship between nodes to ensure that the direction of force application matches the spatial orientation of the fixed point, thereby avoiding the generation of lateral force that causes the waste to shift. Finally, through multiple rounds of feature aggregation and combination optimization, the model can generate multiple sets of differentiated parameter combinations. Each set of combinations includes a complete set of selected key fixed points, a set of selected auxiliary fixed points, the direction of force application for each fixed point, and the magnitude of force application for each fixed point, ultimately forming multiple test fixing schemes that meet the test requirements.

[0071] Step S5: Based on the multiple test fixing schemes, control the high torque density robot joint module to fix and shake the waste to be compressed, and obtain the shaking video of the waste to be compressed after fixing by each test fixing scheme.

[0072] The shaking video of the waste to be compressed after each set of test fixing schemes refers to the video captured by the video shooting equipment when the waste is shaken by controlling the high torque density robot joint module after each set of test fixing schemes is used to fix the waste. Figure 3 This is a schematic diagram of a high torque density robot provided in an embodiment of the present invention.

[0073] The video recording of the shaking of the waste material to be compressed after each set of test fixing schemes can record the displacement changes, deformation process, and loosening of the fixing points of the waste material during the shaking process.

[0074] Step S6: Determine the initial fixing scheme and preliminary compression depth based on the shaking video of the waste to be compressed after each set of test fixing schemes.

[0075] In some embodiments, Figure 4 This is a flowchart illustrating the process of determining an initial fixing scheme and a preliminary compression depth according to an embodiment of the present invention. The determination of the initial fixing scheme and the preliminary compression depth includes steps S61 to S63:

[0076] Step S61: Based on the shaking video of the waste to be compressed after each set of test fixing schemes is fixed, use the reasonable fixing point determination model to determine multiple reasonable fixing point information for each set of test fixing schemes.

[0077] The reasonable fixing point determination model is a recurrent neural network model. The input of the reasonable fixing point determination model is the shaking video of the waste to be compressed after each set of test fixing schemes is fixed, and the output of the reasonable fixing point determination model is the information of multiple reasonable fixing points for each set of test fixing schemes.

[0078] Recurrent Neural Network (RNN) models are a type of neural network capable of processing sequential data. RNNs possess memory capabilities, and connections exist between the nodes in their hidden layers, ensuring that the output at the current time step depends not only on the current input but also on the hidden layer state from the previous time step. This recurrent connection structure allows RNNs to capture the temporal dependencies and dynamic patterns within sequential data.

[0079] The information on multiple reasonable fixing points for each test fixing scheme is derived from a reasonable fixing point determination model. This data, verified through actual shaking tests, shows good stability and no detachment or excessive displacement in the shaking video of the waste material to be compressed after fixing with each scheme. The reasonable fixing point information includes the three-dimensional coordinates of the points that have passed stability verification, the maximum displacement during shaking, and the rate of change of the gripping contact area.

[0080] The video recordings of the shaking of the waste material to be compressed after each set of test fixing schemes documented the continuous motion trajectory and morphological changes of the waste material under dynamic external forces. The time-series frames of the videos contain visual evidence of whether the fixing points experienced relative slippage, vibration, or detachment. This temporal visual data can provide dynamic behavioral characteristics for determining the reliability of fixing points under actual physical disturbances in a reasonable fixing point determination model.

[0081] Recurrent neural networks (RNNs) can analyze the motion characteristics of fixed points over time by processing continuous frame sequences of swaying videos. Leveraging their memory mechanism, RNNs can focus on the temporal correlations between video frames, thereby identifying which points remain relatively stationary or experience minor displacement throughout the swaying cycle. Furthermore, RNNs can extract temporal features of stability from dynamic videos, effectively distinguishing between stable and loose points, and thus determining multiple reasonable fixed points for each test fixation scheme.

[0082] Step S62: Cluster the multiple reasonable fixed points information of each set of test fixed schemes to obtain multiple clusters.

[0083] The clustering method described is K-means clustering, a commonly used unsupervised learning clustering algorithm. The value of K can be preset manually. K-means clustering divides the dataset into K predetermined clusters using a pre-defined number of clusters K. Through iterative calculation, it finds the optimal cluster centers, minimizing the sum of distances from data points within a cluster to the cluster center while maximizing the distances between data points in different clusters, thus achieving data grouping and classification.

[0084] Multiple clusters refer to multiple data groups obtained by clustering the reasonable fixing point information of all multiple test fixing schemes using the K-means clustering algorithm. Each cluster contains reasonable fixing point information of several test fixing schemes, and the reasonable fixing point information within the same cluster has a high degree of similarity. However, the reasonable fixing point information between different clusters shows significant differences in core features such as the three-dimensional coordinate distribution of the points, the maximum displacement during the shaking process, and the rate of change of the gripping contact area, corresponding to different fixing effect performance types.

[0085] The process of clustering reasonable fixed-point information from multiple test fixed schemes using the K-means clustering algorithm is as follows: First, K groups of reasonable fixed-point information are randomly selected from the reasonable fixed-point information dataset as initial cluster centers. Next, for each group of reasonable fixed-point information in the dataset, Euclidean distance is used to measure and calculate its distance to these K initial cluster centers, and the reasonable fixed-point information is assigned to the corresponding cluster according to the principle of closest proximity. After all reasonable fixed-point information has been partitioned, the average value of each feature of the reasonable fixed-point information within each cluster is recalculated to update the cluster center of each cluster. This process of partitioning and updating cluster centers is repeated until the change in cluster centers is minimal. At this point, the clustering process is considered to have converged, thus completing K-means clustering.

[0086] Clustering effectively isolates invalid interference items from the data by dividing the reasonable fixing point information of all multiple test fixing schemes into multiple clusters, focusing on the core stability characteristics under different fixing strategies. The reasonable fixing point information of all multiple test fixing schemes includes multiple sets of three-dimensional coordinates of points, maximum displacement during shaking, and rate of change of the grasping contact area, among other characteristic parameters. This data is multi-dimensional and scattered, making it susceptible to local case influences when directly deriving the baseline fixing strategy. Clustering, however, groups reasonable fixing point information with similar characteristics into the same cluster, forming a clear fixing pattern classification. This allows the originally scattered parameters to form a directional set of patterns. Through feature analysis of multiple clusters, the adaptation conditions and stability performance of different fixing patterns can be quickly distinguished.

[0087] Step S63: Determine the initial fixing scheme and preliminary compression depth based on the multiple clusters.

[0088] In some embodiments, a fixed-scheme determination model can be used to determine the initial fixed scheme and the initial compression depth. The fixed-scheme determination model is a deep neural network model. The input to the fixed-scheme determination model is the plurality of clusters, and the output of the fixed-scheme determination model is the initial fixed scheme and the initial compression depth.

[0089] The initial fixation scheme is the robotic arm operation strategy generated from the model to fix the waste to be compressed during the initial compression attempt. The initial fixation scheme includes the spatial coordinates of the preferred fixation points, the direction of force application by the robotic arm end effector, and the magnitude of the applied force.

[0090] The initial compression depth is the vertical displacement value determined by the fixed scheme model for the trial compression of the waste material to be compressed.

[0091] Multiple clusters are the result of cluster analysis of reasonable fixed point information from multiple test fixing schemes. Each cluster represents a set of fixed points with similar stability performance and spatial distribution characteristics. The data of multiple clusters include the average displacement, contact area stability, and spatial clustering of different points under shaking tests. These characteristic data can directly reflect the actual stability of different parts of the waste material under stress.

[0092] Deep neural networks, through weighted analysis of feature data from multiple clusters, can identify which cluster's set of fixed points exhibits the highest stability characteristics during shaking tests, such as minimal displacement and lowest contact area fluctuation. The deep neural network can map the cluster center of this optimal cluster to specific spatial coordinates and, combined with the anti-slip characteristics of reasonable fixed points within the cluster, calculate the optimal force direction and magnitude to maintain equilibrium, thus generating an initial fixing scheme. Simultaneously, based on the drag characteristic data contained in the optimal cluster, the deep neural network can analyze the potential deformation trend of the region under pressure, and then use regression algorithms to predict a displacement limit value that produces effective deformation without causing immediate failure of the fixed points, determining this as the initial compression depth.

[0093] Step S7: Based on the initial fixing scheme and the initial compression depth control high torque density robot joint module, fix the waste to be compressed, and obtain the initial compression video of the compressor on the waste to be compressed at the initial compression depth.

[0094] The initial compression video of the compressor on the waste material to be compressed at the initial compression depth refers to the video captured by a video recording device when the compressor compresses the waste material to be compressed at the initial compression depth after the waste material is fixed by the initial fixing scheme.

[0095] The compressor can record the initial compression video of the waste material to be compressed at the initial compression depth, including the compression deformation and the stress stability of the fixed point.

[0096] Step S8: Based on the initial fixing scheme and the initial compression video of the waste to be compressed at the initial compression depth, the compressor determines the adjustment fixing scheme.

[0097] In some embodiments, an adjustment scheme determination model can be used to adjust the fixing scheme. The adjustment scheme determination model is a Transformer model. The inputs to the adjustment scheme determination model are the initial fixing scheme and the initial compression video of the waste to be compressed by the compressor at the initial compression depth. The output of the adjustment scheme determination model is the adjusted fixing scheme.

[0098] The Transformer model is a deep learning model based on the self-attention mechanism. It primarily consists of an encoder and a decoder. The encoder transforms the input sequence into a sequence of feature vectors containing contextual information, while the decoder uses these features to generate the output sequence. The Transformer model abandons traditional recurrent and convolutional structures, relying entirely on attention mechanisms to capture long-range dependencies and global features in the input sequence. It possesses extremely strong parallel computing and sequence modeling capabilities.

[0099] The adjusted fixation scheme is an optimized robotic arm fixation strategy generated by correcting deficiencies in the initial compression process using the adjusted scheme determination model. The adjusted fixation scheme includes the corrected coordinates of the fixation points, the adjusted direction and magnitude of the applied forces, and compensation torque parameters to enhance stability.

[0100] The initial fixing scheme provides the baseline operating parameters for compression testing, clarifying the initial settings of the force points and force application states. The initial compression video of the compressor on the waste material to be compressed at the initial compression depth visually records the actual dynamic response of the waste material under the baseline operating parameters. The video includes temporal visual information on the stretching of the waste material surface texture, the slight slippage around the fixing points, and the non-uniform deformation of the structure.

[0101] The Transformer model, through its encoder, extracts frame-level features from the initial compression video of the waste material to be compressed at the initial compression depth. Utilizing a self-attention mechanism, it accurately captures the time and spatial locations of minute displacements at fixed points or unexpected tilting of the waste material within the video sequence. The Transformer model correlates these dynamic failure features in the video with the input initial fixing parameters, learning the mechanical response of the fixed points under specific compression loads. If the model detects a slippage trend at a fixed point on one side of the video, it calculates a compensating displacement in the opposite direction or increases the normal pressure in that direction using the decoder. If the model identifies a torsional moment generated during compression, it adjusts the force direction to counteract the moment. Ultimately, the Transformer model integrates these visual feedback-based correction strategies to output an adjusted fixing scheme that adapts to the actual deformation characteristics of the waste material.

[0102] Step S9: Based on the adjustment and fixing scheme, control the high torque density robot joint module to fix the waste to be compressed and control the compressor to compress it.

[0103] Once the adjustment and fixing scheme is determined, the high torque density robot joint module is controlled to precisely clamp the waste to be compressed according to the combination of fixing points, the direction and magnitude of force applied to each fixing point in the scheme, ensuring that the waste is fixed firmly and without deviation. At the same time, the compressor is linked to perform the compression operation on the waste to be compressed.

[0104] Based on the same inventive concept Figure 5 This is a schematic diagram of a control system for a high torque density robot joint module provided in an embodiment of the present invention. The control system of the high torque density robot joint module includes:

[0105] The acquisition module 71 is used to acquire three-dimensional image information of the waste to be compressed;

[0106] Compression resistance determination module 72 is used to determine the three-dimensional distribution map of the compression resistance of the waste material to be compressed based on the three-dimensional image information of the waste material to be compressed;

[0107] The fixed point determination module 73 is used to determine multiple key fixed point information and multiple auxiliary fixed point information based on the three-dimensional distribution map of the compression resistance of the waste to be compressed;

[0108] The test scheme generation module 74 is used to generate multiple test fixing schemes based on the three-dimensional distribution map of the compressibility resistance of the waste to be compressed, the information of the multiple key fixing points, and the information of the multiple auxiliary fixing points.

[0109] The shaking test module 75 is used to control the high torque density robot joint module to fix and shake the waste to be compressed based on the multiple test fixing schemes, and to acquire the shaking video of the waste to be compressed after fixing each test fixing scheme.

[0110] The initial scheme determination module 76 is used to determine the initial fixing scheme and the initial compression depth based on the shaking video of the waste to be compressed after each set of test fixing schemes is fixed.

[0111] The initial compression module 77 is used to fix the waste to be compressed based on the initial fixing scheme and the initial compression depth control high torque density robot joint module, and to acquire the initial compression video of the waste to be compressed by the compressor at the initial compression depth.

[0112] The adjustment scheme determination module 78 is used to determine the adjustment and fixing scheme based on the initial fixing scheme and the initial compression video of the waste to be compressed by the compressor at the initial compression depth.

[0113] The compression execution module 79 is used to control the high torque density robot joint module to fix the waste to be compressed and control the compressor to compress it based on the adjustment and fixing scheme.

[0114] It should be noted that, in order to simplify the descriptions disclosed herein and thus aid in the understanding of one or more embodiments of the invention, the foregoing description of embodiments of this specification may sometimes combine multiple features into a single embodiment, drawing, or description thereof. However, this method of disclosure does not imply that the subject matter of this specification requires more features than those mentioned in the claims. In fact, the embodiments contain fewer features than all the features of a single embodiment disclosed above.

[0115] Finally, it should be understood that the embodiments described in this specification are merely illustrative of the principles of the embodiments described herein. Other variations may also fall within the scope of this specification. Therefore, alternative configurations of the embodiments described herein are intended to be illustrative rather than limiting, and should be considered consistent with the teachings of this specification. Accordingly, the embodiments described herein are not limited to those explicitly introduced and described herein.

Claims

1. A control method for a high torque density robot joint module, characterized in that, include: Obtain three-dimensional image information of the waste to be compressed; Based on the three-dimensional image information of the waste to be compressed, a three-dimensional distribution map of the compressibility resistance of the waste to be compressed is determined; Based on the three-dimensional distribution map of the compressibility resistance of the waste material to be compressed, information on multiple key fixed points and multiple auxiliary fixed points is determined. Multiple test fixing schemes are generated based on the three-dimensional distribution map of the compressibility resistance of the waste material to be compressed, the information of the multiple key fixing points, and the information of the multiple auxiliary fixing points. The generation of multiple test fixing schemes based on the three-dimensional distribution map of the compressibility resistance of the waste material to be compressed, the information of the multiple key fixing points, and the information of the multiple auxiliary fixing points includes: Construct a map of the waste to be compressed, which includes multiple nodes and edges between the nodes. The multiple nodes include multiple key fixed nodes and multiple auxiliary fixed nodes. The node features of the key fixed nodes are key fixed point information and a three-dimensional distribution map of the compressibility resistance of the waste to be compressed. The node features of the auxiliary fixed nodes are auxiliary fixed point information. Multiple test fixation schemes are generated by processing the graph of the waste to be compressed based on graph neural networks; Based on the multiple test fixing schemes, the high torque density robot joint module is controlled to fix and shake the waste to be compressed, and the shaking video of the waste to be compressed after fixing by each test fixing scheme is obtained. The initial fixing scheme and preliminary compression depth are determined based on the shaking video of the waste material to be compressed after fixing each set of test fixing schemes. This determination includes: Based on the shaking video of the waste to be compressed after each set of test fixing schemes is fixed, a reasonable fixing point determination model is used to determine multiple reasonable fixing point information for each set of test fixing schemes. The reasonable fixing point determination model is a recurrent neural network model. Multiple clusters are obtained by clustering based on the information of multiple reasonable fixed points for each set of test fixed schemes; Based on the aforementioned clusters, an initial fixation scheme and preliminary compression depth are determined. Based on the initial fixing scheme, the high torque density robot joint module for controlling the initial compression depth fixes the waste to be compressed, and obtains the initial compression video of the compressor on the waste to be compressed at the initial compression depth. Based on the initial fixing scheme and the initial compression video of the compressor on the waste to be compressed at the initial compression depth, the adjustment fixing scheme is determined. This determination includes: using an adjustment scheme determination model to determine the adjustment fixing scheme. The adjustment scheme determination model is a Transformer model. The input to the adjustment scheme determination model is the initial fixing scheme and the initial compression video of the compressor on the waste to be compressed at the initial compression depth. The output of the adjustment scheme determination model is the adjustment fixing scheme. The adjustment fixing scheme is an optimized robotic arm fixing operation strategy generated by correcting deficiencies in the initial compression process through the adjustment scheme determination model. The adjustment fixing scheme includes corrected fixing point coordinates, adjusted force direction and magnitude, and compensation torque parameters to enhance stability. Based on the aforementioned adjustment and fixing scheme, the high torque density robot joint module is controlled to fix the waste to be compressed and the compressor is controlled to compress it.

2. A control system for a high torque density robot joint module, characterized in that, include: The acquisition module is used to acquire three-dimensional image information of the waste to be compressed; The compression resistance determination module is used to determine the three-dimensional distribution map of the compression resistance of the waste material to be compressed based on the three-dimensional image information of the waste material to be compressed. The fixed point determination module is used to determine multiple key fixed point information and multiple auxiliary fixed point information based on the three-dimensional distribution map of the compressibility resistance of the waste to be compressed; The test plan generation module is used to generate multiple test fixing schemes based on the three-dimensional distribution map of the compressibility resistance of the waste to be compressed, the information of the multiple key fixing points, and the information of the multiple auxiliary fixing points. The test plan generation module is also used for: Construct a map of the waste to be compressed, which includes multiple nodes and edges between the nodes. The multiple nodes include multiple key fixed nodes and multiple auxiliary fixed nodes. The node features of the key fixed nodes are key fixed point information and a three-dimensional distribution map of the compressibility resistance of the waste to be compressed. The node features of the auxiliary fixed nodes are auxiliary fixed point information. Multiple test fixation schemes are generated by processing the graph of the waste to be compressed based on graph neural networks; The shaking test module is used to control the high torque density robot joint module to fix and shake the waste to be compressed based on the multiple test fixing schemes, and to acquire the shaking video of the waste to be compressed after fixing by each test fixing scheme. The initial scheme determination module is used to determine the initial fixation scheme and preliminary compression depth based on the shaking video of the waste to be compressed after each set of test fixation schemes. The initial scheme determination module is also used for: Based on the shaking video of the waste to be compressed after each set of test fixing schemes is fixed, a reasonable fixing point determination model is used to determine multiple reasonable fixing point information for each set of test fixing schemes. The reasonable fixing point determination model is a recurrent neural network model. Multiple clusters are obtained by clustering based on the information of multiple reasonable fixed points for each set of test fixed schemes; Based on the aforementioned clusters, an initial fixation scheme and preliminary compression depth are determined. The initial compression module is used to fix the high torque density robot joint module to be compressed based on the initial fixing scheme and the initial compression depth, and to acquire the initial compression video of the compressor on the waste to be compressed at the initial compression depth. The adjustment scheme determination module is used to determine an adjustment and fixing scheme based on the initial fixing scheme and the initial compression video of the compressor on the waste to be compressed at the initial compression depth. This determination includes: using an adjustment scheme determination model (a Transformer model) to determine the adjustment and fixing scheme; the input to the adjustment scheme determination model is the initial fixing scheme and the initial compression video of the compressor on the waste to be compressed at the initial compression depth; the output of the adjustment scheme determination model is the adjustment and fixing scheme. The adjustment and fixing scheme is an optimized robotic arm fixing operation strategy generated by correcting deficiencies in the initial compression process through the adjustment scheme determination model. The adjustment and fixing scheme includes corrected fixed point coordinates, adjusted force direction and magnitude, and compensation torque parameters to enhance stability. The compression execution module is used to control the high torque density robot joint module to fix the waste to be compressed based on the adjustment and fixing scheme and to control the compressor to compress it.

3. An electronic device, characterized in that, include: processor; Memory; And a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement the control method for the high torque density robot joint module as claimed in claim 1.

4. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the control method for the high torque density robot joint module as described in claim 1.

Images