68 results about "Lie algebra" patented technology

The embodiment of the invention provides an aircraft motion state determination method based on Lie group unscented Kalman filtering, and the method comprises the steps: building a state model of a target aircraft, and a measurement model of the target aircraft; obtaining a final Lie group state estimation value of the target aircraft through iterative calculation according to the state model and the measurement model; and determining the motion state of the target aircraft according to the final Lie group state estimation value. According to the technical scheme, a Lie group unscented Kalman filtering navigation framework based on external measurement information is provided, specific forms of sigma point error propagation on Lie algebra and state updating on the Lie group are given, and a self-adaptive updating method of a state noise covariance matrix and a measurement noise covariance matrix is designed. Therefore, the self-adaptive Lie group unscented Kalman filtering method is provided, and the multi-sensor multi-beacon target state estimation precision is greatly improved by means of the self-adaptive Lie group unscented Kalman filtering method.

The invention discloses a low-overlapping-rate point cloud registration method based on topology-metric decoupling, and belongs to the technical field of computer vision and three-dimensional reconstruction, and the method comprises the steps: scanning an object through a three-dimensional laser scanning collection terminal, collecting point clouds at two different angles as a source point cloud and a target point cloud, and achieving high-robustness registration in a difficult scene with an extremely low overlapping rate; according to the method, through the asymmetric association strategy, the zero solution problem under the low overlapping rate is solved through the one-way matching union set, and the recall rate is remarkably increased; meanwhile, through a DGTG geometric pruning and probability manifold optimization mechanism, a voxel quantization error is eliminated by using anisotropic covariance and Lie algebra iteration, a translation error and a rotation error are greatly reduced, and a precision bottleneck caused by voxelization is effectively broken through. According to the method, the existence of a solution can be ensured under the condition that the overlapping rate is extremely low, voxel quantization errors can be eliminated through manifold optimization based on covariance, and high-precision and high-efficiency registration of low-overlapping-rate point clouds is achieved.

The invention discloses a robot fruit grabbing method based on a multi-mode time sequence collaborative prediction algorithm, and aims to solve the problems that the short-time future trajectory is difficult to accurately predict and the grabbing opportunity is difficult to determine due to fruit and branch swinging caused by wind, branch elasticity and platform advancing. According to the method, a multi-modal time sequence is unified through micro-time alignment, uncertainty is quantified, a three-dimensional special Euclidean group and other variable graph converters are constructed, a spiral shaft constraint projection layer is arranged on the edge between the fruit and a fruit stem, and conditional diffusion short-time trajectory prediction of plum algebra parameterization and uncertainty gating triggering are combined; motion planning and closed-loop control of time delay compensation and collision avoidance constraint are executed in a linkage mode, and the technical effects of high-precision short-time three-dimensional pose prediction, grabbing opportunity self-adaptive triggering and high-success-rate stable grabbing are achieved.

The invention discloses a satellite group collaborative prediction method based on Lie group differential manifold and covariant optimization, and mainly solves the problems of inaccurate disturbance modeling, poor physical interpretation and large prediction error in the prior art. According to the scheme, the method comprises the following steps: mapping the position and attitude of a satellite to an SE (3) manifold, and describing the movement speed of the satellite by using Lie algebra; establishing a multi-level Lie algebra disturbance decomposition mechanism including absolute perturbation, relative coupling and observation noise; constructing a Riemannian loss function by using the orbit state after Lie group manifold embedding and the orbit state deviation obtained by disturbance decomposition; defining an orbital state updating rule by using a covariant derivative of the function, and iteratively solving the orbital state updating rule on the SE (3) manifold to output an orbital state prediction result of the satellite; for an abnormal state of inter-satellite link interruption, a differential geometric connection form is defined, and virtual relative position observation of an integral generation orbit state is carried out. According to the method, the physical interpretability and long-term prediction stability of orbit prediction can be improved, prediction errors are reduced, and the method can be used for low orbit communication constellation deployment of frequency band inter-satellite links.

The invention relates to the technical field of industrial data processing, and provides an industrial internet production scheduling method and system based on multiple modes. Industrial data in the production process are obtained through an intelligent sensor in the industrial internet, modal parameters corresponding to all modals are calculated, and space-time alignment is carried out to generate an industrial tensor; decomposing and combining the industrial tensor through a preset first decomposition mode and a preset second decomposition mode to obtain a data tensor; generating a scheduling strategy based on to-be-optimized parameters screened from a parameter space of production scheduling and a preset Lie algebra base, and constructing a strategy function based on the scheduling strategy and the data tensor; and adjusting a modal weight factor in the strategy function to optimize a scheduling strategy in the strategy function to generate a target strategy, and performing distributed control on each industrial device through the industrial control terminal. The digital twin base constructed through the intelligent sensor ensures the real-time performance of data, the scheduling and production efficiency is improved, and the idle rate of equipment is reduced.

The invention relates to the technical field of computer vision, and discloses a precise medicine delivery method and system, and the method effectively strips illumination artifacts through the construction of a local gradient structure tensor field and an anisotropic screening mechanism, remarkably improves the robustness of a visual front end in a light and dark alternating environment, and improves the accuracy of medicine delivery. Rigid body motion constraint and Lie algebra continuous modeling are utilized, high-precision space-time alignment between heterogeneous sensors is realized on the premise that scene depth does not need to be recovered, hardware delay errors are eliminated, a physical-driven probability weight dynamic allocation mechanism is constructed by introducing a multipath scattering index and a structural information entropy flux, and a dynamic space-time alignment algorithm is established. Environment degradation is sensed in real time, the observation weight is adjusted in a self-adaptive mode, the influence of the non-line-of-sight multipath effect and visual texture missing is effectively restrained, high-precision inertial dead reckoning can still be maintained in a blind area where a sensor is in full failure in combination with momentum prior information generated through Schel complement marginalization operation, and the method has the advantages of being high in precision and high in precision. And the navigation continuity and reliability of the distribution robot in a complex scene are ensured.

The invention discloses a sneaking and crawling integrated navigation method and system for an amphibious robot, and relates to the technical field of robot navigation and control, and the method comprises the steps: extracting environment features through a deep learning model based on preprocessed environment information data, and carrying out the feature fusion through employing a Bayesian network, and generating a comprehensive feature vector; based on the fused comprehensive feature vector, establishing a state transition equation by using a Lie group Lie algebra method, and predicting the position and posture change of the amphibious robot; and dynamically adjusting navigation parameters of the amphibious robot through an AI algorithm based on the position and posture change of the amphibious robot. According to the invention, a Bayesian network is adopted to fuse various environment features, such as angular velocity features, acceleration features, depth features and the like, extracted from various sensors, and a comprehensive feature vector is generated. Through the step, effective integration of multi-source information is realized, and the understanding ability of the system to a complex environment is enhanced.

The invention relates to the technical field of a two-wheeled vehicle system, in particular to a two-wheeled vehicle state estimation method and device based on a Pinochio dynamics library and a Lie group, and the method comprises the steps: carrying out the modeling of a two-wheeled vehicle based on the Lie group and the Lie algebra, and constructing a motion state; the method comprises the following steps: acquiring a plurality of kinetic parameter matrixes of a two-wheeled vehicle in real time by adopting a Pinocochio kinetic library, and constructing a discrete kinetic equation; determining a predicted motion state and a predicted covariance matrix of the two-wheeled vehicle at the next moment; and according to the observation data, carrying out posteriori updating on the predicted motion state and the predicted covariance matrix by adopting a Kalman gain and an observation matrix, and determining a posteriori motion state and a posteriori covariance matrix of the two-wheeled vehicle at the next moment. According to the method provided by the embodiment of the invention, the kinetic parameter matrix is obtained by adopting the Pinocochio kinetic library, so that the manual workload is greatly avoided, the complexity degree in the modeling process is reduced, and the modeling efficiency is remarkably improved while the estimation precision is improved.

The invention discloses a modal modeling method for a tendon-driven continuum robot. The method comprises the following steps: carrying out differential modeling on a configuration under a SE (3) Lie group-se (3) Lie algebra framework; the strain field is subjected to spectrum parameterization and is decomposed into active / passive orthogonal subspaces, and an inner product adopts energy weighted Hilbert measurement containing material and section parameters. The active mode is constructed by a linear continuous Jacobian of a tendon path; legendre polynomial expansion is adopted under the collinear layout, and complex Fourier harmonic expansion is adopted under the non-collinear layout; the passive mode is selected according to a spectrum slot position mutual exclusion principle so as to ensure that the passive mode is orthogonal to the active mode in energy and suppress spectrum leakage. Lie group integration and spectral integration are adopted for numerical value realization; newton-Rafson iteration is used in the quasi-static state, and generalized-alpha time integration or Newmark-beta integration is used in the dynamics. The method has high precision and robustness under nonlinear large deformation and multi-tendon coupling conditions, and is suitable for scenes such as software operation, minimally invasive intervention and precise detection.

The invention discloses a data processing method and system based on Lie algebraic dynamics and an entropy modulation mechanism, and relates to the technical field of artificial intelligence and data processing, and the method comprises the steps: constructing a group action operator on a physical manifold, and carrying out the manifold enhancement of input data; the Hilbert subspace decomposition and decoupling of the feature channel are completed in parallel through a learnable decorrelation kernel and tensor contraction operation by using a matrix parallel elastic orbit layer; mapping the features to a symplectic geometric phase space, and introducing an adaptive symplectic respiration operator and a Hamiltonian kinetic equation to carry out physical conservation evolution; constructing a Lie algebra classifier, and determining a data category by calculating a Lie bracket transpose norm of a feature generator and a category base; and guiding model training by using an entropy modulation action quantity loss function containing physical potential energy. The method can solve the problems of poor generalization and weak anti-noise capability of a traditional deep learning model, and can be widely applied to the fields of power equipment monitoring, computer vision, natural language processing and the like.

The invention provides a double-layer safe trainable quantum machine learning method based on polynomial dynamics Lie algebra, and the method comprises the following steps: 1, constructing core variation simulation meeting the constraint of the polynomial dynamics Lie algebra, and guaranteeing the trainability of a model; 2, performing truncated Chebyshev graph coding at an input end, and constructing a rugged loss function landscape by using graph state entanglement and a Chebyshev tower strategy to prevent a snapshot inversion attack; 3, executing dynamic local scrambling at an output end, applying time-varying random local unitary transformation before measurement, and confusing a linear relation between gradient and a snapshot to prevent recovery attack of the snapshot; and 4, measuring and calculating a loss function, and updating parameters. According to the method, an orthogonal decoupling strategy is adopted, a privacy protection mechanism is externally arranged on an input / output interface, and trainability is anchored to core configuration, so that the capability of resisting algebraic attacks is remarkably improved while model convergence is ensured.

The invention discloses an anti-interference multi-label webpage identification method and system under an encrypted traffic condition, and the method comprises the steps: constructing a dynamic association graph through a burst transmission structure of the encrypted traffic, precisely deconstructing aliasing traffic through an entropy weight punishment mechanism and a node aggregation algorithm, and outputting an independent substream sequence; extracting non-stationary time-frequency features for the substreams, fusing the non-stationary time-frequency features with graph topology features extracted based on a graph structure in a tangent space, generating disturbance samples by using an adversarial training environment, adjusting feature weights, and generating composite fingerprints; executing long time sequence modeling on the Lie group manifold by using a Bi-Mama network, reserving a modulation interface, and outputting a preliminary recognition probability; and auditing causal consistency through a webpage loading logic state machine, converting an abnormal state into geodesic line correction in a Lie algebra tangent space, and correcting network parameters on line through a reserved interface. According to the method, the problems that aliasing traffic is difficult to deconstruct, fingerprint robustness is insufficient and an identification result lacks logic consistency constraint are solved.

The invention discloses a dynamic motion primitive coding method and system based on Lie group Lie algebra, and the method comprises the steps: uniformly representing the tail end pose and contact force spinor of a robot in a Lie group form, and constructing a unified mathematical framework; the method comprises the following steps: acquiring a pose and a force spinor sequence through a teaching phase, and converting the force spinor sequence into a virtual pose sequence through scaling mapping; establishing a dynamic motion primitive (DMPs) model based on the Lie group Lie algebra theory, and learning a model weight matrix and a primary function parameter by adopting local weighted regression; and according to the initial and target poses of the new task, generating a smooth trajectory through phase synchronization control and Euler integration, and outputting the smooth trajectory to a robot controller. According to the method, the problems of posture singularity and motion distortion caused by splitting of a posture channel in a traditional DMPs method are solved, the learning fidelity and reproduction stability of the six-degree-of-freedom posture track of the robot are improved, the smoothness and coordination of the whole motion process are guaranteed, and the debugging complexity of a force control system is reduced.

This invention discloses a pose estimation method for industrial parts that integrates neural implicit representation and MGC-SR anti-reflection constraints. The method first acquires RGB-D images and a 3D CAD model of the part to be assembled; it then extracts multimodal features using a pose prediction model based on neural implicit representation and generates an initial 6 DoF pose of the part in the camera coordinate system; it converts the depth image into an observation point cloud using camera intrinsic parameters and performs coarse alignment with the CAD model point cloud in 3D space; it constructs an anti-reflection weighted model (MGC-SR) based on multi-source geometric confidence, and introduces an adaptive weight allocation strategy based on normal consistency and depth confidence to address noise interference in industrial high-reflectivity scenarios. Using Lie group and Lie algebra theory, it iterative fine-tuning minimizes the point cloud registration error in the tangent space, outputting the final high-precision pose. This invention combines the zero-shot generalization of deep learning with the physical accuracy of 3D geometric registration, effectively solving the problems of pose jitter and insufficient accuracy of industrial metal parts in high-reflectivity, low-texture environments, and has high engineering application value.

The application relates to a three-dimensional continuous shape estimation method and system of a flexible robot, wherein the method combines the arc length of the flexible robot and unknown external loads from the environment with a Cosserat elastic rod model and a Cosserat elastic string model to generate a Lie algebra coupled statics model representing a tendon-driven model, models the external loads as Gaussian process noise to construct a prior probability model, obtains a prior probability distribution of the flexible robot in a Lie algebra space, constructs a measurement noise model in the Lie algebra space according to discrete measurement poses captured by a sensor and measurement noise, obtains a predicted measurement value, fuses the prior probability distribution and the predicted measurement value to obtain a posterior shape estimation of the flexible robot, and calculates a continuous three-dimensional shape estimation of the flexible robot and a first credibility of the continuous three-dimensional shape estimation according to the posterior shape estimation and a Gaussian process interpolation method. Therefore, the application realizes accurate perception of the continuous three-dimensional shape of the flexible robot.

The invention discloses a fire source positioning method and system based on combination of thermal imaging and laser ranging, and relates to the field of fire prevention and control. Acquiring a thermal image, an original image and point cloud data of a target area, identifying a fire source candidate area, constructing a cylindrical coordinate system taking one end of a central axis of the fire source candidate area as an original point, and generating a three-dimensional model containing fire source space attitude and obstacle information; adjusting the shooting angle of the thermal imager in combination with the three-dimensional model, and planning a surrounding path containing an effective detection point location; the mobile carrier is controlled to move along the path, and a three-dimensional space constraint matrix is generated; performing Lie algebraic fusion on the thermal imaging signal features and a three-dimensional space constraint matrix in a symplectic geometric space, dynamically adjusting a laser ranging covariance weight through a confrontation recognition model, and mapping the laser ranging covariance weight into a space coordinate of a fire source in a cylindrical coordinate system; and calculating the Riemannian manifold distance between the positioning coordinate and the pre-path data, and when the Riemannian manifold distance exceeds a threshold value, re-sampling the data, updating the model weight and re-positioning. According to the invention, the fire source positioning precision is greatly improved.

The invention discloses a Lie algebra-based variable component subalgorithm layered training method, which comprises the following steps of: optimizing and freezing ZX on a generation element subset G < 1 > = {XX, YY and XY} through a layered activation-freezing training schedule, remarkably relieving a barren plateau, improving trainability, reducing an instantaneous reachable algebra dimension, and increasing gradient variance and optimizing stability; after the switching criterion is met, ZX is introduced and expanded to G2 = {XX, YY, XY, ZX}, and the final expression ability is maintained and recovered; the switching basis is consistent with the dynamic Li algebra structure, and is jointly triggered by gradient statistics and sub-algebra scale indexes, so that the training process is schedulable, explainable and convenient to audit and reproduce.

The application discloses a Lie algebra-based variable quantum algorithm hierarchical training method, through a hierarchical activation-freezing training schedule, first optimization and freezing of ZX on a generating element set G 1 ={XX, YY, XY}, barren plateaus are significantly alleviated and trainability is improved, instantaneous reachable algebraic dimension is reduced, gradient variance is increased and optimization stability is improved; after meeting switching criteria, ZX is introduced to expand to G2={XX, YY, XY, ZX}, and final expression capability is maintained and restored; switching is consistent with dynamic Lie algebra structure, and is jointly triggered by gradient statistics and subalgebra size indexes, so that the training process is schedulable, interpretable, and convenient for auditing and reproduction.

The invention provides a method for iteratively optimizing a 6D pose of an object through ConvGRU in a complex scene. In a complex scene with serious shielding and insufficient texture information, image information of an object is collected through an industrial camera, and a corresponding rendering image is generated by using PyTorch3D. An observation image and rendered images under multiple postures are used as input, and cross-view geometric consistency image features are constructed by using a weight-shared Resnet network and a bidirectional RNN network. Then, correlation features between the images are coded through correlation volume, and the correlation features and semantic features are input into ConvGRU together for iterative updating; a confidence coefficient weight output by the network is used for dynamically suppressing interference of a low-quality area, an output corresponding field correction amount constructs a re-projection error through a differentiable Perspective-n-Point (PnP) module, a Gaussian-Newton method is introduced into a Lie group space to carry out nonlinear least square solution, a pose increment is calculated in a corresponding Lie algebra space, and the pose of the low-quality area is calculated. And completing iterative updating of the attitude.

The invention discloses an unmanned aerial vehicle live-action three-dimensional modeling method based on LiDAR assistance, and the method is characterized in that the method comprises the following steps: (1) data collection: employing an IMU + GNSS integrated navigation system to collect the pose data of an unmanned aerial vehicle in real time, carrying out the filtering, intensity value normalization and down-sampling processing of a LiDAR point cloud, and obtaining the pose data of the unmanned aerial vehicle; performing multi-view image feature point extraction and feature point descriptor calculation and matching on the image data; (2) adjusting a conversion matrix through a dynamic weight calculation model; and (3) performing iterative optimization by adopting a nonlinear least square method, converting a registration problem into nonlinear optimization, obtaining an initial conversion matrix from a dynamic weight stage, setting Lie algebra parameterized expression, determining an optimization boundary condition, establishing an iterative optimization process, and gradually and finely adjusting a coordinate conversion relation to enable an overall matching error to be minimum and ensure that a registration error is less than or equal to 0.5 cm. The method has the beneficial effects that the offset of the coordinate system is corrected in real time through the dynamic weight registration algorithm, and the registration error precision is improved by more than three times.

The invention discloses a self-adaptive laser mapping method fusing historical poses and point cloud characteristics, and belongs to the technical field of robot positioning and map construction, and the method comprises the steps: reading point cloud data of a laser radar sensor, and carrying out the preprocessing; modeling the change trend of the obtained historical pose as a Lie algebraic vector field, and constructing a dynamic correction guide field in combination with historical error information; classifying the point cloud into planar points and non-planar points; calculating adaptive weights of different classification points; and respectively establishing point-to-point and point-to-plane residual errors according to point cloud classification, solving the pose of the robot through an adaptive weighted Gauss-Newton optimization method, and updating a local map and a historical pose sequence. According to the method, the historical motion trend and real-time deviation correction are fused, and a geometric feature self-adaptive registration strategy is combined, so that the pose estimation precision and robustness in a complex motion mode and a feature degradation scene are remarkably improved, and an effective solution is provided for reliable navigation of a robot in a GPS denial environment.

The invention relates to a lamp ring-free handle positioning method and system based on Lie group state estimation and dynamic offset calibration, and the method comprises the steps: defining the pose state of a handle based on Lie group SE (3), defining an error state in a corresponding Lie algebra se3 space, and building a system kinetic equation based on error state Kalman filtering; representing a multi-source sensor observation model as an observation function on the Lie group; modeling dynamic offset transformation between the handle and the hand into a time-varying random process, and expanding the state of the time-varying random process into a system state vector for online joint estimation; updating and optimizing the system state vector by using a nonlinear optimization method based on multi-sensor observation to obtain an optimized handle positioning result; the system is implemented based on the method. According to the method, the consistency between the positioning output and the actual physical position of the handle is remarkably improved, the pose jumping phenomenon is effectively eliminated, the immersion and continuity of user experience are improved, and the continuous track which is consistent in time and accurate in space is finally output.

The present application relates to photovoltaic material processing detection technical field, specifically disclose a kind of solar cell panel busbar laser build-up welding defect detection method, comprising: after the pretreatment of busbar laser build-up welding image sample is labeled with defect category, to obtain training set;Extract each covariance matrix in training set to constitute Lie algebra training point set;Lie algebra space point in Lie algebra training point set is clustered in Lie algebra space, and the center point of each category Lie algebra space point is obtained;The Euclidean distance between the Lie algebra test point of the busbar laser build-up welding image to be detected respectively to each category Lie algebra center point is calculated, and the category of the shortest Euclidean distance Lie algebra center point belongs is judged as the category of the busbar laser build-up welding image to be detected.The present application further discloses a kind of solar cell panel busbar laser build-up welding defect detection system.The present application can improve the accuracy and learning efficiency of solar cell panel busbar laser build-up welding defect identification.

The invention belongs to the field of robot motion control and trajectory planning, and particularly relates to a quaternion limited trajectory online re-planning method based on a Lie algebra space, which comprises the following steps of: mapping a current quaternion attitude read by an attitude sensor in real time to the Lie algebra space to obtain rotation vector representation; a trajectory planner is adopted and based on a preset kinematics constraint set, a trajectory with continuous time is planned; target attitude update is monitored in real time, and when a new target quaternion attitude is received, online re-planning is carried out based on a current track state so as to ensure the continuity of the attitude, the angular velocity and the angular acceleration at a switching point; restoring the planned trajectory into a quaternion form through inverse mapping; and performing normalization processing on the recovered quaternion, outputting a quaternion trajectory with continuous time and satisfactory constraint to a trajectory actuator, and ending the current planning cycle. The constraint of the preset maximum angular velocity, angular acceleration and jerk is met in the whole process, and overrun and sudden change are avoided.

This invention proposes a Lie group-based hand-eye calibration method based on optimal transmission theory, comprising the following steps: S1, after the robotic arm performs multiple non-coplanar pose transformations, synchronously acquire the pose data of the actuator end effector and the camera after each pose transformation; S2, map the rotation matrix in the acquired pose data to the Lie algebra space to construct discrete probability distributions of the two sets of pose data; S3, calculate the weights of the pose data based on optimal transmission theory according to the discrete probability distributions of the two sets of pose data; S4, solve the rotation and translation matrices in the hand-eye calibration matrix using weighted averages to obtain the hand-eye calibration matrix. This invention utilizes optimal transmission theory to achieve global consistency optimization of calibration parameters, overcoming the limitations of local optimization in traditional least squares methods, ensuring that the solution results of the rotation matrix and translation vector are not affected by abnormal initial values, maintaining extremely high calibration accuracy even under complex motion trajectories, and completely avoiding systematic failures caused by error accumulation.

The invention relates to the field of music artificial intelligence, and discloses an intelligent music generation algorithm dynamically adapting to environment emotions, which comprises the following steps: acquiring emotional state information of a user in a current environment, and constructing an emotional state function changing along with time for describing an emotional change trend of the user in continuous time; inputting the emotional state function into a pre-trained emotional mapping model to obtain a corresponding music control vector sequence, the control vector sequence being defined in a Lie algebra space of music control parameters; the invention also discloses an intelligent music generation system dynamically adapting to the environmental emotion. The system comprises an emotion recognition module; an emotion mapping module; a track embedding module; a music generation module; and a feedback control module. According to the method, the Lie group trajectory modeling and user emotion feedback fine tuning mechanism is introduced, so that the structural continuity of melody generation and dynamic self-adaption of emotion response are realized.

The application provides a Bessel curve posture interpolation trajectory optimization method, system and computer device, a mapping relationship between a posture trajectory in a Cartesian space and a Lie algebra space tangent vector is established, a Bessel curve interpolation trajectory in the Lie algebra space is constructed, and based on a curvature radius change model and a speed change model of the Bessel curve interpolation trajectory, the smoothness and stability of a multi-section posture interpolation trajectory when a rotation axis changes are improved. Furthermore, based on a Jacobian matrix mapping of a mechanical arm joint space and the Cartesian space, a joint space speed constraint is added, and a joint space maximum angular velocity constraint problem of the mechanical arm is considered to further constrain the Cartesian space speed, bidirectional feedback of the joint space and the Cartesian space is realized, the stability of each joint motion of the mechanical arm is improved, the speed curve of each joint of the mechanical arm is continuous, and the precision of the mechanical arm trajectory tracking is improved.

The invention discloses a class-level object grabbing method based on an equivariant grabbing pose field. The method solves the problems that an existing method is low in reasoning speed, poor in pose generalization and the like. The invention provides an equivariant grabbing pose field which comprises four parts. Firstly, a grabbing pose field is constructed, object shape point cloud serves as a condition, class-level shape features are extracted through a neural network, and mapping from any input pose to an object feasible grabbing pose is learned. Secondly, training a grabbing pose field, quantifying a grabbing pose data set of an object by using L1 norm loss, and supervising the difference between a predicted pose and a grabbing pose true value; then, on the basis of reasoning of a grabbing pose field, grabbing pose prediction is used as guidance for updating a grabbing pose estimated value, and interpolation is carried out in the Lie algebra space; and finally, constructing a grabbing pose field to be equal, so that the network can be generalized to any object pose through the property of the construction instead of data enhancement.

The invention provides an intelligent grabbing method based on covariant Hamiltonian optimization, and the method comprises the steps: obtaining the position information and posture information of a target object through an RGB-D camera installed at the tail end of a robot, optimizing a grabbing path in a robot configuration space based on a covariant Hamiltonian optimization algorithm, constructing an optimization objective function, and obtaining a target object through the optimization objective function. Iterative solution is carried out based on gradient information, and an optimal grabbing path is obtained; in combination with a target position, a rotation matrix of an end effector of the robot is optimized, and based on calculation torque control and impedance control strategies, cooperative control over the position and force of the robot in the grabbing process is achieved, the contact force and the motion path of the end effector of the robot are adjusted in real time, and the grabbing stability and safety are ensured. According to the method, by combining covariant Hamiltonian optimization and Lie algebra optimization, autonomous grabbing of the robot in a complex environment is achieved, and the calculation efficiency of path planning, the stability of the grabbing posture and the adaptability of the grabbing process are improved.