26 results about "Robot action" patented technology

This invention provides a welding line detection system that improves the efficiency of teaching welding robot actions and enhances welding quality. It comprises: an imaging unit (211) for capturing an image of the welding object; a coordinate system setting unit (212) for setting a user coordinate system based on markers contained in the captured image; a point group data drawing unit (213) for detecting specific positions of markers in the image, setting the detected specific positions onto point group data obtained by a distance measurement sensor measuring the distance to the welding object, and drawing the point group data, assigned coordinates to the user coordinate system with the set specific positions as the origin, onto the user coordinate system; and a welding line detection unit (214) for detecting the welding line of the welding object based on the point group data drawn on the user coordinate system.

The application provides a robot action control method, system, computer device and storage medium. The robot action control method comprises the following steps: extracting periodic action features from a historical action feature sequence, performing self-attention processing based on the periodic action features to obtain a self-attention processing result; extracting periodic image-text fusion features from image-text fusion features, performing cross-attention processing based on the periodic image-text fusion features and the periodic action features to obtain a first cross-attention processing result; extracting periodic splicing features from splicing features, performing cross-attention processing based on the periodic splicing features and the periodic action features to obtain a second cross-attention processing result; and outputting a robot execution action according to the three attention processing results. The application improves the modeling capability of the model for periodic operation actions, enhances the utilization efficiency of action data, and improves the accuracy, continuity and environmental adaptability of action generation of the robot in a complex scene.

The application provides a robot action control method and device, a medium, a control equipment and a robot, and relates to the technical field of control. The method comprises the following steps: obtaining a control voice signal input by a user for a target robot; performing voice recognition on the control voice signal to obtain voice text corresponding to the control voice signal; performing structured semantic analysis on the voice text by using a preset large language model to obtain structured semantic information, wherein the structured semantic information comprises an action semantic unit; obtaining corresponding target action parameters by using a preset mapping table from semantic units to action parameters according to the action semantic unit; generating an action control instruction for the target robot according to the target action parameters; and the action control instruction is used for action control of the target robot. The application improves the accuracy of robot control and avoids safety problems of the robot.

This disclosure provides a robot intelligent visualization method, device, storage medium, and program product. The method includes: acquiring robot pose data, joint data, and point cloud data of a real-world scene; using a coordinate transformation method to convert the position data in the pose and joint data from the robot's body coordinate system to a three-dimensional scene coordinate system; based on the coordinate-transformed pose and joint data, performing real-time posture and skeletal synchronization of the robot to generate a virtual robot model that accurately reproduces the robot's real movements; constructing a three-dimensional scene based on the point cloud data and embedding the virtual robot model into the three-dimensional scene; and triggering a corresponding task mode when the virtual robot model is within the task point sensing range of the three-dimensional scene. This method achieves precise spatiotemporal synchronization and fusion between the physical robot and the virtual scene through multi-source data fusion and coordinate system unification, realistically reproducing robot movements and intelligently triggering tasks.

The application provides a binocular structured light micro-target positioning recovery method for an intelligent robot, and belongs to the field of positioning measurement of robot visual perception. The method solves the problem of invalidation of an existing self-supervised attention mechanism on micro-targets. The method comprises an inertial measurement component, a mechanical arm, a multi-axis operation end provided on the mechanical arm, a binocular structured light depth camera provided on the robot body and a control calculation module. The control calculation module is used for completing robot action control and image depth data processing. A self-supervised depth refinement network is embedded in the control calculation module. The binocular structured light depth camera and the inertial measurement component are used for collecting an RGB image and an original depth image of a target. The self-supervised depth refinement network is used for solving fusion to generate a cooperative control instruction of the mechanical arm and the multi-axis operation end, so that the mechanical arm and the multi-axis operation end complete targeted operation. The method is mainly used for industrial precision operation and outdoor service robots such as ocean beach garbage cleaning.

A method and apparatus are disclosed for a clutter tidying robot utilizing floor segmentation for its mapping and navigation system, whereby a perception module and navigation module transform lidar and image data from lidar sensors and cameras of a robot sensing system using segmentation and pseudo-laserscan or point cloud transformations to generate global and local maps. The robot pose and maps are transmitted to a robot brain that directs an action module to produce robot action commands controlling the operation of a clutter tidying robot using the pose and map data. In this manner multi-stage planning and sophisticated obstacle avoidance techniques may be incorporated into autonomous robot operations.

The application provides a robot action sequence generation method, a generation device, equipment and a medium. Real-time multi-modal observation data and a current joint state are input into a pre-trained action prediction model to generate an initial action prediction sequence containing multiple future time steps; a spatiotemporal heterogeneous guide mask matrix is constructed using a current end line speed and an execution state of an end operating component in a target robot; the initial action prediction sequence is corrected using the spatiotemporal heterogeneous guide mask matrix and the historical action sequence queue to obtain a target action sequence, and the target action sequence is sent to a controller of the target robot. Through the method and device, the problems of mechanical arm shaking, end effector non-deterministic state switching and the like in the trajectory splicing process of the traditional method are effectively solved, and the stability and response accuracy of the target robot control are improved.

The application discloses an AI robot dialogue understanding method based on causal reasoning, comprising the following steps: obtaining user dialogue text, and splitting semantic roles, event elements, emotional expressions and key sentences into semantic microparticles; establishing a semantic charge conservation constraint in a semantic microparticle set, and adaptively adjusting microparticle charges; generating causal bias change information for the semantic microparticles, and updating the moving trend of the microparticles in the semantic space; constructing a semantic microparticle density field, identifying a semantic area with a density exceeding a threshold, and determining the semantic meaning corresponding to the area as the user intent of the current dialogue; and generating a robot action instruction or a natural language reply matched with the user intent and the moving trend of the semantic microparticles in the area. The application realizes fine-grained semantic tracking and stable robot response generation in multi-round dialogue by splitting dialogue text into semantic microparticles and applying a semantic charge conservation and causal bias driving mechanism.

The application provides a robot action control method and device and electronic equipment, the action control method comprises: encoding natural language operation instructions and image data to obtain image feature sequence, target embedding vector and text embedding vector; performing dot product operation based on embedding retrieval on the image feature sequence and the target embedding vector to obtain a similarity score matrix, and performing mask reconstruction on the image feature sequence to obtain a fine-grained perception feature sequence; performing feature vector linear modulation aggregation on the image feature sequence and the text embedding vector, and performing pruning routing on the fine-grained perception feature sequence to obtain a target visual feature sequence; based on the target visual feature sequence, outputting an action trajectory sequence, and controlling a target robot to perform a corresponding action by an action expert layer. Through the above method, the accuracy of image representation and the processing capability of complex graphics are improved, and the logical close coupling of the action trajectory sequence and the operation instructions is ensured.

The application discloses a robot operation method and system based on task perception virtual visual angle re-rendering, and the method comprises the steps of acquiring robot working scene multi-modal data, and constructing a global scene point cloud; predicting the position of an end effector at the next moment through a rough positioning model, and intercepting a local point cloud; inputting the local point cloud into a multi-view exploration strategy network model to obtain a virtual exploration visual angle; re-rendering the global point cloud according to the virtual visual angle to obtain a 2D observation image containing complete visual information; inputting the re-rendered image and a natural language instruction into a task perception hybrid expert model to extract task-specific fine visual features; and inputting the features into a self-recurrent action strategy network to predict a robot action sequence and complete a task operation. Through multi-view exploration and re-rendering, the application acquires complete visual information of a key target, solves the problems of occlusion and limited field of view under a fixed visual angle, and improves the operation success rate.

This disclosure provides a model training method, action generation method, apparatus, and electronic device. During the model training phase, by introducing a causal-based attention mask in the diffusion transformer, access to future visual latent variables by action sequence latent variables is restricted. However, future visual latent variables can access action sequence latent variables, ensuring that the action sequence prediction branch is independent of future visual information during training. This forces the model to learn the causal relationship between actions and environmental changes based on multimodal data. As a result, future visual latent variables and related computational paths can be directly removed during the inference phase, achieving decoupling between training and inference. This allows the model to stably output high-quality action sequences without relying on predicting future visual images during the inference phase, fundamentally solving the problems of high inference costs and severe error accumulation, significantly reducing inference latency, and improving model inference efficiency and robot motion control accuracy.

Systems and methods for immutable robotic procedure outcome provenance and reimbursement binding are disclosed. A hardware-isolated provenance engine cryptographically binds robotic actions to clinical outcomes and authorization context. Verified provenance is deterministically linked to reimbursement, liability attribution, and regulatory compliance.

The application relates to the technical field of intelligent robots and discloses a patrol robot voice interaction system combining multi-round dialogue context modeling, which comprises the following units: a voice and multi-modal perception access unit, a natural language understanding unit, a multi-modal dialogue state tracking unit, a collaborative dialogue strategy learning unit, a natural language and robot action collaborative generation unit used for synchronously generating a natural language reply and a corresponding robot control instruction sequence according to a robot collaborative strategy instruction, a cross-dialogue continuous learning and strategy evolution unit used for executing the robot control instruction sequence and taking an execution result and real-time multi-modal perception data flow as feedback to update a unified dialogue context model. The application realizes accurate and real-time analysis of the fuzzy reference in a medical scene by constructing a medical adaptive physical word slot and a multi-dimensional alignment fusion mechanism, so that the robot can accurately understand the natural language closely related to a clinical scene.

This invention discloses a deep learning-based method for recognizing robot action intentions, comprising the following steps: collecting multimodal perception data of the robot during task execution; performing semantic analysis on task context information and voice command data; constructing a semantically guided differentiable neural computer model, inputting the multimodal input feature vector and the memory vector read in the previous round into the controller submodule; updating the short-term memory matrix and long-term memory matrix in the write controller submodule; reading content from the short-term memory slot and long-term memory slot in the read controller submodule; inputting the control signal vector, read vector, and task semantic label vector into the decoding network; dynamically adjusting the semantic priority threshold in the write controller submodule and updating the utilization of the memory slots. This invention employs a semantically guided differentiable neural computer to achieve high-precision recognition of robot action intentions.

Provided is a real-time control method for robot manipulation actions based on collaboration between a large model and a small model. The method includes combining environmental data with instruction text data to form multi-modal data, and performing preprocessing on the multi-modal data to obtain preprocessed multi-modal data; performing data encoding on the preprocessed multi-modal data to obtain a feature vector; aligning the preprocessed multi-modal data using cross-modal token alignment technology to obtain a feature representation; training a neural network model using the feature vector and the feature representation to obtain a trained large model; performing a pruning operation, a distillation operation, and a quantization operation on the trained large model to generate a small model; deploying the small model to an edge computing device to obtain robot action planning; and performing real-time control of robot manipulation actions using robot action planning and feedback signals from a plurality of sensors.

The embodiment of the application provides a robot control method and system, a robot and a storage medium, and relates to the fields of artificial intelligence and robot technology. The method comprises the following steps: acquiring a historical space-time scene graph; acquiring an original image information by image acquisition on a target scene through a sub-agent; performing semantic analysis on the original image information to obtain a current scene description; comparing and judging a robot action execution rule, the current scene description and the historical space-time scene graph through a brain agent to obtain a scene comparison result; and in response to the fact that the scene comparison result represents that the current scene description is different from the historical space-time scene graph and the difference conforms to the robot action execution rule, distributing a task to the sub-agent through the brain agent to enable the sub-agent to execute the task. The embodiment of the application can improve the dynamic decision-making capability of the robot control method.

This invention belongs to the field of robot control, specifically relating to an adaptive power allocation method for biomimetic skin heating in humanoid robots. It aims to address the problem of inadequate heating control for humanoid robots performing complex dynamic tasks. The method includes: acquiring target data; inputting the target data into a pre-constructed digital twin model to predict the thermal state change trend of each skin region within a preset future period, and identifying key areas for priority heating based on the predicted motion state; constructing and solving a multi-objective optimization model using the heating power allocated to each skin region as a decision variable; and determining the heating control command for each skin region based on the obtained power allocation value. This invention utilizes a digital twin model and predictive motion state to anticipate thermal state trends caused by future robot actions and environmental changes, effectively improving the heating control performance of robots under complex working conditions.

The application provides a robot action control method and device, electronic equipment and computer readable storage medium; the method comprises the following steps: through a pre-trained learning strategy network, action reasoning is carried out based on the state observation value of the robot at the current time, and the action mean value of the robot at the current time is obtained; a preset gating function is called to determine the gating value at the current time; based on the gating value, the action mean value and the historical action parameters determined by the action prediction model at the first time are weighted and fused to obtain the sampling mean value at the current time; the sampling mean value is used as the sampling baseline of the action prediction model, the action parameter prediction is carried out based on the state observation value through the action prediction model, and the target action parameter at the current time is obtained; and the robot is controlled to perform the target action according to the target action parameter. Through the application, the accuracy of robot action control can be improved.

This invention discloses a robot control method based on large model-driven and multimodal data fusion. The method specifically includes: a user inputting commands via natural language; the robot acquiring scene information of the current environment through visual sensors; deep enhancement of the user commands using the large model Deepseek-V3, combined with visual scene information for fusion; inputting the fused commands into CLIP and T5 text encoders to extract global and local text features, which are then fused using MMDiT for multimodal processing; optimizing the fused features using Mamba L*Blocks to finally generate robot action commands, and generating specific action plans to execute the task through a DiT layer. This invention, by deeply fusing visual and text data and combining large models for command enhancement and scene understanding, enables more accurate and flexible task execution, overcoming the limitations of traditional methods in complex environments and dynamic tasks, and possessing significant intelligent and adaptive value.

The utility model discloses a ductile cast iron pipeline timbering robot timber placing rack, including vertical rod, crosspiece, web and timber, vertical rod sets up vertically, and adjacent vertical rod is fixedly connected through web, and crosspiece sets up horizontally, and one end of crosspiece is fixedly arranged on vertical rod, and at least two crosspieces of same height are provided, and the timber is set up on each crosspiece of same height to place in horizontal state, and the timber of same height is linearly arranged along the extension direction of crosspiece, and two ends of timber are limit end and free end respectively, and the limit end is the end part of setting up triangular block, and the limit end and free end of adjacent timber are staggered arrangement, the utility model discloses firm structure, makes fast, and the setting of timber on the utility model can greatly simplify robot action, reduce the programming difficulty of robot, and can effectively reduce the occurrence of robot failure, greatly improve robot timber grabbing rate, reduce equipment waiting time, thereby improve production efficiency.

The application belongs to the technical field of robots, and discloses a method and system for controlling embodied agents by using a reward function based on optimized reinforcement learning. An interaction trajectory is formed by using the state and control instruction of the embodied agent, a plurality of random interaction trajectories are obtained by randomly sampling the interaction trajectories of a plurality of gaits of the embodied robot, the gradient of the reward function is calculated by using the plurality of random interaction trajectories, and the reward value of the reward function is updated by using the gradient; the updated reward value and the current state of the embodied agent are input into a policy network of robot actions to obtain the current control instruction of the robot. Through the application, the problem of low control precision of the agent caused by the fact that the reward function cannot be automatically updated is solved.

This application provides a method for generating robot actions and a robot, relating to the field of robot control. The method includes: in response to an operation task instruction, acquiring point cloud data of the robot's current working scene and the robot's state features; scoring the importance of each point in the point cloud data to obtain an importance score for each point; encoding pairwise geometric relationships in the point cloud data to obtain the geometric relationship between two points; determining the attention weight of each point based on its importance score and the geometric relationship between the two points; performing a weighted transformation on the point cloud data based on the attention weights to generate enhanced point cloud data; and generating an action instruction corresponding to the operation task instruction based on the enhanced point cloud data and the robot's state features to control the robot to execute the corresponding action instruction. The method provided in this application can improve the accuracy of the robot in performing operation tasks, thereby increasing the success rate of the operation task execution.

This invention discloses a machine grasping method based on scene understanding and intelligent grasping. It generates a human grasping image from an input RGB image and user prompts using a basic generative model. The generated human grasping image is then transformed into specific robot actions. The robot, based on a grasping control network (GCN), controls its manipulator to perform grasping actions according to the target posture map and stress distribution map, and uses force sensors for real-time stress control. Visual and tactile sensors determine whether the manipulator's grasping is successful, and the weights of the GCN are adaptively adjusted based on the results, achieving adaptive learning and optimization of the model. This method, while completing the core grasping task, solves secondary problems such as strategy reliability, environmental adaptability, and grasping precision, and significantly reduces the deployment cost and cycle time for complex grasping tasks.