Industrial mechanical arm intelligent operation system with intelligent capacity
By introducing an analytical decision-making layer, a motion planning layer and an autonomous learning layer in the industrial robot arm system, combining large-scale pre-training language model and environment perception, the decomposition and path planning of complex tasks are achieved, and the robot arm in the existing technology is solved. The lack of independent operation capabilities in multi-task and complex environments has improved the intelligence and multi-task adaptability of the system.
Patent Information
- Application Number
- CN202411363678.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-28
- Publication Date
- 2024-11-22
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing industrial robotic arm operating system is designed in a single way, and it is impossible to achieve high autonomous operation and intelligence in multi-tasking and complex environments, and it is impossible to truly possess intelligent capabilities, especially in task scenarios other than assembly.
Designed an industrial robotic arm intelligent operating system with embodied intelligent capabilities, including analytical decision-making layer, motion planning layer, ontology execution layer and autonomous learning layer, and pre-trained language models, environment perception, adaptive learning and modules through large-scale pre-training. Designed to realize complex task decomposition, path planning and action optimization, with multi-task adaptability and high-precision control.
It improves the autonomous operation capability and intelligence level of the robotic arm, realizes efficient and accurate operation in various application scenarios, and enhances the system's multi-task adaptability and human-computer interaction friendliness.
Smart Images

Figure CN118990507A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of intelligent operation technology, and in particular to an industrial robot arm intelligent operation system with embodied intelligence capabilities. Background Art
[0002] As one of the core equipment of intelligent manufacturing, industrial robotic arms are increasingly used in a wide range of applications. However, the existing industrial robotic arm operation system has a single design and can only adapt to specific tasks. When faced with multiple tasks and complex environments, its autonomous operation capability and intelligence level are not high.
[0003] The control method of the existing technical solution includes: importing prior information; generating assembly sequence and process; generating the assembly action flow of each part based on the prior information and the assembly sequence and process; disassembling the assembly action flow so that each action flow corresponds to the motion control of one or more robotic arms; and uniformly using the state machine model to control the motion of each robotic arm. At the same time, the underlying design of multiple robots is unified into a path planning, motion, and operation state machine model, and a blockchain-style message sharing mechanism is used to synchronize their respective state information. Finally, an embodied intelligence framework is used for information collection and feedback processing. The collected information includes three-dimensional point cloud data, sensor data, robot feedback status data, and system history record data, and finally completes the intelligent assembly decision. However, the above technical solution only targets the task of assembly, and cannot truly possess embodied intelligence capabilities to achieve intelligent planning of multiple tasks. Summary of the invention
[0004] In order to solve the problems existing in the above-mentioned prior art, the purpose of the present invention is to propose an industrial robot arm intelligent operation system with embodied intelligence capabilities, which can understand human instructions, autonomously identify the environment, plan paths, perform diversified tasks, and demonstrate excellent performance in various application scenarios such as scanning measurement, handling, processing, and assembly.
[0005] To achieve the above object, the present invention provides the following solutions:
[0006] An industrial robot arm intelligent operation system with embodied intelligence capability, characterized by comprising:
[0007] The analysis and decision layer is used to decompose complex tasks into basic action primitive sequences, generate execution plans based on the basic action primitive sequences, and optimize the execution plans based on environmental constraints and resource limitations;
[0008] The motion planning layer is used to convert the optimized execution plan into robot arm actions;
[0009] The main body execution layer is used to plan the movement trajectory of the robot arm through the robot arm action, control the movement of the robot arm, and complete the operation;
[0010] The autonomous learning layer is used to optimize the action strategy online during the process of controlling the movement of the robotic arm.
[0011] Optionally, the analysis and decision layer includes:
[0012] A plan generation unit, used to fine-tune the large-scale pre-trained language model, decompose the complex task into a sequence of basic action primitives based on the fine-tuned large-scale pre-trained language model, and generate an execution plan;
[0013] The environmental perception unit is used to integrate various perception information and build a semantic understanding model of the scene as the environmental constraint.
[0014] Optionally, the analysis and decision layer further includes:
[0015] The optimization unit is used to construct a domain-specific knowledge base, make reasoning decisions based on the constructed domain-specific knowledge base, and optimize the execution plan in combination with the environmental constraints and resource limitations.
[0016] Optionally, the motion planning layer includes:
[0017] The motion planning unit is used to decompose the optimized execution plan into multiple functional modules, convert the multiple functional modules into robot arm movements, and optimize the control parameters of the robot arm movement in the main body execution layer through adaptive learning.
[0018] Optionally, the motion planning layer further includes:
[0019] A motion control unit is used to control the joints of the robot arm using a control algorithm and to perform smooth control of the joints of the robot arm through impedance control;
[0020] The path planning unit is used to plan the path of the intelligent controller in the main body execution layer and optimize the trajectory in combination with the dynamic model.
[0021] Optionally, the motion planning layer further includes:
[0022] The data processing unit is used to process the sensor data in the ontology execution layer, and to filter and denoise the data to improve the data quality, and provide perception information for the analysis and decision-making layer.
[0023] Optionally, the ontology execution layer includes:
[0024] The robot arm adopts modular design and can be flexibly configured according to different planning requirements;
[0025] Sensors, used to collect environmental information and characteristic data of target objects;
[0026] The end effector is used to select different plans and complete the work;
[0027] The intelligent controller is used to plan the motion trajectory of the robotic arm through the motion of the robotic arm, and control the motion of the robotic arm in combination with the human-computer interaction data.
[0028] Optionally, the ontology execution layer further includes:
[0029] The human-computer interaction interface is used for interactive operations based on multimodal interaction methods, and provides a visual interface to facilitate manual intervention and supervision.
[0030] Optionally, the autonomous learning layer includes:
[0031] An autonomous learning unit, used for online optimization of the action strategy based on an online reinforcement learning method during the process of controlling the movement of the robotic arm;
[0032] The recording unit is used to record the trajectory and sensor data of each planned execution, and use the large model capability to optimize the decision-making and planning capabilities of the analysis and decision-making layer, motion planning layer, and ontology execution layer.
[0033] The beneficial effects of the present invention are:
[0034] Strong autonomous operation capability: The present invention realizes the autonomous learning and decision-making capabilities of the robotic arm through embodied intelligent technology, thereby improving the autonomous operation capability and intelligence level of the system.
[0035] Good multi-task adaptability: The present invention adopts multi-task planning algorithm and environmental perception and adaptation technology to enable the robotic arm to adapt to the needs of different tasks and environments and achieve efficient and accurate operations.
[0036] High operating accuracy: The present invention uses high-precision sensors and advanced motion control algorithms to achieve accurate recognition and positioning of target objects as well as high-precision operation and control by the robotic arm.
[0037] Friendly human-computer interaction: The present invention realizes convenient interactive operation between the user and the system through the human-computer interaction interface, improving the usability of the system and the user experience. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0039] Figure 1A schematic diagram of an industrial robot arm intelligent operation system with embodied intelligence capabilities according to an embodiment of the present invention;
[0040] Figure 2 This is a schematic diagram of the body execution layer of an embodiment of the present invention; wherein, 1, the robot body, 2, the sensor, 3, the end effector, 4, the intelligent controller, 5, the human-computer interaction interface. DETAILED DESCRIPTION
[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0042] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0043] like Figure 1 As shown, this embodiment discloses an industrial robot arm intelligent operation system with embodied intelligence capabilities, including: an analysis and decision-making layer, which is used to decompose complex tasks into basic action primitive sequences, generate execution plans based on the basic action primitive sequences, and optimize the execution plans based on environmental constraints and resource limitations; a motion planning layer, which is used to convert the optimized execution plan into robot arm movements; a body execution layer, which is used to plan the robot arm motion trajectory through the robot arm movement, control the robot arm movement, and complete the operation; an autonomous learning layer, which is used to optimize the action strategy online during the process of controlling the robot arm movement.
[0044] The analysis and decision-making layer includes: a plan generation unit, which is used to fine-tune the large-scale pre-trained language model, decompose complex tasks into a sequence of basic action primitives based on the fine-tuned large-scale pre-trained language model, and generate an execution plan; an environmental perception unit, which is used to integrate multiple perception information and build a semantic understanding model of the scene as an environmental constraint.
[0045] The analysis and decision-making layer also includes: an optimization unit, which is used to build a specific domain knowledge base, make reasoning decisions based on the built specific domain knowledge base, and optimize the execution plan in combination with environmental constraints and resource restrictions, specifically:
[0046] Analysis and decision-making layer: This layer is responsible for high-level decision-making, task planning, and environmental understanding. First, a large-scale pre-trained language model (such as GPT) is used for fine-tuning to improve the ability to understand specific task areas, thereby realizing the understanding and decomposition of natural language task instructions. Then, by integrating multiple perceptual information such as vision, hearing, and touch, a semantic understanding model of the scene is constructed to achieve comprehensive perception of the environment. In terms of task planning, this layer decomposes complex tasks into a sequence of basic action primitives, and uses methods such as symbolic reasoning to generate execution plans, while considering environmental constraints and resource limitations for optimization. In addition, this layer also builds a knowledge base in a specific field to support reasoning decisions, realize experience accumulation and knowledge transfer, and also uses reinforcement learning to continuously optimize decision-making strategies through interaction with the environment.
[0047] Fine-tuning using large-scale pre-trained language models (such as GPT) includes: selecting a suitable pre-trained large language model as the basis, such as the GPT series, BERT, etc. Collecting professional data sets related to industrial robot arm tasks, including descriptions of various complex tasks, basic action sequences, execution plans, etc. Adopting a partial fine-tuning (Repurposing) strategy, only updating the top layer or a few layers of the model, keeping the underlying parameters unchanged. This can adapt the model to specific industrial tasks while retaining the general knowledge of the model; using low-rank adaptation methods (such as LoRA) to reduce computing resource consumption. This method only fine-tunes a small part of the parameters of the weight matrix, which can ensure model flexibility and improve efficiency. Set appropriate hyperparameters such as learning rate, batch size, number of training rounds, and use the validation set for evaluation and adjustment during the training process. For specific tasks of industrial robot arms, such as decomposing complex tasks into basic action primitive sequences, focus on optimizing this ability of the model. In the fine-tuning process, the scene semantic understanding model provided by the environmental perception unit is combined to improve the adaptability of the model to the industrial environment. The model performance is evaluated regularly using the test set to ensure that the fine-tuned model can accurately decompose complex tasks into basic action sequences suitable for the robotic arm to perform.
[0048] The domain-specific knowledge base includes: Task-related knowledge: rules and methods for decomposing complex tasks into sequences of basic action primitives. This knowledge helps to generate and optimize execution plans. Environmental constraint information: various constraints on the working environment, such as space constraints, obstacle locations, boundaries of the working area, etc. This information is used to optimize the motion path and task execution strategy of the robot arm. Resource constraint information: the capacity limitations of the robot arm (such as maximum load, movement speed, accuracy, etc.) and limitations of other resources (such as electricity, time). Domain-specific rules and standards: operating standards, quality control requirements, and safety regulations in specific fields (such as automotive manufacturing, electronic assembly, etc.). Historical data and experience: records of past task execution, case analysis of successes and failures, evaluation of the effectiveness of optimization strategies, etc. These data can help the system make better decisions in future tasks.
[0049] Resource limitation information is stored in a domain-specific knowledge base. These limitations include, but are not limited to: the capacity limitations of the robot (such as maximum load, movement speed, accuracy, etc.); other resource limitations (such as power consumption, operation time, etc.); system hardware and software performance limitations, etc.
[0050] The optimization process of optimizing the execution plan is as follows: Plan analysis: Analyze the initial execution plan and identify the parts that may exceed the resource constraints. Constraint application: Apply resource constraints as constraints to the execution plan. Adjustment strategy: Including, adjusting the action sequence to more effectively utilize available resources; rescheduling the order of task execution to balance resource usage; splitting complex tasks to adapt to resource constraints. Performance prediction: Use simulation or prediction models to evaluate the performance of the adjusted plan. Iterative optimization: Perform multiple iterations based on the prediction results, and continuously adjust the plan until the resource constraints are met and the optimal performance is achieved.
[0051] The motion planning layer includes: a motion planning unit, which is used to decompose the optimized execution plan into multiple functional modules, convert the multiple functional modules into robot arm actions, and optimize the control parameters of the robot arm motion in the body execution layer through adaptive learning.
[0052] The motion planning layer also includes: a motion control unit, which is used to control the joints of the robot arm using a control algorithm, and to perform smooth control of the joints of the robot arm through impedance control; a path planning unit, which is used to plan the path of the intelligent controller in the body execution layer, and to optimize the trajectory in combination with the dynamic model.
[0053] The use of control algorithms to control the joints of the robot arm includes: the motion control unit and the intelligent controller of the body execution layer work together to control the joints of the robot arm. The responsibilities and functions of the two are different, but the ultimate goal is to achieve precise control of the robot arm. The motion control unit is located in the motion planning layer and is responsible for using control algorithms to control the joints of the robot arm. Its main functions include: 1) Control algorithm application: Use advanced control algorithms (such as impedance control) to accurately control the joints of the robot arm to ensure the smoothness and stability of the robot arm's movements. 2) Smooth processing: Through impedance control and other methods, the control of the joints of the robot arm is smoothed to adapt to different working environments and task requirements. The intelligent controller of the body execution layer is responsible for the specific execution of the robot arm's motion instructions. Its main functions include: 1) Motion trajectory planning: According to the instructions provided by the motion planning layer, the motion trajectory of the robot arm is planned. 2) Real-time control: Combined with human-computer interaction data, the movement of the robot arm is controlled in real time to ensure the accurate execution of the task. The motion control unit decomposes the optimized execution plan into specific robot arm movements and controls the joints of the robot arm through the control algorithm. The intelligent controller at the execution layer of the main body receives instructions from the motion control unit, plans and executes specific motion trajectories, and ensures that the robotic arm completes the task according to the predetermined path and method.
[0054] The motion planning layer also includes: a data processing unit, which is used to process the sensor data in the body execution layer, and filter and denoise to improve the data quality, and provide perception information for the analysis and decision-making layer, specifically:
[0055] The processing of sensor data in the ontology execution layer includes: the data processing unit is responsible for processing the sensor data in the ontology execution layer, which mainly includes the following steps:
[0056] Data preprocessing: 1) Remove outliers: Detect and remove outliers caused by sensor failure or environmental interference to ensure data accuracy. 2) Fill missing values: Use interpolation methods or model-based methods to fill missing data to ensure data integrity. 3) Filtering: Remove noise or uninteresting frequency components through low-pass, high-pass or band-pass filters to improve data quality. 4) Calibration: Correct the sensor output according to a known standard to improve measurement accuracy.
[0057] Feature extraction: 1) Time domain features: Extract time domain features such as mean, variance, standard deviation, etc. to reflect the basic statistical characteristics of the data. 2) Frequency domain features: Extract frequency domain features such as power spectrum density and spectrum peak to analyze the frequency components of the data. 3) Wavelet transform: Decompose the signal into sub-signals of different frequencies to extract local features. 4) Principal component analysis: Reduce high-dimensional data to low-dimensional space and retain the most representative features.
[0058] Data fusion: 1) Weighted average: Different weights are given according to the accuracy and reliability of the sensor to obtain the final result. 2) Model fusion: The data of different sensors are input into the model for joint estimation. 3) Decision-level fusion: The decision results of different sensors are integrated to obtain the final decision.
[0059] Motion planning layer: This layer is responsible for the underlying control, motion planning, and sensor data processing. First, this layer implements precise control of each joint of the robot arm, using advanced control algorithms such as adaptive control and robust control to ensure the stability and accuracy of the robot arm, and achieves smooth operation through impedance control. In terms of motion planning, planning algorithms such as RRT* are used for path planning, and trajectory optimization is combined with dynamic models to ensure that the robot arm can perform collision detection and obstacle avoidance in real time when performing tasks. Sensor data processing is another important function of this layer. It processes data from multiple sensors such as vision, force / tactile, improves perception accuracy through sensor fusion, and performs filtering and denoising to improve data quality. In addition, this layer adopts feedforward processing, that is, the signal is processed unidirectionally from input to output, to achieve a fast and clear response to specific information. Its modular structure decomposes complex functions into multiple functional modules, realizing efficient collaboration and information transmission between modules. At the same time, through adaptive learning, it can continuously optimize control parameters to adapt to different working environments and task requirements. The motion planning layer plays a key role in connecting the analysis and decision-making layer and the body execution layer. It can convert high-level instructions from the analysis and decision-making layer into precise robotic arm movements, while also processing and integrating various sensor data from the body execution layer to provide accurate environmental and status information for the analysis and decision-making layer.
[0060] The ontology execution layer includes:
[0061] The robot arm 1 is used for adopting modular design and can be flexibly configured according to different planning requirements;
[0062] Sensor 2, used to collect environmental information and characteristic data of target objects;
[0063] The end effector 3 is used to select different plans and complete the work;
[0064] The intelligent controller 4 is used to plan the motion trajectory of the robot arm through the robot arm action, and control the movement of the robot arm in combination with the human-computer interaction data.
[0065] Collecting environmental information and characteristic data of target objects includes: Sensors in the ontology execution layer are used to collect environmental information and characteristic data of target objects. These characteristic data play an important role in multiple levels of the system. The following are the application scenarios of each layer: Analysis and decision-making layer: Sensor data is used to build a semantic understanding model of the scene to help the system understand the state and constraints of the current environment, thereby decomposing complex tasks and generating execution plans. Motion planning layer: Sensor data is used to perceive obstacles and dynamic changes in the environment in real time, optimize the motion trajectory of the robot arm, and ensure the accuracy and safety of path planning. Ontology execution layer: Sensor data is used to adjust the motion of the robot arm in real time to ensure the accuracy and flexibility of task execution. For example, sensors can detect the position and state changes of the target object, and the intelligent controller adjusts the action of the robot arm accordingly. Autonomous learning layer: Sensor data records the trajectory and environmental feedback of each planned execution. These data are used for online reinforcement learning to optimize action strategies and decision-making capabilities of each layer of the system.
[0066] The robot arm motion trajectory is planned by the robot arm action, and combined with the human-computer interaction data, the control of the robot arm motion includes: the intelligent controller is used to process sensor data, and this processing complements the data processing of the motion planning layer. The intelligent controller and the motion planning layer work together when processing sensor data: Motion planning layer: responsible for high-level data fusion and path planning, and provides optimized motion instructions. Intelligent controller: at the execution level, it processes sensor data in real time, combines human-computer interaction data, and accurately controls and adjusts the specific movements of the robot arm. This division of labor ensures the flexibility and accuracy of the system. The motion planning layer provides macroscopic path and trajectory planning, while the intelligent controller performs real-time adjustment and control at the micro level. The two cooperate with each other to achieve precise operation and intelligent control of the robot arm.
[0067] The ontology execution layer also includes: a human-computer interaction interface 5, which is used to perform interactive operations based on a multimodal interaction method and provide a visual interface to facilitate manual intervention and supervision, specifically:
[0068] Ontology execution layer: This layer is the actual execution part, responsible for converting the control instructions of the motion planning layer into specific mechanical movements. This layer contains the system hardware and has automatic system calibration functions such as robot hand-eye calibration and sensor center point positioning. Figure 2 As shown, it includes a robot body 1, a sensor 2, an end effector 3, an intelligent controller 4 and a human-computer interaction interface 5.
[0069] The robot body 1 adopts a modular design and can be flexibly configured according to different task requirements; the sensor 2 is used to collect environmental information and characteristic data of the target object in real time; the end effector 3 can be selected for different tasks; the intelligent controller 4 is responsible for understanding human-computer interaction data, processing sensor data, planning the robot arm's motion trajectory, controlling the robot arm's motion, etc.; the human-computer interaction interface 5 is used for the interaction between the user and the system, supports multimodal interaction methods, and provides a visual interface for manual intervention and supervision.
[0070] The autonomous learning layer includes: an autonomous learning unit, which is used to optimize the action strategy online based on the online reinforcement learning method during the process of controlling the movement of the robotic arm; a recording unit, which is used to record the trajectory and sensor data of each planned execution, and use the large model capability to optimize the decision-making and planning capabilities of the analysis decision layer, motion planning layer, and ontology execution layer, specifically:
[0071] Autonomous learning layer: This layer has online learning and self-adaptation capabilities. First, during the task execution, this layer will use online reinforcement learning methods such as PPO to optimize the action strategy online. Second, after the task is executed, this layer will record the trajectory and sensor data of each task execution, use the large model capability to continuously optimize the system's decision-making and planning capabilities, and precipitate the decision-making and planning results of different tasks to achieve rapid adaptation to new environments and tasks.
[0072] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.
Claims
1. An industrial robot arm intelligent operation system with embodied intelligence, characterized in that: include: The analysis and decision layer is used to decompose complex tasks into basic action primitive sequences, generate execution plans based on the basic action primitive sequences, and optimize the execution plans based on environmental constraints and resource limitations; The motion planning layer is used to convert the optimized execution plan into robot arm actions; The main body execution layer is used to plan the movement trajectory of the robot arm through the robot arm action, control the movement of the robot arm, and complete the operation; The autonomous learning layer is used to optimize the action strategy online during the process of controlling the movement of the robotic arm.
2. The industrial robot arm intelligent operation system with embodied intelligence according to claim 1, characterized in that: The analysis and decision-making layer includes: A plan generation unit, used to fine-tune the large-scale pre-trained language model, decompose the complex task into a sequence of basic action primitives based on the fine-tuned large-scale pre-trained language model, and generate an execution plan; The environmental perception unit is used to integrate various perception information and build a semantic understanding model of the scene as the environmental constraint.
3. The industrial robot arm intelligent operation system with embodied intelligence according to claim 2, characterized in that: The analysis and decision-making layer also includes: The optimization unit is used to construct a domain-specific knowledge base, make reasoning decisions based on the constructed domain-specific knowledge base, and optimize the execution plan in combination with the environmental constraints and resource limitations.
4. The industrial robot arm intelligent operation system with embodied intelligence according to claim 1, characterized in that: The motion planning layer includes: The motion planning unit is used to decompose the optimized execution plan into multiple functional modules, convert the multiple functional modules into robot arm movements, and optimize the control parameters of the robot arm movement in the main body execution layer through adaptive learning.
5. The industrial robot arm intelligent operation system with embodied intelligence according to claim 4, characterized in that: The motion planning layer also includes: A motion control unit is used to control the joints of the robot arm using a control algorithm and to perform smooth control of the joints of the robot arm through impedance control; The path planning unit is used to plan the path of the intelligent controller in the main body execution layer and optimize the trajectory in combination with the dynamic model.
6. The industrial robot arm intelligent operation system with embodied intelligence according to claim 5, characterized in that: The motion planning layer also includes: The data processing unit is used to process the sensor data in the ontology execution layer, and to filter and denoise the data to improve the data quality, and provide perception information for the analysis and decision-making layer.
7. The industrial robot arm intelligent operation system with embodied intelligence according to claim 1, characterized in that: The ontology execution layer includes: The robot arm adopts modular design and can be flexibly configured according to different planning requirements; Sensors, used to collect environmental information and characteristic data of target objects; The end effector is used to select different plans and complete the work; The intelligent controller is used to plan the motion trajectory of the robotic arm through the motion of the robotic arm, and control the motion of the robotic arm in combination with the human-computer interaction data.
8. The industrial robot arm intelligent operation system with embodied intelligence according to claim 7, characterized in that: The ontology execution layer also includes: The human-computer interaction interface is used for interactive operations based on multimodal interaction methods, and provides a visual interface to facilitate manual intervention and supervision.
9. The industrial robot arm intelligent operation system with embodied intelligence according to claim 1, characterized in that: The autonomous learning layer includes: An autonomous learning unit, used for online optimization of the action strategy based on an online reinforcement learning method during the process of controlling the movement of the robotic arm; The recording unit is used to record the trajectory and sensor data of each planned execution, and use the large model capability to optimize the decision-making and planning capabilities of the analysis and decision-making layer, motion planning layer, and ontology execution layer.
Citation Information
Patent Citations
Real-time multi-task multi-person human-machine interaction system based on environment perception
CN113829343A
Automatic driving decision planning cooperation method and system
CN117032219A
Robot control method and device
CN117283549A
Part sorting processing method, device and equipment based on body intelligence and medium
CN117772648A
Intelligent strategy generation system and method based on generative AI
CN117993518A
Cited By
Deformable object interactive operation control method based on visual touch-language-action multi-mode model
CN119526422A
Parameter acquisition method of mechanical arm
CN119704198A
Design method of intelligent system with body based on artificial intelligence
CN119795199A
Distributed self-organizing body-building intelligent system and method
CN120103804A
A distributed self-organizing embodied intelligence system and method
CN120103804B
