Operation injection type intelligent mechanical device and control method thereof

Abstract

The invention is suitable for the technical field of artificial intelligence, and provides an operation injection type intelligent mechanical device and a control method thereof.The device comprises a main body structure which is a physical carrier of the intelligent mechanical device and provides mechanical support and an integrated interface; the sensing module is used for collecting environment information and a user instruction and outputting the collected information as sensing data; the AI processing module is used for receiving the sensing data and the operation injection information of the communication module and performing user intention recognition based on the sensing data and the operation injection information; the decision control module is used for performing task planning according to the intention recognition result of the AI processing module and generating an action control instruction for driving the multi-limb execution module; the communication module is used for receiving externally input operation injection information in a wired or wireless mode; and the multi-limb execution module is used for executing physical operation according to the received action control instruction. Complex reprogramming does not need to be carried out on each task change, and intelligence and generalization of the physical interaction capability of the device are achieved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an operation injection-type intelligent mechanical device and its control method. Background Technology

[0002] Currently, the integration of artificial intelligence and physical devices has become an important development trend, giving rise to a series of products represented by smart speakers and service robots. At present, the automation of equipment operation is mainly based on software-based automation scripts, such as macro command programs running on computers or mobile devices, which execute predetermined tasks by simulating events such as mouse clicks and keyboard inputs at the operating system level; another approach is to use specialized robotic arms based on fixed programs, which complete repetitive operations in specific scenarios through pre-programmed trajectories and movements.

[0003] However, software scripts are highly dependent on specific operating systems, application programming interfaces (APIs), and even screen pixel positions, making them unsuitable for generalization to new, unprogrammed devices. Dedicated robotic arms, on the other hand, are typically designed for single tasks and fixed environments; any change in task or environment necessitates complex reprogramming, which is time-consuming and labor-intensive. Furthermore, existing solutions can only passively execute preset processes, unable to understand user natural language commands, let alone perform real-time perception and decision-making based on dynamic environments. This results in weak physical interaction capabilities of existing devices, preventing them from flexibly integrating into diverse daily life scenarios as general-purpose intelligent assistants. Summary of the Invention

[0004] This invention provides an operation injection-type intelligent mechanical device and its control method to solve the problem that existing devices have weak physical interaction capabilities and cannot be flexibly integrated into diverse daily generation scenarios as general-purpose intelligent assistants.

[0005] The first aspect of the present invention provides an injection-type intelligent mechanical device, comprising: Main structure, perception module, AI processing module, decision control module, communication module, and multi-limb execution module; The main structure serves as the physical carrier of the intelligent mechanical device, providing mechanical support and integration interfaces; The sensing module is mounted on the main structure and is communicatively connected to the AI ​​processing module. It is used to collect environmental information and user commands, and output the collected information as sensing data. The AI ​​processing module is located within the main structure and is communicatively connected to the perception module, the communication module, and the decision control module. It is used to receive perception data from the perception module and operation injection information from the communication module, and to perform user intent recognition based on the perception data and operation injection information. The decision control module is located within the main structure and is communicatively connected to the AI ​​processing module and the multi-limb execution module, respectively. It is used to perform task planning based on the intent recognition results of the AI ​​processing module and generate action control commands to drive the multi-limb execution module. The communication module is located within the main structure and is communicatively connected to the AI ​​processing module. It is used to receive operation injection information from external input via wired or wireless means. The multi-limb execution module is mechanically connected to the main structure and communicatively connected to the decision control module, and is used to perform physical operations according to the received motion control commands.

[0006] Furthermore, the multi-limb execution module includes at least one mechanical limb unit, each of which is driven by a micro-actuator and equipped with a functional actuator at its end; the functional actuator is at least one of a gripper, an adsorption head, a touch head, or a writing head.

[0007] Furthermore, the sensing module includes a microphone array for acquiring user commands in the form of voice; and / or, the sensing module includes a camera for acquiring environmental information in the form of images.

[0008] Furthermore, the intelligent mechanical device also includes a safety monitoring module; The safety monitoring module is communicatively connected to the AI ​​processing module and the decision control module, and is used to monitor abnormal situations in the operation process in real time. When an abnormal situation is detected, the safety monitoring module sends an emergency stop signal to the decision control module, causing the multi-limb execution module to stop its current operation; the abnormal situation includes at least one of the following: incorrect identification of the operation object, obstruction of the operation path, abnormal operation force, and user intervention command.

[0009] A second aspect of the present invention provides a control method for operating an injection-type intelligent mechanical device, applied to operating the injection-type intelligent mechanical device, the method comprising controlling the AI ​​processing module to perform the following steps to recognize user intent: Semantic parsing is performed on user commands in the received perception data to extract semantic elements of operation actions and operation object identifiers; The semantic elements of the operation action and the identifier of the operation object are matched with the atomic operation template library in the operation injection information; The target atomic operation sequence corresponding to the user instruction is determined based on the matching results.

[0010] Furthermore, the step of matching the semantic elements of the operation action and the operation object identifier with the atomic operation template library in the operation injection information includes: Calculate the semantic similarity between the semantic elements of the operation action and the predefined operation type descriptions in each atomic operation template; Calculate the correlation degree between the operation object identifier and the predefined operation object description in each atomic operation template; By integrating the semantic similarity and correlation, the overall matching degree of each atomic operation template is obtained; The atomic operation template with the highest overall matching degree is selected as the matching result.

[0011] Furthermore, the establishment and updating of the atomic operation template library includes the following processes: Generate initial atomic operation templates based on preset demonstration data; During the use of the device, user commands and the final operation sequence that has been confirmed by the user to be executed correctly are continuously collected; The atomic operation template library is updated based on the collected data. The update operations include adding new templates or adjusting the matching parameters of existing templates.

[0012] A third aspect of the present invention provides a control method for operating an injection-type intelligent mechanical device, applied to operating the injection-type intelligent mechanical device, the method comprising controlling the decision control module to perform the following steps to perform task planning and generate motion control commands: Receive the sequence of atomic operations from the AI ​​processing module, and determine the execution parameters and order of each atomic operation based on the environmental information in the perceived data; Based on the environmental information, determine whether the current environmental state meets the execution conditions of the atomic operation; If the execution conditions are met, then motion control instructions to drive the multi-limb execution module are generated; If the execution conditions are not met, the execution parameters are adjusted based on environmental constraints and the judgment is made again. If the conditions are still not met after adjustment, the current task is paused and an exception message is issued.

[0013] Furthermore, the step of determining whether the current environment state meets the execution conditions of the atomic operation based on the environment information includes: When atomic operations involve physical contact, the spatial location, orientation, and accessibility of the target object are determined based on the visual information in the perceived data. When atomic operations involve force control, the force feedback information in the sensing data is used to determine whether the current contact force is within a safe range. When atomic operations involve continuous motion, the presence of obstacles or interference risks is determined based on the motion trajectory prediction information in the sensed data.

[0014] Furthermore, if the execution conditions are not met, the process of adjusting the execution parameters based on environmental constraints and then re-evaluating includes: If the spatial location does not meet the requirements, the motion path of the multi-limb execution module should be replanned. If the contact force exceeds the safe range, reduce the output force of the multi-limb actuator module and try contact again; If there are obstacles interfering, adjust the operation sequence or introduce obstacle avoidance actions.

[0015] As can be seen from the above technical solutions, the present invention has the following advantages: This invention integrates a sensing module into the main structure to collect environmental information and user commands. An AI processing module, combined with operation-injected information received from the communication module, identifies the user's intent. A decision control module then plans the task based on the identification results, generates control commands, and ultimately drives the multi-limb execution module to complete the physical operation. This invention achieves skill scalability and cross-device transferability through operation injection. Users can easily inject new operational logic into the device, enabling it to adapt to different devices and scenarios, overcoming the limitations of traditional automation scripts that rely on specific systems. It can understand natural language commands and autonomously adjust operations according to the dynamic environment, eliminating the need for complex reprogramming for each task change, thus solving the problems of poor flexibility and high deployment costs associated with dedicated robotic arms. Furthermore, it achieves intelligent and universal physical interaction capabilities, enabling it to seamlessly integrate into diverse daily physical interaction scenarios as an autonomous and flexible universal intelligent assistant. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the structure of an injection-type intelligent mechanical device provided by the present invention; Figure 2 A schematic diagram illustrating the process of controlling the AI ​​processing module to perform user intent recognition in the control method provided by the present invention; Figure 3 This is a flowchart illustrating the process by which the decision control module performs task planning and generates action control commands in the control method provided by the present invention. Detailed Implementation

[0017] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “corresponding to,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0018] Example 1 Please see Figure 1 The operation-injection intelligent mechanical device provided in this application embodiment includes: a main structure A, a sensing module 1, an AI processing module 2, a decision control module 3, a communication module 4, and a multi-limb execution module 5; the main structure A is the physical carrier of the intelligent mechanical device, providing mechanical support and an integration interface; the sensing module 1 is disposed on the main structure A and is communicatively connected to the AI ​​processing module 2, used to collect environmental information and user commands, and output the collected information as sensing data; the AI ​​processing module 2 is disposed within the main structure A and is communicatively connected to the sensing module 1, the communication module 4, and the decision control module 3, used to receive sensing data from the sensing module 1 and from the communication module 5. Block 4 injects operation information and performs user intent recognition based on perception data and operation injection information; Decision control module 3 is located within the main structure A and is communicatively connected to AI processing module 2 and multi-limb execution module 5, respectively, for task planning based on the intent recognition results of AI processing module 2 and generating motion control commands to drive multi-limb execution module 5; Communication module 4 is located within the main structure A and is communicatively connected to AI processing module 2, for receiving externally input operation injection information via wired or wireless means; Multi-limb execution module 5 is mechanically connected to the main structure A and communicatively connected to decision control module 3, for performing physical operations based on the received motion control commands.

[0019] Furthermore, the multi-limb execution module 5 includes at least one mechanical limb unit, each mechanical limb unit being driven by a micro-actuator and equipped with a functional actuator at its end; the functional actuator is at least one of a gripper, an adsorption head, a touch head, or a writing head.

[0020] Furthermore, the perception module includes a microphone array for acquiring user commands in the form of voice; and / or, the perception module includes a camera for acquiring environmental information in the form of images.

[0021] Furthermore, the intelligent mechanical device also includes a safety monitoring module 6; the safety monitoring module 6 is communicatively connected to the AI ​​processing module 2 and the decision control module 3, and is used to monitor abnormal situations in the operation process in real time; when an abnormal situation is detected, the safety monitoring module 6 sends an emergency stop signal to the decision control module 3, causing the multi-limb execution module 5 to stop the current operation; the abnormal situation includes at least one of the following: incorrect identification of the operation object, obstruction of the operation path, abnormal operation force, and user intervention command.

[0022] The working principle of the injection-type intelligent mechanical device is described below using specific scenarios: Taking the application scenario of a smart desktop assistant as an example, this operation-injected intelligent mechanical device is designed to be clamped onto the edge of a desk. The main structure uses a lightweight alloy frame, with dimensions of approximately 8 cm long, 5 cm wide, and 4 cm high. It integrates circuit board mounting positions internally and features standard screw holes and magnetic surfaces as integrated interfaces externally. When a user needs to change channels on the smart TV in front of them, they issue a voice command: "Switch to the next channel." The device's sensing module activates, its built-in microphone array picks up the voice command, and its camera captures image information from the smart TV and TV remote control. This environmental information and user command are output as sensing data to the AI ​​processing module. It should be noted that the user has previously injected the skill of operating the TV remote control into the device through a teaching recording method using a mobile application. The specific process is as follows: the user selects to record a new operation on the application, and then manually uses the robotic arm of the device's multi-limb execution module to press the volume up button on the remote control. The application records this action as an atomic operation, for example, the action type is "tap," the operation object is the remote control button, and the parameters include a force of 0.3 Newtons and a duration of 0.5 seconds. After the user repeatedly records button operations such as adding channels, the application software binds these atomic operation sequences with voice command tags to change channels. Through the Wi-Fi unit of the communication module, this set of operation injection information is sent and stored in the device's local AI processing module.

[0023] In this scenario, the AI ​​processing module simultaneously receives real-time perception data from the perception module and pre-stored operation injection information. It performs natural language understanding on the voice command, recognizing the user's intent as changing channels, and matches it with an operation template related to channel changing from the operation injection information. This template includes a series of atomic operation logics such as locating the remote control, moving to the target button, and performing a tap. Upon receiving this intent recognition result, the decision control module begins task planning. Combining real-time camera footage, it plans the motion path of the robotic arm from its initial position to the remote control's channel-adding button on the table, and generates specific motion control commands, including the rotation angle sequence of each joint servo and the end effector tap command. Upon receiving the command, the single three-degree-of-freedom robotic arm of the multi-limb execution module begins to move, and its end effector precisely moves above the remote control's channel-adding button and performs the tap operation, thus successfully completing the channel change.

[0024] Throughout the entire process, the safety monitoring module operates continuously. For example, if an obstacle, such as a water glass, suddenly appears in the robotic arm's path, the camera will detect this change. The safety monitoring module, by analyzing real-time sensing data, determines that the operation path is obstructed and immediately sends an emergency stop signal to the decision control module. The robotic arm then hovers to avoid a collision. After the user removes the water glass, the device can resume its task or wait for new instructions.

[0025] This embodiment fully demonstrates, through a specific desktop assistant scenario, how the technical solutions of operation injection, multimodal perception, intent recognition, task planning, and security control work together, thus fully illustrating the feasibility and practicality of the device.

[0026] Example 2 Please see Figure 2 The control method for the operation-injection-type intelligent mechanical device provided in this embodiment of the invention includes controlling the AI ​​processing module to perform the following steps to identify user intent: 201. Perform semantic parsing on the user instructions in the received sensing data to extract the semantic elements of the operation actions and the identifier of the operation object; The semantic parsing process first performs basic processing on the user command text, such as word segmentation and part-of-speech tagging. Using pre-defined parsing rules or a lightweight semantic understanding model, it identifies and extracts the core verbs or verb phrases from the command as semantic elements representing the type of action the user wishes to perform, such as tapping, swiping, grabbing, and rotating. Simultaneously, the parsing process identifies the nouns or noun phrases in the command that act as the recipient of the action, using them as object identifiers. These identifiers clarify the physical or logical target to which the action is performed, such as a TV remote control, an air conditioner switch, or an application icon on a mobile phone screen. For example, for the user command to turn on the living room light, the parsed semantic element of the action is "turn on," and the object identifier is "living room light."

[0027] 202. Match the semantic elements of the operation action and the identifier of the operation object with the atomic operation template library in the operation injection information; The Atomic Operation Template Library is a pre-built and continuously optimized knowledge base that stores a large number of standardized atomic operation templates. Each template defines an independently executable basic physics, such as a robotic arm moving to a coordinate, an end effector applying a specific force to press, and a communication module simulating a mouse click event. The creation and updating of the Atomic Operation Template Library involves the following processes: 1. Generate initial atomic operation templates based on preset demonstration data; 2. During device use, continuously collect user commands and the final operation sequence confirmed by the user to be executed correctly; 3. Update the atomic operation template library based on the collected data. Update operations include adding new templates or adjusting the matching parameters of existing templates.

[0028] Specifically, the preset demonstration data comes from standard operating procedures entered during the development phase through expert definition or user instruction, such as a complete set of steps for turning a TV on and off with a remote control. Secondly, during actual device use, the system continuously collects valid user commands and device operation sequences that have been ultimately confirmed by the user as correct. This data constitutes the learning samples for optimizing the template library. Based on the collected new sample data, the system dynamically updates the atomic operation template library. The update process includes not only adding new atomic operation templates for entirely new or rare tasks, but also fine-tuning the matching parameters or the generalization range of the operation object descriptions of existing templates based on user feedback, thereby making the template library increasingly intelligent and personalized.

[0029] To accurately find the operation template that best matches the current user command from the atomic operation template library, the matching process is as follows: 1. Calculate the semantic similarity between the semantic elements of the operation action and the predefined operation type descriptions in each atomic operation template; 2. Calculate the correlation degree between the operation object identifier and the predefined operation object description in each atomic operation template; 3. By integrating semantic similarity and relevance, the overall matching degree of each atomic operation template is obtained; 4. Select the atomic operation template with the highest overall matching degree as the matching result.

[0030] Specifically, the semantic similarity between the semantic elements of the operation extracted from the current user command and the predefined operation type descriptions within each atomic operation template in the library is calculated. The operation type description is the textual definition of the encapsulated action in the template, such as "press." Semantic similarity is calculated by comparing the distance between word vectors in the semantic space; closer distances indicate greater semantic similarity. Next, the correlation between the operation object identifier in the current user command and the predefined operation object descriptions in each template is calculated. The operation object description defines the target object features to which the template applies, such as an object with rectangular buttons. The correlation is calculated based on a comprehensive assessment of text keyword matching, visual feature similarity comparison, or relationships in a knowledge graph. Then, the semantic similarity and correlation calculated for each template are weighted and fused according to preset weights to obtain a comprehensive matching score representing the overall fit between the template and the current command. Finally, the entire template library is traversed, and the atomic operation template with the highest comprehensive matching score is selected as the matching result for the current user command.

[0031] 203. Determine the target atomic operation sequence corresponding to the user instruction based on the matching results.

[0032] The atomic operation template with the highest overall matching degree not only encapsulates basic atomic operations but also associates with an operation sequence composed of multiple atomic operations arranged in a specific logical order. This sequence is the target atomic operation sequence. For example, the matched template might be for changing channels on a TV remote control, and its associated target atomic operation sequence might be: visually positioning the remote control, moving the robotic arm above the remote control, aligning the end effector with the channel increment button, and performing the click action. Therefore, determining the matching result also determines the specific executable target atomic operation sequence required to execute the user command. This sequence will be output to the decision control module for subsequent task planning and action execution.

[0033] Example 3 Please see Figure 3 The control method for an operation-injection-type intelligent mechanical device provided in this embodiment of the invention includes a control decision control module performing the following steps to plan tasks and generate action control commands: 301. Receive the sequence of atomic operations from the AI ​​processing module, and determine the execution parameters and order of each atomic operation based on the environmental information in the perception data; After receiving the target atomic operation sequence output by the AI ​​processing module, the decision control module needs to instantiate the sequence in conjunction with real-time perception data. Execution parameters refer to the quantified values ​​required to drive the actuator, which concretize the abstract action description in the atomic operation. For example, for a click atomic operation, its execution parameters include the three-dimensional spatial coordinates to which the robotic arm's end effector needs to move, the precise force required to contact the target, and the duration of the press. The sequence refers to the logical temporal relationship between steps when the atomic operation sequence contains multiple steps; for example, the target must be located before movement can be performed, and the click must be performed last. The decision control module analyzes the perception data to interpret the environment, thereby determining these parameters and the sequence. For example, through visual recognition and spatial calculation, the precise position of the channel + button on the remote control in the current coordinate system is determined, and this is used as the coordinate parameter for the click operation.

[0034] 302. Determine whether the current environmental state meets the execution conditions of the atomic operation based on the environmental information; Before generating specific action instructions, it is necessary to assess whether the current environment allows for the safe and efficient execution of the planned atomic operations. The judgment process employs different perception dimensions and judgment criteria depending on the type of atomic operation: 1. When atomic operations involve physical contact, the spatial location, orientation, and accessibility of the target object are determined based on visual information from the sensory data; 2. When atomic operations involve force control, the system determines whether the current contact force is within a safe range based on force feedback information from the sensor data; 3. When atomic operations involve continuous motion, the presence of obstacles or interference risks is determined based on motion trajectory prediction information in the sensing data.

[0035] Specifically, when atomic operations involve physical contact, the decision control module, based on visual information from the perceived data, uses image recognition and target tracking algorithms to determine whether the spatial position of the target object is within the reachable workspace of the robotic arm, whether its posture facilitates vertical force application by the end effector, and whether there is a reachable, collision-free trajectory for the movement path from the current position to the target position. When atomic operations involve force control, the decision control module, based on force feedback information from the perceived data, monitors the current contact force when the robotic arm contacts the object in real time. The judgment is based on whether the contact force remains within a preset safety range. The upper limit of this range is the maximum permissible force to prevent damage to the equipment or object, and the lower limit is the minimum effective force to ensure the operation is effective. When atomic operations involve continuous motion, the decision control module, based on motion trajectory prediction information from the perceived data, simulates or quickly detects the planned motion path to determine whether there are obstacles or risks of interference with other limbs or environmental structures during the motion.

[0036] 303. If the execution conditions are met, generate motion control instructions to drive the multi-limb execution module; If, as determined in step 302, all necessary conditions are met, meaning the environmental conditions allow for safe and accurate operation, the decision control module converts the execution parameters and sequence determined in step 301 into low-level motion control instructions that can directly drive the micro-actuators in the multi-limb execution module. These instructions are typically pulse-width modulated signals or specific motor rotation angle and speed commands, controlling the coordinated movement of the robotic arm's joints to complete the specified atomic operations.

[0037] 304. If the execution conditions are not met, adjust the execution parameters based on environmental constraints and re-evaluate. If the conditions are still not met after adjustment, pause the current task and issue an exception message.

[0038] 1. If the spatial location does not meet the requirements, replan the motion path of the multi-limb execution module; 2. If the contact force exceeds the safe range, reduce the output force of the multi-limb actuator module and retry the contact; 3. If there are obstacles interfering, adjust the operation sequence or introduce obstacle avoidance actions.

[0039] Specifically, if the judgment result indicates that the execution conditions are not met, it means that there is a deviation between the current environmental state and the expectations, and direct execution may lead to failure or danger. At this time, the system will not give up immediately, but will try to adapt by adjusting the execution parameters based on environmental constraints: if the spatial position does not meet the requirements, the system will replan the motion path of the multi-limb execution module based on new visual information and calculate new reachable trajectory coordinates. If the contact force exceeds the safe range, the system will instruct the multi-limb execution module to reduce the output force and retry the contact in a gentler manner. If there is obstacle interference, the system may adjust the operation sequence or introduce obstacle avoidance actions to plan a new path around the obstacle.

[0040] After parameter adjustments are completed, the system will return to step 302 to reassess whether the execution conditions are met under the new parameters. If the environmental constraints still cannot be met after several adjustment attempts, the system will pause the current task and issue an error message to the user via indicator lights, sounds, or the communication module, reporting the reason for the failure and awaiting further user instructions or intervention. This ensures the safety and reliability of the entire operation process.

[0041] This invention achieves scalability, transferability, and rapid deployment of operational skills through an operation injection method. It combines the intelligent perception and decision-making capabilities of artificial intelligence with multi-degree-of-freedom physical execution capabilities to create a universal intelligent interactive device capable of understanding user intentions, adapting to dynamic environments, and performing complex physical operations.

[0042] It is understood that those skilled in the art can combine various implementation methods in the above embodiments under the guidance of the above examples to obtain technical solutions with multiple implementation methods.

[0043] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An injection-type intelligent mechanical device, characterized in that, include: Main structure, perception module, AI processing module, decision control module, communication module, and multi-limb execution module; The main structure serves as the physical carrier of the intelligent mechanical device, providing mechanical support and integration interfaces; The sensing module is mounted on the main structure and is communicatively connected to the AI ​​processing module. It is used to collect environmental information and user commands, and output the collected information as sensing data. The AI ​​processing module is located within the main structure and is communicatively connected to the perception module, the communication module, and the decision control module. It is used to receive perception data from the perception module and operation injection information from the communication module, and to perform user intent recognition based on the perception data and operation injection information. The decision control module is located within the main structure and is communicatively connected to the AI ​​processing module and the multi-limb execution module, respectively. It is used to perform task planning based on the intent recognition results of the AI ​​processing module and generate action control commands to drive the multi-limb execution module. The communication module is located within the main structure and is communicatively connected to the AI ​​processing module. It is used to receive operation injection information from external input via wired or wireless means. The multi-limb execution module is mechanically connected to the main structure and communicatively connected to the decision control module, and is used to perform physical operations according to the received motion control commands.

2. The operation injection-type intelligent mechanical device according to claim 1, characterized in that, The multi-limb execution module includes at least one mechanical limb unit, each of which is driven by a micro-actuator and equipped with a functional actuator at its end; the functional actuator is at least one of a gripper, an adsorption head, a touch head, or a writing head.

3. The operation injection-type intelligent mechanical device according to claim 1, characterized in that, The sensing module includes a microphone array for acquiring user commands in the form of voice; and / or, the sensing module includes a camera for acquiring environmental information in the form of images.

4. The operation injection-type intelligent mechanical device according to claim 1, characterized in that, The intelligent mechanical device also includes a safety monitoring module; The safety monitoring module is communicatively connected to the AI ​​processing module and the decision control module, and is used to monitor abnormal situations in the operation process in real time. When an abnormal situation is detected, the safety monitoring module sends an emergency stop signal to the decision control module, causing the multi-limb execution module to stop its current operation; The abnormal situations include at least one of the following: incorrect identification of the operation object, obstruction of the operation path, abnormal operation force, and user intervention instructions.

5. A control method for an injection-type intelligent mechanical device, characterized in that, Applied to operating injectable intelligent mechanical devices, the method includes controlling the AI ​​processing module to perform the following steps to identify user intent: Semantic parsing is performed on user commands in the received perception data to extract semantic elements of operation actions and operation object identifiers; The semantic elements of the operation action and the identifier of the operation object are matched with the atomic operation template library in the operation injection information; The target atomic operation sequence corresponding to the user instruction is determined based on the matching results.

6. The control method for the injection-type intelligent mechanical device according to claim 5, characterized in that, The step of matching the semantic elements of the operation action and the operation object identifier with the atomic operation template library in the operation injection information includes: Calculate the semantic similarity between the semantic elements of the operation action and the predefined operation type descriptions in each atomic operation template; Calculate the correlation degree between the operation object identifier and the predefined operation object description in each atomic operation template; By integrating the semantic similarity and correlation, the overall matching degree of each atomic operation template is obtained; The atomic operation template with the highest overall matching degree is selected as the matching result.

7. The control method for the injection-type intelligent mechanical device according to claim 5, characterized in that, The creation and updating of the atomic operation template library includes the following processes: Generate initial atomic operation templates based on preset demonstration data; During the use of the device, user commands and the final operation sequence that has been confirmed by the user to be executed correctly are continuously collected; The atomic operation template library is updated based on the collected data. The update operations include adding new templates or adjusting the matching parameters of existing templates.

8. A control method for an injection-type intelligent mechanical device, characterized in that, Applied to operating injection-type intelligent mechanical devices, the method includes controlling the decision control module to perform the following steps to plan tasks and generate motion control commands: Receive the sequence of atomic operations from the AI ​​processing module, and determine the execution parameters and order of each atomic operation based on the environmental information in the perceived data; Based on the environmental information, determine whether the current environmental state meets the execution conditions of the atomic operation; If the execution conditions are met, then motion control instructions to drive the multi-limb execution module are generated; If the execution conditions are not met, the execution parameters are adjusted based on environmental constraints and the judgment is made again. If the conditions are still not met after adjustment, the current task is paused and an exception message is issued.

9. The control method for the injection-type intelligent mechanical device according to claim 8, characterized in that, The step of determining whether the current environment state meets the execution conditions of the atomic operation based on the environmental information includes: When atomic operations involve physical contact, the spatial location, orientation, and accessibility of the target object are determined based on the visual information in the perceived data. When atomic operations involve force control, the force feedback information in the sensing data is used to determine whether the current contact force is within a safe range. When atomic operations involve continuous motion, the presence of obstacles or interference risks is determined based on the motion trajectory prediction information in the sensed data.

10. The control method for the injection-type intelligent mechanical device according to claim 8, characterized in that, If the execution conditions are not met, the execution parameters will be adjusted based on environmental constraints and a new judgment will be made, including: If the spatial location does not meet the requirements, the motion path of the multi-limb execution module should be replanned. If the contact force exceeds the safe range, reduce the output force of the multi-limb actuator module and try contact again; If there are obstacles interfering, adjust the operation sequence or introduce obstacle avoidance actions.

Images