Autonomous operation robot and method for power frequency line parameter tester in high induced electricity environment
By using a modular hardware platform and standardized software architecture, combined with 3D visual perception and full-body control, the safety and operational precision issues of robots in high-voltage power grid infrastructure scenarios have been solved, enabling precise operation and rapid migration capabilities in highly induced electric environments.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-25
- Publication Date
- 2026-04-07
AI Technical Summary
Existing robots face safety risks, insufficient operational precision, inadequate system intelligence, and difficulties in training and deployment in high-voltage power grid infrastructure scenarios, making it impossible for them to stably perform wiring, disconnection, and testing processes in high-inductive-electric environments.
By adopting a modular hardware platform and standardized software architecture, and combining 3D vision perception, whole-body control and task planning modules, the robot can achieve precise operation in a highly induced electric environment through modular architecture, standardized training process and behavior tree task planning.
It achieves safe and precise operation in high-inductive electric environments, with a success rate of over 95%, significantly improving the robot's safety and reliability, and enabling it to be quickly transferred to other high-risk work scenarios.
Smart Images
Figure CN121798581A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of robotics, and in particular to an autonomous robot and method for testing power frequency line parameters in a high-inductive electrical environment. Background Technology
[0002] Existing humanoid robot bodies typically provide humanoid-like mobility and manipulation capabilities, with dexterous hands possessing multi-degree-of-freedom finger structures to support fine movements. ROS (Robot Operating System), as a middleware layer, supports modular software development and is widely used in industrial automation scenarios, but it has not yet been optimized for standardized operation in high-voltage induced voltage environments. Specifically, existing technologies face the following main problems when applied to fine-grained operations in high-voltage power grid infrastructure scenarios: 1) Safety risks: In operations such as testing power frequency line parameters, the high induced electric environment poses a serious risk of electric shock to manual operation. Existing passive protection measures such as insulating blankets and insulating boots cannot fundamentally eliminate the risk. 2) Operational precision issues: Commercial robots or general-purpose robotic arms lack customized design and sub-millimeter precision operation capabilities for power frequency line parameter testers (such as terminal blocks and knife switches), making it difficult to achieve precise alignment, insertion and removal of terminal blocks and accurate pressing of switches; 3) System intelligence issues: Existing robot systems lack accurate environmental perception, dexterous whole-body coordinated control, and robust task planning capabilities in unstructured and interference-prone power operation environments, and cannot stably and autonomously execute complete wiring, disconnection, and testing processes. 4) Training and deployment issues: Traditional robot simulation training differs significantly from the real physical environment and does not fully consider factors such as changes in lighting, occlusion, and physical disturbances in the real environment. This results in low reliability of the control strategy in actual scenarios and difficulty in guaranteeing the success rate of operation. Summary of the Invention
[0003] Purpose of the invention: The purpose of this invention is to provide an autonomous operating robot and method for a power frequency line parameter tester in a high-inductive electric environment, enabling the robot to perform a series of precise operations such as accurate alignment of wiring terminals, plugging and unplugging, and precise pressing of switches in a high-inductive electric environment.
[0004] Technical solution: An autonomous operating robot for a power frequency line parameter tester in a high-inductive electrical environment, comprising: The physical layer includes the humanoid robot body, a high-precision multi-degree-of-freedom dexterous hand, a data acquisition module, and a working frequency circuit parameter tester that serves as the operational target. The middleware layer enables real-time and secure data exchange and communication between modules in the physical, cognitive, and control layers. The cognition and control layer includes a 3D vision perception module, a whole-body control module, and a task planning module, which are used to realize active perception, autonomously output control commands, drive the robot body or head to adjust its position to re-observe the target, and form a closed-loop perception. The monitoring and training layer, including operator monitoring stations, digital twin simulation environments, and domain-randomized machine learning pipelines, is used for training, validating, and iteratively optimizing control strategies.
[0005] Furthermore, the data acquisition module is used to achieve 360° panoramic scanning and depth perception of the target being operated on; The three-dimensional vision perception module is used to identify targets on the testing instrument and accurately estimate the 6D pose of the targets. The whole-body control module adopts a unified whole-body control framework to achieve decoupling and coordinated control of lower body balance and upper body fine operation. The task planning module uses behavior trees to model, schedule, and handle exceptions for the entire job process.
[0006] An autonomous operation method for a power frequency line parameter tester in a high-inductive electrical environment, comprising, as described above, a closed-loop collaborative operation of an autonomous robot through a cognitive and control layer, encompassing perception, planning, and control operations, to achieve precise alignment, insertion / removal of wiring terminals, and accurate pressing of switches; the implementation steps are as follows: S1 uses a 3D vision perception module to perceive and locate the operating environment and adjust the robot's pose. S2, based on the adjusted target pose, the task planning module completes specific task planning and execution; S3, the whole-body control module receives instructions from the task planning module in real time and converts them into physical actions to complete the coordinated control of the whole body.
[0007] Furthermore, the detailed steps for adjusting the robot's pose are as follows: S11 collects point cloud data and RGB image data of the operating environment through the robot's data acquisition module; S12, the RGB image data is input into a pre-trained 2D target detection model to identify the two-dimensional bounding box information of at least one target; S13, Based on the two-dimensional bounding box information and the point cloud data, project the two-dimensional bounding box onto the point cloud data, extract the point cloud subset of the target region, and calculate the 6D pose of the target through a point cloud processing algorithm; S14, evaluate the confidence score of the 6D pose. If the confidence score is higher than or equal to a preset threshold, it is directly used for subsequent tasks. If the confidence score is lower than the preset threshold, output an adjustment command to adjust the robot's pose and return to step 11 to re-collect data until the confidence score reaches or exceeds the preset threshold.
[0008] Furthermore, the 6D pose of each target is [x, y, z, roll, pitch, yaw], where x, y, and z represent the coordinates of the object in three-dimensional space, roll represents the rotation angle of the object around its own x-axis, pitch represents the rotation angle of the object around its own y-axis, and yaw represents the rotation angle of the object around its own z-axis.
[0009] Furthermore, the detailed steps for the task planning module to complete specific task planning and execution are as follows: S21, Receive the "Execute Test" instruction from the operator; S22, the top-level behavior tree is activated and sub-tasks are scheduled in a preset order: the robot first moves to the operating position; executes a wiring sub-sequence; verifies the test status through the 3D vision perception module; and returns to the safe position after the task is completed. If a failure is encountered during execution, the current process is interrupted via the Fallback node, and the exception handler is triggered. S23 continuously outputs specific task instructions to the underlying control module to drive the robot to complete physical operations.
[0010] Furthermore, the detailed steps by which the whole-body control module completes whole-body coordinated control are as follows: S31, take the real-time status data input by the robot body sensor, the current task instruction input by the task planning module, and the target 6D pose input by the 3D vision perception module as input data; S32, the whole body control module uses the SoFTA framework for decoupling. The upper body agent completes the high-frequency upper body joint motion commands a_t^U, and the lower body agent completes the low-frequency lower body joint motion commands a_t^L. S33 combines the upper body joint motion command a_t^U and the lower body joint motion command a_t^L into a complete full-body joint command at, as shown in the following expression: at = π(st, dt) Among them, π is a control policy network pre-trained based on the monitoring and training layers; S34, the full-body joint command at is output to each servo motor of the robot body, driving the robot to perform physical operations.
[0011] Furthermore, the control policy network π is trained in the NVIDIA Isaac Sim simulation environment through a monitoring and training layer, and the implementation process is as follows: S331, a simulation environment is built based on the robot's URDF model, the operator console 3D model, and physical parameters; S332 randomizes lighting, texture, object position, and physical properties in simulation, and adopts a training strategy from easy to difficult, allowing the robot to first master basic skills and then learn complex tasks. S333 uses a comprehensive reward function R_total to guide the learning direction, while optimizing task completion, body balance and motion smoothness, and trains a control policy network π. The control strategy network π directly calculates the joint commands at the bottom layer based on the current state, driving the robot's servo motors to perform physical operations.
[0012] Furthermore, the expression for the comprehensive reward function is as follows: R_total = w1 * R_task + w2 * R_balance + w3 * R_penalty Where R_task represents the higher the reward as the end effector gets closer to the target; R_balance represents the higher the reward as the body posture is more stable; R_penalty represents the lower the reward as the joint acceleration is greater; and w1, w2, and w3 are the weights.
[0013] Compared with the prior art, the significant advantages of this invention are as follows: 1. The modular hardware platform (such as Unitree G1 + Dex5) and standardized software architecture (ROS 2 + MHC) adopted by the robot of this invention have good versatility. It adopts a modular architecture, standardized training process, and visual configuration of task tree. By retraining and adjusting the task tree, it can be quickly migrated to other high-risk operation scenarios, such as equipment inspection and live-line work. When migrating to new scenarios, "there is no need to reconstruct the core architecture, only fine-tuning of training and task configuration is required", and the migration cycle is shortened to 1-2 weeks. This solves the difficulties of traditional robots, such as "poor scene adaptability and high migration cost". It can completely replace human beings in high-inductive-electricity areas to perform wiring, plugging and unplugging and pressing operations, fundamentally eliminating the risk of electric shock from manual operation. 2. The robot of this invention is based on high-precision perception, decoupled collaborative control and behavior tree task planning, and can achieve sub-millimeter level fine operation. Experimental tests (simulated and real environment tests ≥100 times) show that the operation success rate is consistently above 95%, which is significantly better than the consistency and reliability of manual operation. 3. The robot of the present invention, through an active perception strategy, is trained in a randomized environment based on the monitoring and training layers. Combined with the control logic of the collaborative cognition and control layers, it can effectively cope with environmental disturbances such as changes in lighting, slight occlusion, and target pose deviations in real work sites. Attached Figure Description
[0014] Figure 1 This is a schematic diagram of the overall frame of the robot in an embodiment of the present invention; Figure 2 This is a schematic diagram of the humanoid robot (G1) used in an embodiment of the present invention; Figure 3 This is a schematic diagram of the high-precision dexterous hand (Dex5) used in the embodiments of the present invention; Figure 4 This is a schematic diagram of the SoFTA (Slow-Fast Dual Agent Framework) used in the whole-body control module in this embodiment of the invention; Figure 5 This is a schematic diagram of the Isaac Sim simulation and training process used in the monitoring and training layers of this invention. Detailed Implementation
[0015] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.
[0016] like Figure 1 As shown, an autonomous operating robot for a power frequency line parameter tester in a high-inductive electrical environment adopts a layered modular architecture, including: The physical layer includes the humanoid robot body, a high-precision multi-DOF dexterous hand, a data acquisition module (including a 3D LiDAR and a depth camera), and a working frequency circuit parameter tester as the operational target.
[0017] In this embodiment, the humanoid robot body adopts the Unitree Robotics G1 humanoid robot (e.g., Figure 2 As shown, the humanoid robot is about 130cm tall and weighs about 35kg. It has an overall degree of freedom of 23-43 and a single-arm effective payload of 3kg, which meets the requirements for carrying tools and operating testing instruments.
[0018] The high-precision dexterous hand adopts Unitree Technology's Dex5 dexterous hand (such as...). Figure 3 As shown, this hand has 20 degrees of freedom (16 active joints) and can be equipped with 94 tactile sensors. It can simulate the fine movements of a human hand, such as grasping, pinching, and pressing, and is used to operate the terminals and switches of the testing instrument.
[0019] The data acquisition module uses a 3D LiDAR and depth camera integrated into the robot body to achieve 360° panoramic scanning and depth perception of the target being operated on.
[0020] The middleware layer, built on the Robot Operating System (ROS 2), enables real-time and secure data exchange and communication between modules in the physical, cognitive, and control layers.
[0021] The cognition and control layer is the core of the autonomous robot, including a 3D vision perception module, a whole-body control module, and a task planning module. The functions of each module are as follows: The 3D vision perception module is used to identify targets such as terminals and switches on the tester and accurately estimate their 6D pose. Full-body control module: Adopts a unified full-body control framework (such as Masked Humanoid Controller (MHC)) to achieve decoupling and collaborative control of lower body balance and upper body fine manipulation; Task planning module: Behavior trees (BTs) are used to model, schedule, and handle exceptions for the entire job process.
[0022] Monitoring and training layer: This includes operator monitoring stations, digital twin simulation environments (such as NVIDIA Isaac Sim), and domain-randomized machine learning pipelines for training, validating, and iteratively optimizing control strategies.
[0023] An autonomous operation method for a power frequency line parameter tester used in highly induced electrical environments is disclosed. This method utilizes an autonomous robot to achieve closed-loop collaboration from perception, planning, and control operations through a cognitive and control layer, enabling precise alignment, insertion / removal of terminals, and accurate pressing of switches. The implementation steps are as follows: Step 1: The 3D vision perception module performs perception and localization. When the autonomous robot receives the "start operation" command, the 3D vision perception module is activated first and performs the following operations: Step 11, Data Input; The robot moves to the operating position, and its head-mounted 3D LiDAR and depth camera acquire point cloud data and RGB image data from the operating platform (power frequency line parameter tester).
[0024] Step 12, 2D target detection; RGB image data is input into the YOLO model, which then performs forward propagation on the image. Identify pre-trained target categories; identify the two-dimensional bounding box information (e.g., [x_min, y_min, x_max, y_max]) of the identified targets such as "terminal hole" and "start switch".
[0025] Step 13, 6D pose estimation; Using the two-dimensional bounding box output in step 12 and the point cloud data collected in step 11 as input, the three-dimensional vision perception module projects the two-dimensional bounding box onto the point cloud data, extracts a subset of the point cloud in the target region, and uses an algorithm (such as ICP or RANSAC (Random Sample Consensus)) to fit and calculate the precise spatial position and orientation of the point cloud subset.
[0026] Output the 6D pose ([x, y, z, roll, pitch, yaw]) of each operation target (e.g., terminal A). This data will be used as input for subsequent control and planning. Here, x, y, and z are the coordinates of the object in three-dimensional space, roll (rotation around the x-axis) represents the rotation angle of the object around its own x-axis, pitch (rotation around the y-axis) represents the rotation angle of the object around its own y-axis, and yaw (rotation around the z-axis) represents the rotation angle of the object around its own z-axis.
[0027] Step 14, Active sensing (closed loop); Input: Confidence score of 6D pose estimation.
[0028] Processing: The 3D vision perception module determines whether the confidence score is lower than a preset threshold (e.g., 0.9). If the confidence score is higher than or equal to the preset threshold, it means that the estimated result has a small deviation from the actual pose and can be directly used for subsequent tasks; if it is lower than the threshold (e.g., due to occlusion or insufficient lighting), the 3D vision perception module will output an adjustment command to the whole-body control module, requiring the robot to slightly move its body or turn its head.
[0029] Loop: After the robot is adjusted, it returns to step 11 to collect data again until the perception confidence level reaches the target.
[0030] Step 2: The task planning module completes specific task planning and execution (logical scheduling). After obtaining the precise target pose, the task planning module (based on Behavior Trees, BTs) starts and performs the following operations: Step 21, top-level task input; The operator issues the "Execute test" instruction.
[0031] Step 22, Behavior tree scheduling (processing); The top-level behavior tree (BTs) is activated and begins to execute sub-tasks in sequence (including a series of specific action instructions such as grabbing the wire, accurately inserting the wire into the designated terminal hole, and pressing the start switch).
[0032] Subtask 1: [Move to the operation position].
[0033] Subtask 2: [Execute wiring sequence]; This subtask itself is a behavior tree: A21) Output command dt_1 to the control module: "Grab voltage line Ua".
[0034] A22) outputs instruction dt_2 to the control module: "Insert Ua into terminal hole A" (using the 6D pose data of hole A).
[0035] A23)... (Reconnect other current lines)... A24) outputs instruction dt_3 to the control module: "Press to start switch" (using the switch's 6D pose data).
[0036] Subtask 3: [Verify test status] (e.g., visually read the tester screen lights).
[0037] Subtask 4: [Return to a safe position].
[0038] Step 23, Exception handling (Fallback); If the control module reports "insertion failure" during the execution of dt_2 (e.g., the force sensor detects excessive resistance), the [Execute Wiring Sequence] subtree will be interrupted through the Fallback node and an exception handling instruction will be output, such as "Re-execute [Sensing and Positioning] Module" or "Alarm and wait for manual intervention".
[0039] Step 24, Task Output: Throughout the process, the task planning module continuously outputs masked task instructions (dt_1,dt_2,dt_3...) to the control module.
[0040] Step 3: The full-body control module completes full-body coordinated control (command execution); the full-body control module receives commands from the task planning module in real time and converts these commands into physical actions. The implementation is as follows: Step 31, control the input; The input data includes: st: Real-time status data input from the robot's body sensors (joint encoders, IMU).
[0041] dt: The current task instruction input by the task planning module (e.g., dt_2: "Insert Ua into terminal hole A").
[0042] Pose_A: The 6D pose of terminal hole A input from the 3D vision perception module.
[0043] Step 32, Decoupling control (processing); In this embodiment, the core responsibility of the whole-body control module is to "decouple and coordinate control of lower body balance and upper body fine manipulation," explicitly dividing the control task into two independent but coordinated agents: the "upper body" and the "lower body." The upper body agent (fast), as a sub-functional unit of this module, focuses on "high-frequency fine manipulation," complementing the lower body agent (slow)'s "low-frequency balance control," together constituting the complete control logic of the whole-body control module. The whole-body control module adopts the SoFTA framework (e.g., Figure 4 Decoupling is performed (as shown) to complete the following actions: B321) Upper body agent (fast); Input: dt_2 and Pose_A.
[0044] Processing: The calculation of how the dexterous end effector (EE) should move to accurately reach Pose_A is achieved through the "whole-body control module" combined with "sensory data + motion planning + high-frequency closed-loop control". The core logic is "pose error modeling → trajectory planning → joint command generation → real-time correction".
[0045] Output: High-frequency (e.g., 1kHz) upper body joint movement commands a_t^U.
[0046] B322) Lower body agent (slow speed); Input: st (IMU attitude data).
[0047] Solution: Calculate how to adjust the legs and waist to maintain overall balance and counteract the disturbances caused by upper body movements.
[0048] Output: Low-frequency (e.g., 100Hz) lower body joint movement commands a_t^L.
[0049] Step 33, Strategy execution (processing); a_t^U and a_t^L are combined to form a complete whole-body joint command at (at = π(st, dt)), and the control strategy π is pre-trained in the monitoring and training layer.
[0050] The aforementioned control strategy network π is implemented through monitoring and training layers in an NVIDIA Isaac Sim simulation environment (e.g., Figure 5 The training process shown is as follows: Step 331, Training Input; Robot URDF (Unified Robot Description Format) model, console 3D model, physical parameters (such as gravity and friction).
[0051] Step 332, Training Process; Domain randomization: Randomize the inputs for lighting, texture, tester position (±5cm deviation), terminal hole friction coefficient, etc. in the simulation.
[0052] Course Learning: The robot first learns the basics of [standing and balancing], then learns [arm movement], and finally learns [hand-eye coordination for inserting terminals].
[0053] Reward function: Input a comprehensive reward function, the expression of which is as follows: R_total = w1 * R_task + w2 * R_balance + w3 * R_penalty Where R_task represents the higher the reward (positive reward) as the end effector is closer to the target; R_balance represents the higher the reward (positive reward) as the body posture is more stable (IMU data is close to horizontal); R_penalty represents the lower the reward (negative penalty) as the joint acceleration is greater (the more "shaky" the movement); w1, w2, and w3 are the weights.
[0054] Step 333, training output; A robust MHC control policy network π that can adapt to environmental changes.
[0055] Step 34, Action Output; The final commands at (e.g., torque commands) for all joints of the body are output to the various servo motors of the robot body, driving the robot to perform physical operations (such as extending an arm or inserting a finger).
[0056] This invention, through the aforementioned hardware platform, perception-planning-control closed-loop process, and high-fidelity simulation training, enables a humanoid robot to operate a power frequency line parameter tester autonomously, safely, and reliably in a high-inductive electrical environment.
Claims
1. An autonomous operating robot for testing power frequency line parameters in a high-inductive electrical environment, characterized in that, include: The physical layer includes the humanoid robot body, a high-precision multi-degree-of-freedom dexterous hand, a data acquisition module, and a working frequency circuit parameter tester that serves as the operational target. The middleware layer enables real-time and secure data exchange and communication between modules in the physical, cognitive, and control layers. The cognition and control layer includes a 3D vision perception module, a whole-body control module, and a task planning module, which are used to realize active perception, autonomously output control commands, drive the robot body or head to adjust its position to re-observe the target, and form a closed-loop perception. The monitoring and training layer, including operator monitoring stations, digital twin simulation environments, and domain-randomized machine learning pipelines, is used for training, validating, and iteratively optimizing control strategies.
2. The autonomous robot for testing power frequency line parameters in a high-inductive electrical environment according to claim 1, characterized in that, The data acquisition module is used to achieve 360° panoramic scanning and depth perception of the target being operated on. The three-dimensional vision perception module is used to identify targets on the testing instrument and accurately estimate the 6D pose of the targets. The whole-body control module adopts a unified whole-body control framework to achieve decoupling and coordinated control of lower body balance and upper body fine operation. The task planning module uses behavior trees to model, schedule, and handle exceptions for the entire job process.
3. A self-operation method for a power frequency line parameter tester in a high-inductive electrical environment, characterized in that, As described in any one of claims 1-2, the autonomous robot completes a closed-loop collaboration from perception, planning, and control operation through a cognitive and control layer, achieving precise alignment, insertion / removal of wiring terminals, and accurate pressing of switches; the implementation steps are as follows: S1 uses a 3D vision perception module to perceive and locate the operating environment and adjust the robot's pose. S2, based on the adjusted target pose, the task planning module completes specific task planning and execution; S3, the whole-body control module receives instructions from the task planning module in real time and converts them into physical actions to complete the coordinated control of the whole body.
4. The autonomous operation method of the power frequency line parameter tester in a high-inductive electrical environment as described in claim 3, characterized in that, The detailed steps for adjusting the robot's pose are as follows: S11 collects point cloud data and RGB image data of the operating environment through the robot's data acquisition module; S12, the RGB image data is input into a pre-trained 2D target detection model to identify the two-dimensional bounding box information of at least one target; S13, Based on the two-dimensional bounding box information and the point cloud data, project the two-dimensional bounding box onto the point cloud data, extract the point cloud subset of the target region, and calculate the 6D pose of the target through a point cloud processing algorithm; S14, evaluate the confidence score of the 6D pose. If the confidence score is higher than or equal to a preset threshold, it is directly used for subsequent tasks. If the confidence score is lower than the preset threshold, output an adjustment command to adjust the robot's pose and return to step 11 to re-collect data until the confidence score reaches or exceeds the preset threshold.
5. The autonomous operation method of the power frequency line parameter tester in a high-inductive electrical environment as described in claim 4, characterized in that, The 6D pose of each target is [x, y, z, roll, pitch, yaw], where x, y, and z represent the coordinates of the object in three-dimensional space, roll represents the rotation angle of the object around its own x-axis, pitch represents the rotation angle of the object around its own y-axis, and yaw represents the rotation angle of the object around its own z-axis.
6. The autonomous operation method of the power frequency line parameter tester in a high-inductive electrical environment as described in claim 3, characterized in that, The detailed steps for the task planning module to complete specific task planning and execution are as follows: S21, Receive the "Execute Test" instruction from the operator; S22, the top-level behavior tree is activated and sub-tasks are scheduled in a preset order: the robot first moves to the operation position; executes a wiring sub-sequence; and verifies the test status through the 3D vision perception module. Return to a safe location after completing the task; If a failure is encountered during execution, the current process is interrupted via the Fallback node, and the exception handler is triggered. S23 continuously outputs specific task instructions to the underlying control module to drive the robot to complete physical operations.
7. The autonomous operation method of the power frequency line parameter tester in a high-inductive electrical environment as described in claim 3, characterized in that, The detailed steps by which the whole-body control module completes whole-body coordinated control are as follows: S31, take the real-time status data input by the robot body sensor, the current task instruction input by the task planning module, and the target 6D pose input by the 3D vision perception module as input data; S32, the whole body control module uses the SoFTA framework for decoupling. The upper body agent completes the high-frequency upper body joint motion commands a_t^U, and the lower body agent completes the low-frequency lower body joint motion commands a_t^L. S33 combines the upper body joint motion command a_t^U and the lower body joint motion command a_t^L into a complete full-body joint command at, as shown in the following expression: at = π(st, dt) Among them, π is a control policy network pre-trained based on the monitoring and training layers; S34, the full-body joint command at is output to each servo motor of the robot body, driving the robot to perform physical operations.
8. The autonomous operation method of the power frequency line parameter tester in a high-inductive electrical environment as described in claim 7, characterized in that, The control policy network π is obtained by training a monitoring and training layer in the NVIDIA Isaac Sim simulation environment. The implementation process is as follows: S331, a simulation environment is built based on the robot's URDF model, the operator console 3D model, and physical parameters; S332 randomizes lighting, texture, object position, and physical properties in simulation, and adopts a training strategy from easy to difficult, allowing the robot to first master basic skills and then learn complex tasks. S333 uses a comprehensive reward function R_total to guide the learning direction, while optimizing task completion, body balance and motion smoothness, and trains a control policy network π. The control strategy network π directly calculates the joint commands at the bottom layer based on the current state, driving the robot's servo motors to perform physical operations.
9. The autonomous operation method of the power frequency line parameter tester in a high-inductive electrical environment as described in claim 8, characterized in that, The expression for the comprehensive reward function is as follows: R_total = w1 * R_task + w2 * R_balance + w3 * R_penalty Where R_task represents the higher the reward as the end effector gets closer to the target; R_balance represents the higher the reward as the body posture is more stable; R_penalty represents the lower the reward as the joint acceleration is greater; and w1, w2, and w3 are the weights.