Teleoperated Live-Line Distribution Working Robot Based on Human-Robot Shared Control

Summary

A teleoperated live-line distribution working robot based on human-robot shared control has been developed. Laboratory experiments showed that shared-control weighting improved trajectory efficiency while reducing task completion time and submovements. An abrupt teleoperation behavior detection algorithm reduced sudden operator movements and was successfully validated in field tapping operations.

Abstract

Traditional live-line work on distribution power lines relies on manual labor, characterized by high risk, high labor intensity, and high training costs; thus, developing distribution live-line working robots is an essential path to replacing manual operations. However, the working environment is highly unstructured and dynamic, with challenges such as complex outdoor lighting and unpredictable wind conditions. Existing robots primarily use either master-slave teleoperation or fully autonomous modes, which impose heavy mental burdens on operators, are highly susceptible to misoperation, and exhibit limited environmental adaptability.

To address these issues, this study presents a teleoperated distribution live-line working robot based on human-robot shared control. The final manipulator trajectory is obtained by a linearly weighted combination of two reference trajectories separately provided by the human operator and the robot’s automatic planning module. This approach reduces the robot's dependency on the operator, alleviates operator mental burden, improves efficiency, and prevents abrupt misoperations by the human operator. Finally, the paper presents the system design and verifies its effectiveness through application in real-world live-line tapping operations.

Introduction

Live-line working is a critical method for the inspection, testing, and maintenance of high-voltage equipment. The volume of live-line work in China has been growing rapidly. From the perspective of industry demand, solving the issue of personnel safety in live-line working is a long-term and arduous task. Innovation and improvement are still needed in methods, technologies, equipment, and concepts.

Distribution live-line working is one of the most common types of live-line working, which is for power distribution live-line maintenance. Due to the complex lighting environment, dense equipment layout, and small phase-to-phase and phase-to-ground distances in distribution power lines, these operations have largely relied on traditional manual methods. However, this traditional approach faces problems such as high risk, high labor intensity, and high training costs¹. Therefore, developing robots with live-line working capabilities to replace manual labor has become an inevitable trend for future distribution network operation and maintenance.

Since the 1980s, countries such as Japan and the United States have conducted research and trials in this field (as shown in Figure 1). The earliest systems were master-slave tracking teleoperated robots. For example, Japan's Phase II live-line working robot² installed a three-degree-of-freedom (DOF) auxiliary manipulator on the platform, in addition to a dual-DOF electric manipulator. The human operator controlled it via an operating rod at close range.

The US TOMCAT³ model utilized a hydraulic manipulator controlled by a human operator within a working bucket. Similarly, the Shandong Power Grid⁴ in China adopted hydraulic manipulators controlled by operators standing in a working bucket. Furthermore, to increase the allowable distance between the master and slave ends, the Spanish ROBTET^5,6 used fiber-optic cables for remote communication, enabling operators to perform teleoperation from the ground. These robot teleoperation modes can be summarized as a master-slave tracking mode. They suffer from insufficient autonomous intelligence, requiring personnel to observe the environment and control the robot at close range and bear a significant mental burden, thus failing to replace humans in high-voltage, hazardous environments.

To address the lack of robotic intelligence, research into intelligent technologies such as computer vision, trajectory planning, and human-robot remote interaction is required to enhance the robot's autonomous capabilities.

In distribution live-line working, the tapping task accounts for approximately 20%⁷ of the total workload (distribution wires are shown in Figure 2). Therefore, using vision sensors to identify and locate the wire and its stripped region, which is the insulation removed region, to guide the robot is crucial. In 2020, State Grid Ruijia⁸ developed a distribution live-line working robot capable of remote interaction, fusing LiDAR and binocular camera information for target recognition and positioning. In addition, algorithms combining morphological operations and Hough transform are used to facilitate power line extraction^9,10. The Hough transform is used to detect power lines, followed by stereo vision algorithms that convert the detected 2D pixel data into 3D spatial coordinates¹¹. State Grid Hubei Electric Power Research Institute¹² proposed a filter based on color–depth map fusion. This method filters out complex backgrounds surrounding the power lines, then extracts clear power line contours by analyzing pixel-value differences in the neighborhood. Additionally, the University of Science and Technology of China¹³ proposed a deep learning-based power line recognition algorithm, achieving rapid identification and localization of multiple power lines.

Despite the rapid development of computer vision algorithms, outdoor operations remain challenging. Complex lighting conditions and reflective wire surfaces often lead to a loss of depth information, resulting in low recognition rates and large positioning errors for methods that rely heavily on depth data¹³. Simultaneously, due to various irregular deformations existing in distribution wires, a series of straight-line detection methods are no longer applicable, as shown in Figure 3. Furthermore, changes in wind direction at elevated heights cause vertical movement of the lines, leading to centimeter-level errors in the results of existing off-line algorithms.

In such highly dangerous, unstructured, and dynamic environments, which can be called special operations environments, the robot's vision and dynamic decision-making capabilities are constrained, making it difficult to complete complex tasks autonomously.

Therefore, it remains necessary to introduce human intelligent decision-making capabilities and combine them with the robot's high-precision motion control. The early teleoperation technologies mentioned above provided a channel for human intervention. By combining these with rapidly developing remote interaction and teleoperation technologies, a body of research on intelligent teleoperated distribution live-line working robots has emerged.

The University of Science and Technology of China¹⁴ implemented incremental position mapping between the master and slave sides and utilized the Artificial Potential Field (APF) method for robot obstacle avoidance. Subsequently, admittance-based teleoperation was implemented in distribution live-line working robots. Building upon master-slave tracking, this approach allows the robot to dynamically adjust its pose in response to changes in environmental contact forces. The University of Electronic Science and Technology of China¹⁵ implemented virtual constraint forces on the master side to enhance the operator's sense of telepresence. Additionally, the University of Manitoba¹⁶ established a series of virtual fixtures during the teleoperation process to reinforce the immersive experience of the operator's haptic perception.

However, the teleoperated robots mentioned above still face several challenges. The task of the slave side is to track the motion trajectory of the master, making it highly dependent on the reference trajectory generated by the human operator. Due to requirements such as short working times and obstacle avoidance, it imposes a significant mental burden on the master operator, which is detrimental to long-duration and complex operational tasks. The operator may commit improper or incorrect actions, causing abrupt motion in the actual trajectory of the manipulator. This can lead to safety issues such as collisions with the environment or phase-to-phase short circuits.

Shared control is a method that effectively combines human and robot decision-making capabilities^17,18. Generally, it requires both the human and the robot to generate desired trajectories or trajectory points simultaneously, and then calculates a new shared trajectory through a linear weighted combination. This approach enables the system to adjust the preference of the shared trajectory by modifying the decision-making weights of the human and the robot.

Therefore, this paper designs a teleoperated distribution live-line working robot based on human-robot shared control. The main contributions are as follows: first, addressing the issue of excessive frequency of reference trajectory input by the human operator during teleoperation, a human-robot shared control method is studied. By assigning control weights to the human operator and the robot and performing a linear weighted combination on their reference trajectories, the system reduces the dependency of the reference trajectory on the human operator, thereby alleviating the operator's burden. In addition, the impact of different human operator weights is verified through field experiments of distribution wire tapping operations. Second, addressing the dangerous behaviors in the master trajectory, an abrupt operation behavior detection method is designed to ensure the smoothness and safety of the manipulator's actual trajectory. Third, physical experiments are conducted in a special operation environment (live-line working) to verify the effectiveness of the shared control algorithm and the teleoperation system.

The remainder of this article first presents the shared control method for the teleoperated robot, then introduces the system hardware and setup, and finally validates the approach through laboratory experiments and field deployment on a live 10 kV distribution network.

Protocol

All experiments involving operators were conducted in accordance with safety guidelines. The teleoperation experiments were conducted by three trained operators (all right-handed, with at least 1 h of training) who are members of the research team. No external human participants were recruited, and this study did not require institutional ethics review. Volunteers were informed of the experimental procedures, and no personal or sensitive data were collected during the study.

Shared control framework

In live-line working scenarios, factors such as the irregular bending of power lines, the loss of depth information caused by complex outdoor lighting, and the random movement of power lines create a highly unstructured and dynamic environment. These factors make it difficult for the robot to accurately recognize and locate power lines within the environment. Consequently, relying entirely on autonomous robot operation would lead to task failure.

To address the limited adaptability of fully autonomous robot operations, human operators intervene in the robot's decision-making process via teleoperation, leveraging human experience to help the robot adapt to the environment. Furthermore, to address the issue of excessive frequency of reference trajectory input by the human operator during teleoperation, a human-robot shared control method is studied. This method reduces the dependency of the actual trajectory on the human operator, thereby alleviating the operator's mental burden and achieving an effective combination of human intelligent decision-making capabilities and the robot's high-precision control capabilities.

The structure of the teleoperated distribution live-line working robot based on shared control proposed in this paper is shown in Figure 4. Where x_h is the trajectory point output by the teleoperation device, r_h is the reference trajectory point provided by the human operator (mapped to the Cartesian space of the manipulator), r_r is the reference trajectory point autonomously planned by the robot, serving an auxiliary role. r is the integrated reference trajectory point sent to the slave manipulator (which is referred as the Shared Trajectory in the following text), f_o is the raw data collected by the force sensor at the end of the manipulator, f_e is the contact force between the tool set at the end of the manipulator and the environment after gravity compensation and x is the current pose of the manipulator.

The teleoperated distribution live-line working robot is divided into two parts: the master side and the slave side. The master side is located on the ground, where the human operator controls the movement of the teleoperation device. By receiving the force feedback from the device rendered to the operator's hand, and information fed back by the vision system, the operator achieves immersive operation. The slave side is the robot located at an elevated height. After receiving the pose x_h from the master side, the robot performs spatial mapping to obtain r_h, and feeds the collected visual and haptic data back to the master side.

Due to the incomplete alignment of perception between the robot and the operator regarding contact forces, targets, and obstacles in the actual environment, the operator must maintain intense concentration to minimize the risks of r_h. To reduce the dependency of r on r_h and alleviate the mental burden on the operator during long-duration operations, the robot needs to autonomously plan an auxiliary trajectory r_r that is approximately the shortest path and safe based on perception information that contains real-world errors. This r_r is then linearly weighted with r_h to obtain r, which is commanded to the manipulator as the target pose.

Motion spatial mapping

The incremental method was used to calculate the trajectory increment Δx_h given by the human operator:

Δx_h = x_h - x^′_h      (1)
where x^′_h is the pose at the starting point of the teleoperation. For live-line working robots where the motion spaces of the master side and slave side have significant differences, incremental mapping allows the operator to arbitrarily choose a convenient pose and set it as the teleoperation starting pose.

The trajectory increment, Δx_h was then mapped into the Cartesian space of the slave side robot to finally obtain r_h. The mathematical expression for this mapping is as follows:

Δr_h = Δx_h * k, r_h = x^′ + Δr_h       (2)
where k is the linear mapping parameter and x' is the pose of the manipulator at the starting point of the teleoperation.

Autonomous trajectory planning
The autonomous trajectory planning process uses linear interpolation to calculate the shortest trajectory r_r from the current pose to the target pose P. The calculation process is as follows:

      (3)

where P is the target point is calculated by the vision system, which includes a binocular camera and a stereo laser LiDAR. The calculation can be performed by selecting points with the mouse and transforming the coordinates using the hand-eye calibration matrix. N is the number of interpolation points, and i is the sequence number of the interpolation point. It should be noted that under complex lighting conditions, P calculated by the vision system contains errors, which prevents r_r from completing the task independently.

Shared control method

In order to combine r_h's adaptability to the environment and r_r's efficiency, it is necessary to design a shared controller that integrates two trajectories. The shared controller uses fixed parameter arbitration weights α, and the implementation process is as follows:

r = (1 - α) * r_r + α * r_h     (4)

Abrupt operation behavior detection method

During the process of remote teleoperation, there may be erroneous behavior caused by the operator, leading to abrupt changes in the trajectory, which can even cause a collision with the environment. In order to ensure the safe and smooth movement of the manipulator on the slave side, it is necessary to detect and handle such excessive teleoperation behavior on the master side. The shared control algorithm flow, including abrupt behavior detection, is shown in Figure 5.

Teleoperation system construction

The live-line work environment is characterized by being highly unstructured and dynamic. Consequently, robots cannot rely solely on their autonomous perception and decision-making capabilities to complete tasks; they must be augmented by human intelligence. Teleoperation presents an effective technique for integrating human intervention into the robot's trajectory planning process. In such a system, the master side involves an operator controlling a teleoperation device on the ground, while the slave side at height consists of multiple subsystems.

This study details the development of a teleoperation robotic system specifically designed for distribution network operation, with the goal of performing live-line tapping tasks.

Introduction of the teleoperated robot system

The teleoperation system includes the perception subsystem, control subsystem, manipulator subsystem, tool subsystem, insulation carrier subsystem, and master side human-robot interaction devices. The detailed usage of each subsystem carried by the robot in the system is shown in Table 1, and the hardware system is shown in Figure 6.

Perception subsystem: The subsystem is responsible for providing 2D RGB images and spatial depth information to guide the robot's motion and ensure operational safety. To achieve this, it is equipped with a laser LiDAR for large-scale environmental modeling at long distances; a binocular camera for high-precision target recognition and localization at close distances; a pan-tilt-zoom (PTZ) monitoring camera to provide an operational view and feed it back to the operator throughout the entire process. Furthermore, a force sensor is integrated to perceive external forces and provide force feedback to the human operator, which helps prevent system shutdowns caused by excessive contact forces between the robot end and the environment.

Manipulator subsystem: As the primary executor of tasks, this subsystem consists of two manipulators arranged in a homogeneous dual-arm configuration.

Control subsystem: As the core subsystem of the live-line working robot, it functions as the platform's computational, control, and energy supply center, responsible for dispatching operational commands to all other subsystems. To achieve insulation and safety isolation from the ground side, the power module utilizes an independent power supply on the elevated platform. It employs a 48V DC supply for both power and drive control, which reduces inversion losses and consequently minimizes the volume and weight of the power system.

End-effector subsystem: Specialized end-effectors (tools) are mounted on the end of the manipulator to execute specific operations. A series of specialized tools with standard interfaces has been developed for different tasks. For distribution network operations, these include a stripping tool and a crimping tool (as shown in Figure 7), where the stripping tool uses three 80 W, 450 nm wavelength laser generators with a focal spot size of 130 × 180 µm, modulated via 5 V PWM. The average operation time for stripping a 120 mm section on a distribution cable with a diameter of 18–24 mm is approximately 2 min at full rated power under ambient temperature conditions. The crimping tool is responsible for securely fastening the tap conductor wire to the main power wire with a clamp, thereby establishing electrical continuity.

Insulated carrier subsystem: This subsystem is used to transport the robot to the elevated work position. It typically consists of a crawler-type or wheeled insulated aerial work platform (also known as a bucket truck). To enable integrated control of the entire robotic system, it has digital control capabilities (as shown in Figure 8).

Master side human-robot interaction subsystem: During operation, the human operator issues commands to the robot system via a handheld tablet, PC, teleoperation handle, and other computing devices. These devices reproduce the sensor data, which includes vision and contact force fed back from the robot, allowing the operator to monitor the robot's status in real-time, and also map the motion of the operator's hand to the end of the manipulator. The communication links between the subsystems are illustrated in Figure 9.

Teleoperation devices

In this system, the master and slave devices for teleoperation are the haptic device and the manipulator, respectively. The Denavit-Hartenberg (DH) parameters for each will be introduced.

As shown in Figures 10A,B, these are the link coordinate frames of the manipulator. For the six revolute joints of the manipulator (L₁,L₂,L₃,L₄,L₅,L₆), a coordinate frame is established for each joint according to the right-hand rule and the principles of the Denavit-Hartenberg (DH) convention.

In Figure 10B, x₁∼x₆, y₁∼y6, z₁∼z₆, represent the x,y,z axes of each joint's coordinate frame, respectively. Based on these established link frames, the forward kinematics model of the manipulator can be created. The DH notation uses four parameters: a_i, α_i, d_i and θ_i. 

These parameters are defined as follows, a_i: represents the distance from moving the z-axis of the i-th rod along its x-axis to the z-axis of the (i+1)-th rod; α_i: refers to the angle of rotating the z-axis of the i-th rod around its x-axis to the z-axis of the (i+1)-th rod; d_i: represents the distance from moving the x-axis of the i-th rod to its x-axis along the z-axis of the i-th rod;  θ_i: refers to the angle of rotating the x-axis of the i-th rod to its x-axis around the z-axis of the i-th rod.

The kinematic model is established using the DH parameter method, and the DH parameters for the manipulator are shown in Table 2 below.

The teleoperation handle is the haptic device used as the master manipulator, whose joints are configured in a serial chain. The joint coordinate frames are shown in Figure 10C; the modeling method is the same as that of the slave manipulator.

Experiments design

To quantitatively validate the proposed shared control algorithm before field deployment, we conduct two laboratory experiments on a 6-DOF manipulator teleoperation platform operating at 50 Hz, which was achieved by a Python script running on a standard PC. At each control cycle, the script reads the master-side pose, computes the shared trajectory, and sends the target Cartesian pose to the manipulator. A fixed interval of 20 ms is enforced between consecutive commands, yielding an update rate of 50 Hz, which is the same as the laboratory platform. The laboratory platform shares the same control architecture and shared control algorithm (Equation. 4) as the live-line working robot described in the teleoperation system construction, differing only in manipulator kinematics and workspace scale. A simulated tapping task is used: in the presence of obstacles (a water bottle and an aluminum column), the operator controls the manipulator end-effector from a start position A to a target position B, mimicking the wire-tapping operation, which is shown in Figure 11. To ensure the exact repeatability of the laboratory experiments, the physical obstacles and the target endpoint were precisely positioned within the slave manipulator's base coordinate system. Two primary obstacles were installed in the workspace: a plastic water bottle (base dimensions: 8 cm × 8 cm, height: 25 cm) placed with its bottom center at coordinates (X = 55 cm, Y = 5 cm), and an aluminum profile (base dimensions: 2 cm × 2 cm, height > 50 cm) placed with its bottom center at coordinates (X = 35 cm, Y = -30 cm). The target endpoint for the teleoperation task was set at coordinates (X = 50 cm, Y = -17 cm, Z = 26 cm). Subsequently, the validated algorithm is deployed on the live-line working robot for field demonstration.

The experimental procedure is conducted through a continuous, integrated workflow. First, during system initialization, both the slave and master manipulators are powered on, and the control software is launched to verify stable bilateral communication at a frequency of 50 Hz. To prevent sudden control jumps at the onset of teleoperation, a master-slave alignment phase is performed prior to active teleoperation. In this phase, the operator manually guides the master manipulator until its end-effector Cartesian position (defined in the master's base frame) lies within 30 mm of the slave manipulator's current end-effector Cartesian position (defined in the slave's base frame). These Cartesian positions are obtained directly from the built-in controller feedback of each manipulator, while the control software monitors their Euclidean distance in real time. Once this alignment succeeds, the current slave position is recorded as the start position, denoted as position A. Subsequently, target specification is carried out by having the operator manually guide the slave end-effector to the desired goal location to record its Cartesian coordinates, which are designated as target position B. The autonomous planner then generates a linear reference trajectory extending from the start position toward position B. Following this, the operator initiates the shared control teleoperation task. During active execution, the system operates at a 20 ms control cycle, continuously reading the master position as input, performing an online check to detect any abrupt operator behaviors, computing the shared control trajectory as output (as formulated in Equation 4), and issuing corresponding commands to the slave manipulator. As a safety measure, if an abrupt movement is detected, the slave manipulator immediately freezes and holds its last valid position in a dedicated position-hold mode; to recover from this state and resume shared control teleoperation automatically, the operator must manually move the master manipulator back until the master-slave end-effector Euclidean distance falls below a 20 mm recovery threshold. Finally, the trial is completed and deemed successful when the slave end-effector reaches and remains within 5 mm of target position B.

Experiment 1: abrupt operation behavior detection
This experiment validates the detection mechanism on the laboratory platform. During teleoperation at 50 Hz, the per-frame displacement of the master end-effector is monitored. When it exceeds D_th = 6 mm/frame (instantaneous velocity 300 mm/s), the system enters FROZEN state: the slave manipulator ceases tracking and holds its last valid configuration. The slave remains frozen until the operator returns the master to within 20 mm of the frozen position.

Experiment 2: Shared control performance comparison

Three arbitration weights α=0.3, α=0.8, α=1.0 are compared on the laboratory platform, as documented in the Supplementary Video 1, Supplementary Video 2, and Supplementary Video 3. A total of 45 experimental trials were conducted, involving three operators performing five repetitions under each of the three arbitration conditions. The system performance was assessed using three key evaluation metrics. First, completion time T: Total duration from the first operator input to the slave end-effector reaching the target position B; shorter time indicates higher operational efficiency. Second, trajectory efficiency η=L_straight/L_actual, where L_straight is the Euclidean distance from the start position A to the target position B, and L_actual is the total path length traversed by the slave end-effector during the trial, η=1 indicates a perfectly efficient, direct path; lower values indicate more redundant, detour-laden motion. Third, submovements count N_sub represents the number of distinct movement segments, which is computed by counting zero-crossings in the smoothed speed profile of the master end-effector; a smaller count indicates a more continuous, confident operation with lower cognitive load on the operator. Finally, statistical significance across the three arbitration conditions is assessed using the Kruskal-Wallis H-test, with pairwise Mann-Whitney U-tests employed for post-hoc comparisons.

Experiment 3: Field deployment

Following laboratory validation, the shared control algorithm (α=0.3) and abrupt behavior detection are deployed on the live-line working robot described in the teleoperation system construction.

Results

Abrupt operation behavior detection
Figure 12 shows the time-series analysis of a representative trial on the laboratory platform, which includes two intentional sudden movements.
Both abrupt movements were detected (TPR = 100%) at 1.81 s and 5.72 s, respectively, with zero false positives (FPR = 0%) during 13.2 s of normal operation. Freeze response time: 1 frame (20 ms), measured by control loop timing.

Shared control performance...

Discussion

This study proposes a teleoperated distribution live-line working robot system based on human-robot shared control. In contrast to fully autonomous systems that struggle with dynamic outdoor lighting and unstructured backgrounds¹², and traditional master-slave tracking teleoperation that imposes high cognitive and physical loads on operators^14,15,16, our approach fuses human intent with robotic autonomous...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
Hikvision DS-2DC2D20IW-DE3	Hikvision	https://www.hikvision.com/cn/products/pdplist/161056/	pan-tilt-zoom (PTZ) monitoring camera
KUNWEI KWR75	KUNWEItech	https://www.kunweitech.com/products/866.html	force sensor
LivoxAVIA	Livox(DJI)	https://www.livoxtech.com/cn/avia	laser LiDAR
Piper	AGILEX	https://www.agilex.ai/page/690abe7b5e78cfa260412c92?mi=1&rn=PIPER+%E6%A0%87%E5%87%86%E7%89%88	master/slave manipulator for experiment validation in laboratory platform
Python	Python Software Foundation	https://www.python.org/	programming language
RealSenseD435	Intel	https://store.realsenseai.com/buy-intel-realsense-depth-camera-d435.html	binocular camera
Touch	3D Systems	https://cn.3dsystems.com/haptics-devices/touch	haptic device
Ubuntu 24.04	Canonical Ubuntu	https://ubuntu.cn/	operation system
UR10	Universal Robots	https://www.universal-robots.cn/	manipulator of the Live-line working robot