Modular four-foot heavy-load humanoid welding robot

Abstract

The invention provides a modular four-foot heavy-load humanoid welding robot. The modular four-foot heavy-load humanoid welding robot comprises a four-foot robot body, a double-mechanical-arm device arranged on the four-foot robot body and a sensor system. One of two mechanical arms of the double-mechanical-arm device is a cutting and welding machine provided with a cutting and welding all-in-one machine, and the other mechanical arm is an auxiliary mechanical arm provided with a clamping jaw; the four-foot robot and the double-mechanical-arm device cooperatively cut and weld, and the method comprises the steps that the sensor system scans the surrounding environment, and an initialized static grid map is established; the quadruped robot walks to the target welding area from the starting node; the auxiliary mechanical arm is used for assisting, and the cutting and welding mechanical arm is used for cutting and welding operation. According to the welding robot, through the integrated innovative design of the quadruped robot and the double mechanical arms and the cooperative control method, accurate execution of a welding task in a complex environment is achieved, and meanwhile the adaptive capacity, efficiency and stability of a robot system are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of autonomous welding, in particular to a modular quadruped heavy-load humanoid welding robot. BACKGROUND

[0002] Ship, marine engineering, offshore wind power and other marine equipment are key equipment for China's marine economy and national defense construction, exploration of marine environment and exploitation of marine resources. With the advancement of marine power construction, the demand for welding manufacturing and repair of marine equipment is increasing. However, due to the complex structure and poor service environment of marine equipment, about 80% of its manufacturing and repair currently uses manual arc welding, which is very low in efficiency and has high labor intensity. In the case of shortage of welders, the contradiction between large market demand and low production and operation efficiency is increasingly prominent, and it is urgent to use integrated welding robots to replace traditional welders.

[0003] The automatic welding in China still uses a large proportion of traditional teaching type robots with fixed guide rails, and some robots use wheel type chassis with poor passing performance, which is difficult to adapt to the irregular and rugged working environment in the process of marine equipment manufacturing and repair. In recent years, the development of quadruped robots is one of the key directions to solve this dilemma, and some scholars have studied it. However, it is difficult for such robots to be deeply integrated with welding support equipment, and external devices such as cables and communication lines seriously affect the environmental adaptability of the robot. At the same time, the welding repair of marine equipment on water structures is often combined with cutting and other processes, and the existing welding robots have single function and are difficult to be used for actual operation and emergency repair of marine equipment. SUMMARY

[0004] In order to overcome the shortcomings and deficiencies in the prior art, the purpose of the present application is to provide a modular quadruped heavy-load humanoid welding robot. The welding robot realizes accurate execution of welding tasks in complex environment through the innovative design and cooperative control method of quadruped robot and double-robot arm integration, and improves the adaptability, efficiency and stability of the robot system.

[0005] In order to achieve the above purpose, the present application realizes the technical scheme as follows: a modular quadruped heavy-load humanoid welding robot, comprising a quadruped robot and a double-robot arm device arranged on the quadruped robot, and a sensor system; two robot arms of the double-robot arm device, one of which is a cutting and welding robot arm provided with a cutting and welding all-in-one machine, and the other of which is an auxiliary robot arm provided with a gripper;

[0006] The quadruped robot and the double-robot arm device cooperatively cut and weld, and the method comprises the following steps:

[0007] S1, the sensor system scans the surrounding environment to establish an initialized static grid map;

[0008] S2, walking from the starting node to the target welding area by the quadruped robot;

[0009] S3, the sensor system detects the cutting area to be cut of the workpiece to be welded and performs accurate positioning;

[0010] The mechanical arm adjusts the cutting parameters through the large model according to the visual servo algorithm, accurately aligns the cutting area, adjusts the cutting posture, and assists the mechanical arm for cutting operation.

[0011] After cutting, the large model is used to judge the welding parameters, the auxiliary mechanical arm is used to clamp the new workpiece for filling the cutting area, and the cutting and welding mechanical arm is used for welding operation.

[0012] S4, after welding, both mechanical arms are retracted, and the quadruped robot moves to the designated area.

[0013] Preferably, the step S3 refers to: the sensor system scans the damaged area to determine the cutting starting point and boundary of the cutting area for cutting path planning; the cutting and welding mechanical arm accurately executes along the planned cutting path, the cutting and welding all-in-one machine is set to cutting mode to remove the damaged or corroded part; at the same time, the sensor system feedbacks visual data to adjust the end pose of the cutting and welding mechanical arm in real time to adjust the cutting depth.

[0014] After cutting, the cutting and welding dual-purpose all-in-one machine switches to welding mode, and the cutting and welding mechanical arm performs repair welding on the cut seam.

[0015] During cutting and welding, the two mechanical arms are precisely controlled through error compensation strategy.

[0016] Preferably, during cutting and welding, the error compensation strategy of the two mechanical arms is: obtaining real-time image feedback, calculating the spatial error e between the cutting position and the planned cutting path, or the welding position and the planned welding path, and calculating the correction value Δp.

[0017]

[0018] Wherein, J S is the visual Jacobian matrix, and λ is the control gain.

[0019] According to the correction value Δp, the position and angle of the cutting and welding all-in-one machine are dynamically adjusted.

[0020] Preferably, the step S2 refers to: based on the static grid map, taking the target welding area as the target node, using A* algorithm to calculate the optimal path of the quadruped robot from the starting node to the target node.

[0021] The quadruped robot walks along the optimal path and constructs a local grid map through an on-board sensor, and uses a D* algorithm to adjust the obstacle avoidance for the local grid map;

[0022] During the walking of the quadruped robot, the quadruped robot controls the motion trajectory through model predictive control.

[0023] Preferably, in the step S2, the A* algorithm is used to calculate the optimal path from the starting node to the target node of the quadruped robot, including the following steps:

[0024] X1, initializing the starting node and the target node, and defining OpenList and CloseList;

[0025] X2, respectively evaluating the total cost f(x) of all nodes in the OpenList: respectively calculating the actual cost g(x) of moving from the starting node to the evaluated node along the generated path and the estimated cost h(x) of moving from the evaluated node to the target node, and calculating the total cost f(x) = g(x) + h(x);

[0026] X3, selecting the evaluated node with the minimum total cost f(x) in the OpenList as the current node, and judging whether the current node is the target node:

[0027] If the current node is not the target node, it is checked whether the adjacent nodes of the current node are walkable and not in the CloseList; if yes, the current node is added to the node set of the generated path, the adjacent nodes are added to the OpenList, the current node is removed from the OpenList and added to the CloseList; then jump to step X2;

[0028] If the current node is the target node, the current node is put into the CloseList, the predecessor nodes of the generated path are searched in reverse, and the optimal path is generated.

[0029] Preferably, the step X1 refers to:

[0030] X11, setting the starting node and the target node;

[0031] X12, defining two lists: OpenList and CloseList; the OpenList stores the grid points to be explored, and the CloseList stores the grid points that have been explored;

[0032] X13, set two evaluation functions of A* algorithm, g(x) and h(x) respectively; wherein g(x) represents the actual cost of moving from the starting node along the generated path to the evaluation node, and h(x) represents the estimated cost from the evaluation node to the target node; set the g(x) value and h(x) value of the initial starting node as 0, and add the starting node to the OpenList; when there are obstacle grid points, add the obstacle grid points to the CloseList.

[0033] Preferably, in the step S2, the D* algorithm is used for obstacle avoidance adjustment for the local grid map, including the following steps:

[0034] Y1, path initialization, including:

[0035] Y11, set the target node, and set the path cost v(goal) of the target node as 0:

[0036] v(goal) = 0

[0037] Except for the target node, the initial path cost of all other nodes n on the local grid map is set to infinity:

[0038]

[0039] Y12, create a priority queue for storing nodes to be expanded; initialize the priority queue to contain only the target node;

[0040] Y13, perform a reverse search, starting from the target node, traverse each node n in the priority queue, calculate the shortest path from each neighbor node p to the target node, obtain the path cost v(m) of each node in the priority queue, and gradually expand the path:

[0041]

[0042] Wherein v(p) represents the optimal path cost from the target node to the neighbor node p; c(n, p) is the estimated cost from the current node n to the neighbor node p;

[0043] Y14, determine the optimal path according to the minimum value of the path cost v(n);

[0044] Y2, path execution:

[0045] The quadruped robot travels along the optimal path and moves to the next node step by step; in the moving process, the local grid map is obtained through the on-board sensor of the quadruped robot, and it is detected in real time whether there is a new obstacle in the front path; when it is detected that there is a new obstacle in any node on the front path, the path update is triggered;

[0046] Y3, path update refers to:

[0047] Y31, set the path cost v(n) of the new obstacle node n as infinity;

[0048] Y32, according to the new obstacle node, determine the affected area; rejoin all the affected area nodes to the priority queue, and calculate the path cost v(n) of each node in the priority queue;

[0049] Y33, reorder according to the new path cost;

[0050] Y34, according to the reordered new path cost, select the minimum cost path as the new optimal path, so that the quadruped robot can continue to walk around the obstacle.

[0051] Preferably, in the step S2, the model predictive control optimizes the motion trajectory according to the deviation between the current state and the target state of the welding robot, so that the welding robot can walk stably in a dynamic environment; the optimization problem of the model predictive control is:

[0052]

[0053] Wherein, x k is the current state, x d is the target state, u k is the control input, Q and R are weight matrices, representing state error and control cost.

[0054] The constraint condition is as follows:

[0055] x k+1 =f(x k ,u k )

[0056] Wherein, x k+1 is the state corresponding to the next moment of the current state x k .

[0057] Compared with the prior art, the present application has the following advantages and beneficial effects:

[0058] 1. The autonomous welding system of the quadruped robot and the dual-arm collaborative control proposed in the present application can adapt flexibly in a complex and dynamic environment, breaking through the limitations of traditional fixed base welding robots, guide rail type and wheel type welding robots that cannot complete welding tasks on uneven ground or when obstacles exist;

[0059] 2. The application uses a movable cutting and welding integrated machine; after cutting is completed, the cutting area can be repaired; wherein the auxiliary mechanical arm (left arm) is used to clamp the parts during cutting and welding, the cutting and welding mechanical arm (right arm) performs cutting and welding operation, and the double mechanical arms assist operation through cooperation, similar to the operation of human double arms, improve the flexibility of cutting and welding, realize the multifunctionalization of the four-legged double-arm welding robot, and can be widely applied to cutting and welding repair of marine equipment and water structures.

[0060] 3. The application combines A* algorithm and D* algorithm, A* algorithm completes global path planning, and D* algorithm adjusts the path in real time to avoid dynamic obstacles. Meanwhile, a model predictive control method is used to optimize the robot motion trajectory, adjust the robot state in real time, and ensure stability and safety in complex terrain. BRIEF DESCRIPTION OF DRAWINGS

[0061] Figure 1 is a structural schematic diagram of the module type four-legged heavy load humanoid welding robot of the application;

[0062] Figure 2 is a flowchart of the cutting and welding of the four-legged robot and the double-arm device of the application;

[0063] Figure 3 is a flowchart of the A* algorithm for global path planning of the application;

[0064] 1 represents a small leg of the front leg of the four-legged robot; 2 represents a large leg of the front leg of the four-legged robot; 3 represents a leg motor of the front leg of the four-legged robot; 4 represents an RGB-D depth camera; 5 represents a laser radar; 6 represents a control line; 7 represents a welding gun; 8 represents a right mechanical arm; 9 represents a controller of the right mechanical arm; 10 represents a fixed sheet metal of the mechanical arm; 11 represents a controller of the left mechanical arm; 12 represents the left mechanical arm; 13 represents a gripper; 14 represents a rotating base of the double-arm mechanical arm; 15 represents a rear leg motor of the four-legged robot; 16 represents a large leg of the rear leg of the four-legged robot; 17 represents a rear hip motor of the four-legged robot; 18 represents a cooling fan on the four-legged robot body; 19 represents a cutting and welding integrated machine; 20 represents a four-legged robot end computing platform; 21 represents a four-legged robot power supply; and 22 represents a front hip motor of the four-legged robot. DETAILED DESCRIPTION

[0065] The application will be further described in detail in combination with the drawings and specific embodiments.

[0066] EMBODIMENT

[0067] The embodiment is a modular four-legged heavy-load humanoid welding robot, mainly applied to welding operations in the manufacturing, repair and maintenance of offshore platforms, and especially suitable for welding tasks in high-altitude, rugged terrain and complex marine environments. Through the technical closed loop of "cable-free integrated design of welding system (including welding torch, welding power supply, wire feeder, etc.) - four-legged dynamic stability - intelligent welding dual-arm cooperation", the dependence of traditional welding robots on fixed environment and manual operation is broken, providing a high-adaptability, high-precision unmanned welding solution for complex industrial scenarios, solving the problems of insufficient mobility of welding robots in complex offshore platform environments, external power cable interference and poor integration of welding systems in the prior art.

[0068] As shown in Figure 1 , the welding robot comprises a four-legged robot and a dual-arm device arranged on the four-legged robot; the dual-arm device has two mechanical arms, one of which is a cutting and welding mechanical arm provided with a cutting and welding all-in-one machine 19, and the other of which is an auxiliary mechanical arm provided with a gripper 13. In the embodiment, the cutting and welding mechanical arm is the right mechanical arm 8, and the auxiliary mechanical arm is the left mechanical arm 12. The four-legged robot body serves as a mobile platform, is a mechanical dog that can adapt to complex terrain, is equipped with an omnidirectional walking design, has strong terrain adaptability, and can freely move in a complex environment; both mechanical arms are designed with six degrees of freedom, and the ends are installed with a cutting and welding all-in-one machine or a gripper, and can carry a welding torch 7 to complete tasks in multiple postures. In addition, the welding robot further comprises a sensor system, a control system, a welding system and a power supply system. The sensor system comprises an inertial measurement unit (IMU), a visual sensor (RGB-D depth camera 4), a laser radar 5 and the like, and can detect environmental information in real time during welding; the control system comprises two controllers (such as high-performance embedded computers), which are respectively used for controlling the four-legged robot and the mechanical arm, processing information of the sensor system, and executing a welding task by using a visual servo control strategy; the welding system comprises a welding torch, a cutting and welding all-in-one machine (with a built-in air pump, without an air cylinder) installed at the end of the mechanical arm, and a wire feeder; and the power supply system is a battery pack carried on the four-legged robot, which is used for integrated power supply of the welding system and the four-legged robot, so as to realize wireless cable integration design of the welding system.

[0069] The four-legged robot cooperates with the dual-arm device to cut and weld, and the method comprises the following steps, as shown in Figure 2 .

[0070] S1, environmental perception and target positioning:

[0071] The surrounding environment is scanned by sensors such as laser radar and depth camera, and an initialized static grid map is established.

[0072] The welding robot first needs to have a comprehensive perception of the surrounding environment, ensuring that it can identify and locate the welding target. To achieve this, the welding robot is equipped with a sensor system, including depth cameras, lidar, etc. The data from these sensors will work together to generate an environmental map and identify and locate target features such as welding surfaces or work area markers.

[0073] During the environmental perception process, the lidar is responsible for scanning the surrounding obstacles and terrain features, generating a three-dimensional map to provide basic data for path planning. At the same time, the depth camera captures high-definition images of the welding area, identifying the welding position through image processing algorithms. The IMU sensor built into the depth camera is used to measure the robot's own attitude changes in real time, ensuring stability during movement and avoiding control errors caused by attitude deviations.

[0074] In this invention, environmental recognition is optimized through multi-sensor fusion technology. The lidar provides a rough structure of the environment, the depth camera improves the precise positioning information of the welding target through visual algorithms, and the IMU is used to optimize the attitude adjustment of the welding robot, thereby achieving highly reliable and accurate environmental perception.

[0075] In terms of specific implementation, the welding robot uses the SLAM (Simultaneous Localization and Mapping) algorithm for real-time map construction and positioning through image matching and target recognition algorithms. In this process, the welding robot scans the surrounding environment through the lidar and combines the data from the depth camera to identify the cutting, welding position, and other important environmental features of the target area, such as obstacles that need to be bypassed.

[0076] S2, path planning and autonomous navigation:

[0077] After the quadruped robot completes environmental perception and determines the target position, the path planning algorithm comes into play. This algorithm needs to make reasonable path choices based on the target position, surrounding obstacles, and terrain features to ensure that the robot can stably and safely reach the target area for welding.

[0078] This invention is based on a static grid map, with the target welding area as the target node, and uses the A* algorithm to calculate the optimal path for the quadruped robot from the starting node to the target node; the quadruped robot walks along the optimal path and constructs a local grid map through on-board sensors, and uses the D* algorithm to adjust obstacles for the local grid map.

[0079] Specifically, the A* algorithm is used to calculate the optimal path from the starting node of the quadruped robot to the target node, including the following steps:

[0080] X1, initialization:

[0081] X11. Set the starting node and target node. The starting node and target node are identified according to the sensors carried by the quadruped robot.

[0082] X12. Define two lists: OpenList and CloseList; OpenList stores the grid points to be explored, and CloseList stores the grid points that have been explored;

[0083] X13. Set two evaluation functions of the A* algorithm, namely g(x) and h(x); where g(x) represents the actual cost of moving from the starting node along the generated path to the evaluation node, and h(x) represents the estimated cost from the evaluation node to the target node; set the g(x) value and h(x) value of the initialized starting node to 0, and add the starting node to the OpenList; when there is an obstacle grid point, add the obstacle grid point to the CloseList.

[0084] X2. Evaluate the total cost f(x) for all nodes in OpenList: calculate the actual cost g(x) of moving from the start node to the evaluation node along the generated path and the estimated cost h(x) from the evaluation node to the target node;

[0085] The estimated cost h(x) from the evaluation node to the target node is usually calculated using a heuristic function (such as Manhattan distance or Euclidean distance); Manhattan distance is used to measure the distance between two points on a plane and is calculated as:

[0086] d(i,j)=|x i -x j |+|y i -y j |

[0087] Among them, d(i,j) represents the Manhattan distance between point i and point j, (x i ,y i ) represents the two-dimensional coordinates of point i on the plane, (x j ,y j ) represents the two-dimensional coordinates of point j on the plane;

[0088] Alternatively, the Euclidean distance is used to calculate the straight-line distance between two points in a multidimensional space. If the coordinates of point A in n-dimensional space are A=(x1, x2,…, x n ), the coordinates of point B in n-dimensional space are B=(y1,y2,…,y n ), then the Euclidean distance between two points is calculated as:

[0089]

[0090] The Euclidean distance is more generally expressed as:

[0091] d(A, B) = ||A - B||2

[0092] where ||·||2 represents the 2-norm, i.e., the Euclidean distance between two vectors;

[0093] Then, the total cost f(x) = g(x) + h(x) is calculated;

[0094] X3, select the evaluation node with the minimum total cost f(x) in the OpenList as the current node, and determine whether the current node is the target node:

[0095] If the current node is not the target node, check whether the adjacent nodes of the current node are walkable and not in the CloseList; if so, add the current node to the set of nodes of the generated path, add the adjacent nodes to the OpenList, remove the current node from the OpenList, and add it to the CloseList; the nodes in the CloseList represent nodes that have been explored and will not be checked again; then jump to step X2;

[0096] If the current node is the target node, add the current node to the CloseList, search the predecessor nodes of the generated path in reverse, and generate the optimal path.

[0097] If the surrounding environment changes during the process of the quadruped robot from the initial position to the target welding position, the D* algorithm is used for adjustment, the core of which is to efficiently update the path in a dynamic environment without having to recalculate the entire path every time the environment changes. The D* algorithm is used for obstacle avoidance adjustment for the local grid map, including the following steps:

[0098] Y1, path initialization, including:

[0099] Y11, set the target node, and set the path cost of the target node to 0:

[0100] v(goal) = 0

[0101] Except for the target node, the initial path cost of all other nodes n on the local grid map is set to infinity:

[0102]

[0103] Y12, create a priority queue for storing nodes to be expanded; initialize the priority queue to contain only the target node;

[0104] Y13, perform a reverse search, starting from the target node, traverse each node n in the priority queue, calculate the shortest path from each neighbor node p to the target node, get the path cost v of each node in the priority queue, and gradually expand the path:

[0105]

[0106] where v(p) represents the optimal path cost from the target node to the neighbor node p; c(n, p) is the estimated cost from the current node n to the neighbor node p; v(p) and c(n, p) are calculated in the same way as the heuristic function in X2 (such as Manhattan distance or Euclidean distance);

[0107] Y14, determine the optimal path according to the minimum value of the path cost v(n);

[0108] Y2, path execution:

[0109] The quadruped robot moves along the optimal path and moves to the next node step by step; during the movement, the local grid map is obtained through the on-board sensor of the quadruped robot, the surrounding environment changes are monitored, and the presence of new obstacles in the front path is detected in real time; when any node on the front path is detected to have a new obstacle, the path update is triggered;

[0110] Y3, path update refers to:

[0111] Y31, set the path cost v(n) of the new obstacle node n to infinity;

[0112] Y32, determine the affected area according to the new obstacle node; re-add all affected area nodes to the priority queue and calculate the path cost v(n) of each node in the priority queue; this process is only performed locally and does not recalculate the entire path;

[0113] Y33, reorder according to the new path cost;

[0114] Y34, according to the reordered new path cost, select the minimum cost path as the new optimal path, so that the quadruped robot can bypass the obstacle and continue to move.

[0115] Y4, termination condition:

[0116] Y41, if the quadruped robot reaches the target node, the task is completed, and the path planning is terminated;

[0117] Y42, if all possible paths are blocked, the quadruped robot stops and reports the no solution state.

[0118] The principle of the D* algorithm is that if there is a change in the local grid map, the local path needs to be updated. At this time, a priority queue is created, and a priority queue structure is introduced to manage the path nodes to be updated. The affected nodes in the local grid map are added to the priority queue, and the cost from the target node to the nodes in the priority queue is calculated. The node with the minimum path cost is selected from the priority queue for expansion each time until the obstacle is avoided and the minimum cost path is selected as the new optimal path in the local area.

[0119] During the walking process of the quadruped robot, the quadruped robot controls the motion trajectory through model predictive control (MPC) to ensure that the robot is stable in movement and avoids falling.

[0120] Gait control of the quadruped robot is the key to precise navigation. In complex terrain, the robot needs to adjust the pace and posture according to real-time feedback to ensure stable walking. Model predictive control optimizes the motion trajectory according to the deviation between the current state (including position, velocity, and attitude) of the welding robot and the target state, so that the welding robot can walk stably in a dynamic environment; the optimization problem of model predictive control is:

[0121]

[0122] where x k is the current state, x d is the target state, u k is the control input, Q and R are weight matrices representing state error and control cost.

[0123] The constraint conditions are as follows:

[0124] x k+1 = f(x k , u k )

[0125] where x k+1 is the state corresponding to the next time of the current state x k .

[0126] This method can make real-time adjustments when the robot gait is disturbed by external disturbances, making the welding process more stable.

[0127] S3, precise control during welding:

[0128] The camera detects the cutting area required for processing the workpiece to be welded and performs precise positioning;

[0129] The mechanical arm aligns the cutting area accurately and adjusts the cutting posture according to the visual servo algorithm and the cutting parameter determined by the large model, and the auxiliary mechanical arm is used for assistance. The cutting and welding mechanical arm is used for cutting operation.

[0130] After cutting, the large model determines the welding parameters, the auxiliary mechanical arm assists in clamping the new workpiece for filling the cutting area, and the cutting and welding mechanical arm performs welding operation.

[0131] Step S3 refers to:

[0132] S31, cutting process:

[0133] The depth camera monitors the position of the welding point in real time, and adjusts the position and angle of the welding gun according to the image feedback. During the cutting process, the two mechanical arms first perform the cutting task to remove damaged or corroded parts. The cutting tool (such as the built-in air pump plasma cutting machine) is a cutting and welding all-in-one machine in cutting mode; at this time, the target positioning data provided by the environmental perception system is used to accurately determine the position, shape and depth of the cutting area. In this process, the dual mechanical arms adjust the cutting angle, pressure and speed in real time through the large model to ensure the flatness and accuracy of the cutting surface.

[0134] Specifically, the welding robot has a sensor system that first senses the environment, and the target positioning data provided by the environmental perception system accurately determines the position, shape and depth of the cutting area

[0135] Scan the damaged area to determine the cutting starting point and boundary of the cutting area to plan the cutting path; the cutting and welding mechanical arm accurately executes along the planned cutting path, the cutting and welding all-in-one machine is set to cutting mode to remove damaged or corroded parts, and the end pose of the cutting and welding mechanical arm is adjusted in real time through the sensor system visual feedback data to adjust the cutting depth to avoid cutting too deep or too shallow; during this process, the auxiliary mechanical arm clamps the cut parts; the whole cutting process is executed by the mechanical arm autonomously, and the cutting state is detected in real time through the sensor to ensure the accuracy of the operation;

[0136] S32, welding process:

[0137] After cutting, the cutting and welding dual-purpose all-in-one machine switches to welding mode, plans the welding path, and the cutting and welding mechanical arm performs repair welding on the cut seam.

[0138] Before cutting and before welding, the cutting path and welding path planned for the two mechanical arms are optimized:

[0139] For the path to be optimized, the poses of each path in the path are calculated A complete joint trajectory sequence is constructed represent two mechanical arms; optimization includes:

[0140] (I) Calculate the condition number κ(J(θ)) of the Jacobian matrix J(θ) of each point in the joint trajectory sequence, analyze and identify potential singular poses of the mechanical arm; specifically:

[0141] For each joint trajectory point in the joint trajectory sequence Through singular value decomposition of the Jacobian matrix J(θ):

[0142] J(θ) = U∑V T

[0143] ∑ = diag(σ1,σ2,…,σ n )

[0144] Where U and V are orthogonal matrices; ∑ is a diagonal matrix with singular values σ1,σ2,…,σ n on the diagonal.

[0145] Calculate the condition number κ(J(θ)):

[0146]

[0147] Where σ max (J(θ)) represents the maximum singular value of the Jacobian matrix, σ min (J(θ)) represents the minimum singular value of the Jacobian matrix; Set a singularity judgment threshold κ max , if the current condition number κ(J(θ)) exceeds the singularity judgment threshold κ max , it is judged as a potential singular pose, and a speed reduction, avoidance or interpolation adjustment strategy is adopted;

[0148] (II) According to the degree of singularity, calculate the required damping factor λ in real time, and adjust the joint speed of the mechanical arm smoothly when approaching the singular point; Specifically:

[0149] Use the damped least squares method to solve the stability of the joint speed:

[0150]

[0151] Where, is the joint speed vector; is the velocity vector of the end of the mechanical arm; λ is the damping factor, and I is the unit matrix;

[0152] The damping factor λ is dynamically adjusted according to the condition number k(J(θ)):

[0153]

[0154] Where λ0 is the basic damping value, κsafe is the condition number warning threshold, and δ is the growth exponent.

[0155] (Three) Collision avoidance analysis is performed on the two mechanical arms, and the motion speed and timing of the two mechanical arms are adjusted when potential collision is detected; specifically:

[0156] The minimum spatial distance d 12 (t i ) of the two mechanical arms at each time is calculated.

[0157] When d 12 (t) < d safe at any time, d safe is the set minimum safety distance threshold, and it is determined that there is potential two-mechanical-arm interference; the motion speed and timing of the two mechanical arms are adjusted to ensure that d 12 (t i ) ≥ d safe .

[0158] During cutting and welding, the two mechanical arms are precisely controlled through an error compensation strategy; the error compensation strategy is: real-time image feedback is obtained, the spatial error e between the cutting position and the planned cutting path, or the welding position and the planned welding path is calculated, and the correction value Δp is calculated:

[0159]

[0160] Where J s is the visual Jacobian matrix, and λ is the control gain.

[0161] According to the correction value Δp, the position and angle of the cutting and welding all-in-one machine are dynamically adjusted.

[0162] S4, after welding is completed, both mechanical arms are retracted, and the quadruped robot moves to the designated area.

[0163] After the welding task is completed, it needs to be reset to the initial position. The reset process is divided into two steps: first, the mechanical arm is retracted, and then the quadruped robot moves.

[0164] After the welding task is completed, the mechanical arm returns to the initial position through the retraction action, avoiding collision with the surrounding environment. At this time, the mechanical arm needs to pass through the precise path planning and control system to ensure that the welding gun maintains a sufficient distance from other equipment or obstacles, and safely returns to the stowed position.

[0165] The reset of the quadruped robot is re-planned by the path planning algorithm, and the quadruped robot moves back to the original position autonomously. The reset process also uses MPC gait optimization control to ensure that the welding robot is stable in posture during movement, avoiding collision and tilting.

[0166] The welding robot adopts a hierarchical control architecture, specifically including a perception layer, a decision layer and an execution layer. Each layer is responsible for different tasks, and through collaborative work, complex welding tasks are completed. The specific performance is as follows:

[0167] The perception layer is mainly responsible for data acquisition and environment modeling. The integration of sensors such as IMU, depth camera, laser radar, etc. is included, and the environment map is generated through sensor data, target recognition and positioning are performed.

[0168] The decision layer is responsible for path planning and decision control. According to the information of the perception layer, the decision layer uses A* algorithm and D* algorithm for path planning, combines MPC to optimize the robot motion trajectory, and judges the cutting and welding related parameters through a large model. In the process of the quadruped robot reaching the welding area, A* is used for global path planning, and if obstacles appear during movement, local path planning needs to be performed through D* algorithm, so as to update the new path to avoid obstacles. The above steps can only ensure the arrival of the quadruped robot, so it is necessary to fuse the MPC algorithm to control the gait of the quadruped robot, so that the posture of the robot during the arrival process is more stable, and the smooth completion of the cutting and welding task is ensured.

[0169] The execution layer is responsible for executing the instructions of the decision layer, including gait control and welding operation. The path planning and welding control command (including related parameter adjustment) are transmitted to the execution unit through the robot control system, and specific action execution is performed.

[0170] The hierarchical control architecture system ensures the efficiency and stability of the quadruped welding robot when performing the welding task.

[0171] The above embodiments are the preferred embodiments of the present application, but the embodiments of the present application are not limited by the above embodiments, and any changes, modifications, substitutions, combinations and simplifications made without departing from the spirit and principles of the present application are equivalent replacement methods, which are all included in the protection scope of the present application.

Claims

1. A modular quadruped heavy-load humanoid welding robot, characterized by: It includes a quadruped robot and a dual-arm device installed on the quadruped robot, as well as a sensor system; the dual-arm device has two robotic arms, one of which is a cutting and welding robotic arm equipped with a cutting and welding integrated machine, and the other is an auxiliary robotic arm equipped with a gripper; The quadruped robot and the dual-manipulator device cooperate in cutting and welding, and the method comprises the following steps: S1, the sensor system scans the surrounding environment and establishes an initial static grid map; S2, using a quadruped robot to walk from the starting node to the target welding area; S3, the sensor system detects the area to be cut of the workpiece to be welded and accurately locates it; The robotic arm uses the visual servo algorithm to determine the cutting parameters through the large model, accurately aligns the cutting area, and adjusts the cutting posture. The auxiliary robotic arm is used for assistance, and the cutting and welding robotic arm is used for cutting operations. After the cutting is completed, the welding parameters are judged by the large model, the auxiliary robot arm assists in clamping the new workpiece to fill the cutting area, and the cutting and welding robot arm performs the welding operation; S4. After welding is completed, both robotic arms are retracted and moved to the designated area by the quadruped robot.

2. The modular quadruped heavy-load humanoid welding robot according to claim 1, characterized in that: Step S3 includes: the sensor system scans the damaged area to determine the cutting starting point and boundary of the area to be cut, thereby planning the cutting path; the cutting and welding robot arm accurately executes along the planned cutting path, and the cutting and welding integrated machine is set to cutting mode to remove the damaged or corroded part; at the same time, the sensor system visual feedback data is used to adjust the end position of the cutting and welding robot arm in real time to adjust the cutting depth; After cutting is completed, the cutting and welding dual-purpose machine switches to welding mode, and the cutting and welding robot arm repairs the cut seams; During the cutting and welding process, the two robotic arms are precisely controlled using error compensation strategies.

3. The modular quadruped heavy-load humanoid welding robot according to claim 2, characterized in that: During the cutting and welding process, the error compensation strategy of the two robotic arms is to obtain real-time image feedback, calculate the spatial error e between the cutting position and the planned cutting path, or between the welding position and the planned welding path, and calculate the correction value Δp: Among them, J s is the visual Jacobian matrix, λ is the control gain; According to the correction value Δp, the position and angle of the cutting and welding machine are dynamically adjusted.

4. The modular quadruped heavy-load humanoid welding robot according to claim 1, characterized in that: The step S2 is as follows: based on the static grid map, with the target welding area as the target node, the A* algorithm is used to calculate the optimal path of the quadruped robot from the starting node to the target node; The quadruped robot walks along the optimal path, constructs a local grid map through onboard sensors, and uses the D* algorithm to adjust the local grid map for obstacle avoidance. During the walking process of the quadruped robot, the quadruped robot controls the motion trajectory through model predictive control.

5. The modular quadruped heavy-load humanoid welding robot according to claim 4, characterized in that: In step S2, the A* algorithm is used to calculate the optimal path from the quadruped robot's starting node to the target node, including the following steps: X1. Initialize the start node and target node, and define OpenList and CloseList; X2. Evaluate the total cost f(x) for all nodes in OpenList: Calculate the actual cost g(x) of moving from the start node to the evaluation node along the generated path and the estimated cost h(x) from the evaluation node to the target node, and calculate the total cost f(x) = g(x) + h(x); X3. Select the evaluation node with the smallest total cost f(x) in OpenList as the current node, and determine whether the current node is the target node: If the current node is not the target node, check whether the adjacent nodes of the current node are traversable and not in the CloseList; If so, add the current node to the set of nodes in the generated path, add the adjacent nodes to the OpenList, remove the current node from the OpenList, and add it to the CloseList; Then jump to step X2; If the current node is the target node, the current node is placed in the CloseList, and the previous node of the generated path is searched in reverse to generate the optimal path.

6. The modular quadruped heavy-load humanoid welding robot according to claim 5, characterized in that: The step X1 refers to: X11. Set the starting node and target node; X12. Define two lists: OpenList and CloseList; OpenList stores the grid points to be explored, and CloseList stores the grid points that have been explored; X13. Set the two evaluation functions of the A* algorithm, g(x) and h(x); where g(x) represents the actual cost of moving from the starting node to the evaluation node along the generated path, and h(x) represents the estimated cost from the evaluation node to the target node; Set the g(x) value and h(x) value of the initial starting node to 0, and add the starting node to the OpenList; when there is an obstacle grid point, add the obstacle grid point to the CloseList.

7. The modular quadruped heavy-load humanoid welding robot according to claim 4, characterized in that: In step S2, the D* algorithm is used to perform obstacle avoidance adjustment on the local grid map, including the following steps: Y1. Path initialization, including: Y11. Set the target node and set the path cost v(goal) of the target node to 0: v(goal)=0 Except for the target node, the initial path cost of all other nodes n on the local grid map is set to infinity: Y12. Create a priority queue to store the nodes to be expanded; initialize the priority queue to contain only the target node; Y13. Perform a reverse search, starting from the target node, traverse each node n in the priority queue, calculate the shortest path from each neighbor node p to the target node, obtain the path cost v(n) of each node in the priority queue, and gradually expand the path: Where v(p) represents the optimal path cost from the target node to the neighbor node p; c(n,p) is the estimated cost from the current node n to the neighbor node p; Y14. Determine the optimal path based on the minimum value of the path cost v(n); Y2. Path execution: The quadruped robot moves along the optimal path, gradually moving to the next node. During the movement, the quadruped robot obtains a local grid map through its onboard sensors and detects in real time whether there are new obstacles on the path ahead. When a new obstacle is detected at any node on the path ahead, a path update is triggered. Y3. Path update means: Y31. Set the path cost v(n) of the new obstacle node n to infinity; Y32. Determine the affected area based on the new obstacle node; add all nodes in the affected area back to the priority queue and calculate the path cost v(n) of each node in the priority queue; Y33, reorder according to the new path cost; Y34. Based on the reordered new path costs, select the path with the minimum cost as the new optimal path, so that the quadruped robot can bypass obstacles and continue moving forward.

8. The modular quadruped heavy-load humanoid welding robot according to claim 4, characterized in that: In step S2, the model predictive control optimizes the motion trajectory according to the deviation between the current state and the target state of the welding robot, so that the welding robot can walk stably in a dynamic environment. The optimization problem of the model predictive control is: Among them, x k is the current state, x d is the target state, u k is the control input, Q and R are weight matrices, representing the state error and control cost; The constraints are as follows: x k+1 =f(x k ,u k ) Among them, x k+1 is the current state x k The state corresponding to the next moment.

Images