Vision-based photovoltaic panel intelligent installation robot path planning method

Abstract

The invention provides a photovoltaic panel intelligent installation robot path planning method based on vision, and relates to the field of image data processing. Comprising the following steps: collecting photovoltaic panel storage and installation area data in real time through a phase sensitive camera, and generating a holographic data stream; a continuous phase field is constructed by adopting a bidirectional space-time grid discretization algorithm based on a wave equation solver, pre-compensation parameters are generated in combination with time reverse light field prediction, a holographic phase field model is formed after fusion, and a digital elevation model is constructed; topographic relief and obstacles are avoided in global optimal path planning through a phase gradient tracking algorithm, and the path curvature and the movement speed are dynamically optimized; the phase fracture risk is detected in real time through time reverse light field prediction, and a microsecond-level correction instruction is generated to adjust the track of the robot; after installation, phase deviation is verified through differential analysis, time reverse light field prediction and reinforcement learning iterative optimization model parameters are triggered, and high-precision adaptive path planning and obstacle avoidance control are achieved.

Description

Technical Field

[0001] The present invention relates to the field of image data processing, and specifically to a vision-based path planning method for intelligent installation robots of photovoltaic panels. Background Art

[0002] With the rapid development of renewable energy, the large-scale installation demand for photovoltaic panels continues to grow. Traditional manual installation is inefficient and costly, so automated installation technology has become an important direction in the industry. Most existing photovoltaic panel installation robots rely on preset paths or simple environmental perception and are difficult to adapt to complex terrains and dynamic obstacles. There is an urgent need for a higher-precision real-time path planning method to improve installation efficiency and reliability.

[0003] In the prior art, lidar SLAM combined with three-dimensional point cloud reconstruction is a common path planning scheme for photovoltaic panel installation robots. It generates a point cloud map by scanning the environment and plans an obstacle avoidance path. Another scheme is based on the visual inertial odometer of an RGB-D camera, which realizes pose estimation and local path optimization through feature matching and depth information fusion. In addition, some studies use genetic algorithms or A* algorithms to search for the global optimal path in a rasterized map and combine the dynamic window method to achieve real-time obstacle avoidance.

[0004] The deficiencies of the prior art are as follows: The lidar scheme is vulnerable to interference in strong light environments, the processing of point cloud data is time-consuming, and it is difficult to capture subtle phase changes. The RGB-D camera is limited by lighting conditions, and the depth information is prone to distortion in the photovoltaic panel reflection scenario. The algorithm based on the grid map has insufficient high-precision terrain modeling ability and cannot effectively handle the path deviation caused by phase gradients, making it difficult to meet the high-precision requirements of photovoltaic panel installation. Summary of the Invention

[0005] (1) Technical Problems to be Solved

[0006] In view of the deficiencies of the prior art, the present invention provides a vision-based path planning method for intelligent installation robots of photovoltaic panels to solve the problems of distorted point cloud data caused by sensor interference in strong light environments, inaccurate depth information acquisition in the photovoltaic panel reflection scenario, and insufficient high-precision terrain modeling ability of the rasterized map algorithm.

[0007] (2) Technical Solutions

[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions: A vision-based path planning method for intelligent installation robots of photovoltaic panels, including: S1: Through a phase-sensitive camera deployed on the robotic arm and mobile chassis of the robot, the light intensity and phase information of the photovoltaic panel storage area and installation area are collected in real time, and the 6-degree-of-freedom pose data and time stamp of the robot are embedded into the light intensity and phase information to generate a holographic data stream with spatio-temporal tags, where an end effector is equipped at the end of the robotic arm of the robot.

[0009] S2: Based on the holographic data stream, use the bidirectional spatio-temporal grid discretization algorithm in the wave equation solver to solve the optical field wave equation to generate a continuous phase field, construct an optical field distribution map, and then perform dimensionality reduction mapping in combination with the robot motion tangent plane. At the same time, deduce the future phase distortion through time-reversed optical field prediction to generate pre-compensation parameters;

[0010] S3: Integrate the continuous phase field and the pre-compensation parameters to construct a holographic phase field model. Convert the phase information into height values through the preset phase difference and incident angle conversion relationship, construct a digital elevation model of the installation area, and identify the three-dimensional contour of the obstacle based on the diffraction ripples and mutation bands in the holographic phase field, and mark the identified obstacle area as a dynamic no-go zone;

[0011] S4: Set the installation target position of the photovoltaic panel and the phase gradient tolerance parameter, initialize the path planning algorithm in combination with the holographic phase field model, generate the global optimal path along the optimal direction of the phase gradient and transmit it to the phase gradient tracking algorithm. The phase gradient tracking algorithm controls the robot to move along the phase gradient direction through a closed-loop feedback mechanism. The phase gradient tolerance parameter is the allowed phase deviation threshold;

[0012] S5: Control the robot to move along the optical field energy flow direction through the phase gradient tracking algorithm, and dynamically optimize the path curvature and motion speed according to the size and weight of the photovoltaic panel;

[0013] S6: Based on the time-reversed optical field prediction result, detect the phase break risk on the future path in real time, dynamically adjust the phase modulation parameter to reshape the optical field distribution map, and correct the robot motion trajectory by generating a microsecond-level motion trajectory correction instruction;

[0014] S7: After the installation is completed, collect the actual phase field data through a phase-sensitive camera, perform differential analysis with the holographic phase field model. If the deviation exceeds the set phase gradient tolerance threshold, trigger the time-reversed optical field to re-plan the path and iteratively optimize the phase gradient mapping parameter and the obstacle contour recognition parameter of the holographic phase field model through the reinforcement learning algorithm.

[0015] Preferably, the phase break refers to the phase transition phenomenon in the continuous phase field caused by propagation interference or structural mutation, manifested as the sudden change and discontinuous regions in the light field distribution map; the phase break risk refers to the probability value of a phase break occurring in a certain area in the predicted path, reflecting the potential impact of environmental disturbance or path complexity on the continuity of the phase field; the phase break risk value is used to quantify the probability of a phase break occurring at a certain grid node in the predicted path; the phase break zone refers to the spatial area in the predicted continuous phase field where a break may form, a connected area formed by the expansion of phase mutation points; the break zone position refers to the four-dimensional coordinate node in the predicted path, recording the time and spatial position where there is a phase break risk; the predicted path refers to the spatial trajectory that the robot may execute on the future time axis based on the current state of the robot and the set installation target position, through time-reversed light field prediction and combined with a wave equation solver to perform four-dimensional deduction on the continuous phase field.

[0016] Preferably, the phase-sensitive camera is deployed at key positions of the robot's manipulator and mobile chassis, covering a 360-degree field of view of the photovoltaic panel storage area and installation area to ensure the acquisition of light intensity and phase information without dead angles; a micro-nano phase modulation layer composed of a piezoelectric ceramic array is integrated at the front end of the photosensitive element of the camera, and the micro-nano phase modulation layer receives the voltage control signal from the processing unit to dynamically adjust the phase offset of the incident light to cancel the environmental light interference; the camera is connected to the robot's IMU / odometer through a nanosecond-level synchronization bus to obtain the 6-degree-of-freedom pose data of the robot and the synchronous acquisition and embedding of nanosecond-level timestamps in real time; among them, the collected light intensity, phase offset, pose data, and nanosecond-level timestamps are synchronously aligned by an embedded processor to generate a holographic data stream with spatio-temporal tags; the spatio-temporal tags are differentially compressed by an entropy coding algorithm to remove redundant information between adjacent timestamps and spatial positions, and the compressed data stream is transmitted to the processing units of the wave equation solver and the phase gradient tracking algorithm through a dedicated light field channel in a low-latency manner to ensure data real-time and integrity; the 6-degree-of-freedom pose includes displacements along the X, Y, and Z axes and pitch, yaw, and roll angles.

[0017] Preferably, the wave equation solver maps the light intensity and phase information in the holographic data stream to a spatio-temporal grid, and each grid node stores the phase field value at a specific time and position; through a two-way spatio-temporal discretization algorithm, the forward time axis is used to solve the current light field distribution, and the reverse time axis is used to reverse-deduce the future phase distortion from the target end point; the principal component features of the phase field are extracted through the principal component analysis algorithm and projected onto the robot motion tangent plane to generate a two-dimensional dimensionality reduction mapping; based on the reverse time axis deduction result, the positions and compensation amounts of the phase fracture zones on the future path are calculated to generate pre-compensation parameters including the phase modulation intensity, compensation direction, and time window; the pre-compensation parameters are generated through GPU parallel computing and transmitted to the micro-nano phase modulation layer in real time. The piezoelectric ceramic unit adjusts the piezoelectric control signal according to the parameters to change the incident light phase offset and cancel the predicted phase distortion to maintain the spatio-temporal continuity of the light field. The generation formula of the pre-compensation parameters is: , where is the piezoelectric ceramic drive voltage; is the phase amount to be compensated; is the light wavelength; k is the piezoelectric ceramic sensitivity coefficient, and the specific value is 0.5μm / V.

[0018] Preferably, the continuous phase field data and the pre-compensation parameters are fused through a weighted superposition algorithm, and the weights are determined by the phase gradient stability to generate a holographic phase field model including the terrain height, obstacle contour, and phase gradient direction; based on the linear relationship between the phase difference and the incident angle, the phase difference between adjacent positions is converted into a height difference to construct a digital elevation model; by detecting the annular diffraction ripples in the phase field, analyzing the correlation between its wavelength offset and the surface roughness of the obstacle, and combining the refractive index change of the phase mutation zone, the material hardness and geometric size of the obstacle are inferred; the obstacle information is marked as a dynamic restricted area, including the coordinate range, shape boundary, and danger level. The dynamic restricted area data is input into the path planning algorithm in real time to trigger the rearrangement of path nodes and the update of the obstacle avoidance strategy; the digital elevation model is: ; where h(x,y) is the terrain height value; is the phase difference between adjacent positions; is the conversion coefficient, and the specific value is 0.1mm / rad; where the calculation of the danger level adopts a weighted scoring model, and the defined danger level R is: , where the weight coefficients w1, w2, and w3 are 0.5, 0.3, and 0.2 respectively, where material > geometry > dynamics; , , The standardized scores of each dimension ∈ [0, 1]; where the threshold for risk level division is: when 0 ≤ R < 0.4, it is a low risk, and the risk can be ignored, and only fine-tuning of the path planning is required; when 0.4 ≤ R < 0.7, it is a medium risk, and the obstacle avoidance strategy needs to be triggered to generate a detour path; when R ≥ 0.7, it is a high risk, and the movement is immediately stopped and the globally optimal path is re-planned.

[0019] Preferably, the remote monitoring center sets the installation target coordinates and the phase gradient tolerance threshold, and the phase gradient tolerance threshold is 0.05 rad, which is used to define the upper limit of the allowed phase deviation; the path planning algorithm combines the size and weight of the photovoltaic panel to initialize the search conditions, and generates the globally optimal path in the holographic phase field model along the direction of phase gradient descent; the path avoids terrain undulations and dynamic restricted areas, and is decomposed into multiple phase gradient nodes through a discretization algorithm, and each node includes the target phase value, the movement speed and the tolerance range; the phase gradient tracking algorithm obtains the current phase value of the robot in real time through a closed-loop feedback system, calculates the difference with the target node phase value, and if the difference exceeds the tolerance threshold, the tolerance threshold is 10 cm, then adjusts the angular velocity of the robotic arm joints and the moving direction of the chassis to make the robot move along the corrected phase gradient direction until it reaches the target node.

[0020] Preferably, the robotic arm dynamics model pre-defines the mass distribution, inertia matrix and torque limit of each joint, and calculates the load torque of the end effector in combination with the weight of the photovoltaic panel; during the path planning process, the path curvature radius is dynamically adjusted according to the real-time load torque to ensure that the joint torque is always lower than the safe torque threshold of 50 Nm; the distance between adjacent nodes is adjusted in real time based on the phase gradient change rate, and the distance is reduced when the gradient changes violently to improve the control accuracy, and the distance is increased when the gradient is gentle to improve the movement efficiency; the closed-loop feedback system continuously monitors the attitude of the robotic arm and the movement state of the chassis, and dynamically optimizes the path curvature and speed parameters to achieve smooth movement under high load; the formula of the robotic arm dynamics model is: , where M is the mass matrix, C is the Coriolis force term, G is the gravity term, r is the joint torque, q is the joint angle, is the angular acceleration.

[0021] Preferably, the bidirectional spatio-temporal grid discretization algorithm discretizes the optical field wave equation into a four-dimensional grid, and each grid node stores the phase value of the space coordinate and the timestamp; the four-dimensional grid is composed of three-dimensional space coordinates (X, Y, Z) and the time axis (T), and includes discretization nodes of the forward time axis and the reverse time axis; based on the dynamic forbidden zone data and the historical fracture probability, the phase fracture risk value on the future path is calculated in real time; if the risk value exceeds the preset threshold, the pre-compensation mechanism is triggered, and the piezoelectric control signal of the piezoelectric ceramic unit is adjusted through the micro-nano phase modulation layer, and a compensation phase offset is injected to offset the predicted fracture effect; at the same time, the wave equation solver generates a microsecond-level trajectory correction instruction to control the chassis and the robotic arm to cooperate to adjust the movement direction and speed, bypass the fracture area, and ensure that the actual trajectory is consistent with the global optimal path. The formula of the bidirectional spatio-temporal grid discretization equation is: ; where is the phase value of the spatio-temporal grid node (i, j) at time step n; c is the speed of light, which is related to the propagation speed of the optical field; Δt and Δx are the time and space discretization steps; it is discretized into a four-dimensional spatio-temporal grid as: , where is the optical field phase value at the spatial position (i, j, k) at the nth time step, i, j, k are the spatial grid indices, corresponding to the x, y, z coordinate axes, n is the time step index, c is the speed of light, Δt is the reverse time step of 0.01 s; Δx, Δy, Δz are the spatial grid resolutions; i, j, k are the spatial grid indices; n is the time step index; its specific function is to reverse-deduce the optical field phase distortion distribution from the target end point, predict the future fracture zone position, such as t = tcurrent + 0.5 s, where t is the time point, generally used to represent the future prediction step; tcurrent is the current time timestamp, used as a reference point for time reverse deduction; and generate pre-compensation parameters; the formula of the phase fracture risk value is: , where represents the probability value of phase fracture at the four-dimensional coordinate point; M is the number of historical fracture event clusters, the number of different types of fracture events clustered from historical data; is the weight of the mth type of event, with a value in [0, 1], and = 1; is the current observed phase value; is the model-predicted phase value; σ m is the Gaussian kernel bandwidth of the mth type of event; the dynamic update rule is to record the position, time, and phase deviation of historical fracture events; compare the current phase field with the model prediction value and calculate the Mahalanobis distance: , if Dm < 3σm, it is considered that the current point belongs to the mth type of fracture event, The covariance matrix for the characteristics of the m-th type of fracture event, which is used to reflect the distribution shape; the phase fracture risk value ranges from [0, 1]. The higher the value, the greater the probability of a phase fracture occurring; if exceeds the preset threshold of 0.5, the corresponding obstacle avoidance or pre-compensation mechanism is triggered; when ≥0.5, trajectory correction is triggered.

[0022] Preferably, after installation, the actual phase field data of the installed photovoltaic panel area is re-collected through a phase-sensitive camera, and pixel-by-pixel differential comparison is performed with the holographic phase field model. The root mean square value of the phase difference is calculated and stored in the historical error data; if the root mean square value exceeds the tolerance threshold of 10 cm, the error source is traced back through the time-reversed light field prediction equation to locate the phase gradient mapping error or obstacle recognition deviation in the model; the boundary conditions of the light field wave equation are adjusted to optimize the holographic phase field model. The boundary conditions are the initial phase distribution of the light field wave equation and the robot motion tangent plane constraint. The robot motion tangent plane is the projection plane of the local coordinate system of the end effector motion trajectory of the robotic arm, which is dynamically determined by the 6-degree-of-freedom pose data of the robot; the weight coefficient of the phase gradient tracking algorithm is optimized through a reinforcement learning algorithm. The reinforcement learning algorithm is based on the historical error data set, iteratively updates the weight coefficient of the phase gradient tracking algorithm, adjusts the path search priority and tolerance allocation strategy, and gradually reduces the phase deviation of subsequent tasks; the rule for the reinforcement learning algorithm to update the weight is:

[0023]

[0024] where is the state and action value function, s is the current state, and a is the current action; is the learning rate, defaulting to 0.01; is the next action state, is the next possible action, is the discount factor; r is the reward function, which is negatively correlated with the phase deviation.

[0025] Preferably, the piezoelectric ceramic unit of the micro-nano phase modulation layer receives the piezoelectric control signal corresponding to the pre-compensation parameter. The voltage intensity is linearly related to the phase compensation amount, and the optical path length is changed through ceramic deformation to achieve precise adjustment of the incident light phase offset; when the camera collects the light intensity and phase data, it synchronously receives the 6-degree-of-freedom pose data of the IMU / odometer, and ensures the precise alignment of the spatio-temporal tags through the hardware clock synchronization mechanism; the compressed holographic data stream is transmitted through a dedicated channel with a fixed priority to avoid data packet loss or delay, supporting the efficient operation of subsequent algorithms.

[0026] Preferably, the digital elevation model marks the terrain undulation by detecting the area with a steep increase in phase gradient, calculates the height value by combining the conversion relationship between the phase difference and the incident angle, and generates a high-precision three-dimensional topographic map; analyzes the wavelength offset of the circular diffraction ripples in the phase field in real time, infers the surface roughness of the obstacle, and determines the material properties by the refractive index change in the phase mutation zone; the obstacle information is encapsulated as dynamic restricted area data, including coordinate boundaries, shape parameters, and danger levels, and is transmitted to the path planning algorithm in real time to dynamically update the restricted area list, ensuring that the robot's movement path always avoids the newly identified dangerous areas.

[0027] Preferably, the phase gradient tracking algorithm continuously monitors the deviation between the current phase value and the target node through a closed-loop feedback system. If the deviation exceeds the limit, it triggers the coordinated adjustment of the robotic arm joint angle and the chassis speed, and the adjustment amount is determined by the deviation ratio; during the post-installation verification phase, if the deviation between the actual phase field and the model continuously exceeds the limit, the reinforcement learning algorithm optimizes the weight coefficient of the phase gradient tracking algorithm based on historical error data, and adjusts the gradient weight and tolerance allocation ratio of the path search; through multiple rounds of iterative training, the model gradually adapts to the dynamic changes of the environment, improving the robustness and installation accuracy of the path planning.

[0028] Preferably, the path planning method uses a two-way spatio-temporal grid discretization algorithm to decompose the optical field wave equation into discrete grid nodes on the forward time axis and the reverse time axis; the forward time axis is used to solve the optical field phase distribution at the current moment, and the reverse time axis reversely deduces the phase distortion distribution on the future path based on the target end point coordinates, predicting the position, size, and distortion amount of the phase fracture zone; the reverse deduction process is accelerated through GPU parallel computing to generate pre-compensation parameters including phase modulation intensity, compensation direction, and time window; the parameters are transmitted to the micro-nano phase modulation layer of the phase-sensitive camera through a dedicated data channel. This layer is composed of a piezoelectric ceramic unit array, and each unit receives a piezoelectric control signal corresponding to the compensation amount, and changes the optical path length of the incident light through the deformation of the piezoelectric ceramic to dynamically adjust the phase offset; this mechanism realizes the real-time cancellation of future phase distortion, ensures the spatio-temporal continuity of optical field data acquisition, and avoids the lag problem of traditional single-time-axis prediction, significantly improving the accuracy and real-time performance of path planning.

[0029] (III) Beneficial Effects

[0030] The present invention provides a path planning method for a vision-based intelligent photovoltaic panel installation robot. It has the following beneficial effects:

[0031] 1. Real-time collect the light intensity and phase information through a phase-sensitive camera, and generate a high-precision holographic phase field model by combining a bidirectional spatio-temporal grid discretization algorithm; use closed-loop feedback and pre-compensation parameters to dynamically adjust the path, improving the path planning accuracy and real-time performance in complex environments; enhance the safety and stability during the installation process through phase gradient tolerance threshold constraints and dynamic no-go zone avoidance.

[0032] 2. Predict and deduce future phase distortions based on time-reversed light field, and real-time correct the motion trajectory to offset the risk of phase breakage; fuse the reinforcement learning algorithm to iteratively optimize the model parameters, combine dynamic obstacle contour recognition and terrain height mapping to improve the system's adaptability and robustness; ensure the motion efficiency of the robotic arm and equipment safety through microsecond-level trajectory correction instructions and load torque constraints. Specific implementation manners

[0033] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0034] The embodiments of the present invention provide a vision-based path planning method for an intelligent photovoltaic panel installation robot, including: the robot mobile chassis and the robotic arm cooperate to enter the photovoltaic panel installation area, and multiple phase-sensitive cameras deployed thereon are synchronously started to cover a 360-degree field of view to collect light intensity and phase information; the micro-nano phase modulation layer of the camera receives the piezoelectric control signal corresponding to the pre-compensation parameter through the piezoelectric ceramic array, dynamically adjusts the incident light phase offset amount, and offsets the environmental interference; the camera and the robot IMU / odometer are connected through a nanosecond-level synchronization bus to collect 6-degree-of-freedom pose data in real time, and are strictly aligned with the light intensity, phase offset amount, and nanosecond-level time stamp through an embedded processor to generate a holographic data stream with spatio-temporal tags; the spatio-temporal tags are differentially compressed through an entropy coding algorithm to remove redundant data between adjacent times and positions, and the compressed data stream is transmitted to the wave equation solver through a dedicated light field channel at a fixed frame rate to ensure high-real-time input.

[0035] After receiving the holographic data stream, the wave equation solver uses a bidirectional space-time grid discretization algorithm to decompose the light field wave equation into four-dimensional grid nodes. The forward time axis is used to solve the current light field phase distribution, and the reverse time axis reversely deduce the future phase distortion from the path planning target end point; the principal component characteristics of the phase field are extracted by the principal component analysis algorithm, and projected onto the robot motion tangent plane to form a two-dimensional dimensionality reduction mapping, thereby simplifying subsequent path planning calculations; GPU parallel computing accelerates the phase distortion prediction of the reverse time axis, and generates pre-compensation parameters including the fracture zone position, compensation amount and phase modulation parameters; the parameters are transmitted to the micro-nano phase modulation layer in real time through a dedicated channel, and the piezoelectric ceramic unit adjusts the deformation amount according to the piezoelectric control signal intensity, accurately corrects the incident light phase offset, offsets the predicted phase fracture interference, and ensures the spatiotemporal continuity of the light field data.

[0036] The continuous phase field data and pre-compensation parameters are fused through a weighted superposition algorithm, and the weights are dynamically allocated by the phase gradient stability to generate a holographic phase field model including terrain height, obstacle contour and phase gradient direction; based on the preset linear conversion relationship between phase difference and incident angle, the phase difference of adjacent areas is mapped to height values, and a digital elevation model of the installation area is constructed pixel by pixel; at the same time, the annular diffraction ripples in the phase field are detected, and the correlation between their wavelength offset and the surface roughness of the obstacle is analyzed, and the three-dimensional contour and material properties of the obstacle are inverted in combination with the refractive index change of the phase mutation band; the obstacle information is marked as a dynamic restricted area, including coordinate boundaries, shape parameters and hazard levels, and transmitted to the path planning algorithm in real time to trigger the update of the obstacle avoidance strategy.

[0037] The target coordinates of photovoltaic panel installation and the phase gradient tolerance threshold are set through the remote monitoring center. The path planning algorithm combines the size and weight parameters of the photovoltaic panels to generate a global optimal path along the direction of the fastest descent of the phase gradient in the holographic phase field model. The path needs to avoid the terrain undulations and dynamic restricted areas in the digital elevation model, and is decomposed into multiple phase gradient nodes through a discretization algorithm. Each node contains a target phase value, a movement speed, and an allowable deviation range. The phase gradient tracking algorithm obtains the current phase value of the robot in real time through a closed-loop feedback system, and calculates the difference with the phase value of the target node. If the deviation exceeds the tolerance threshold, the angular velocity of the robot arm joint and the movement direction of the chassis are adjusted to make the robot move along the corrected phase gradient direction.

[0038] The dynamic model of the robotic arm predefines the mass, inertia matrix, and torque limit of each joint, and calculates the real-time load torque of the end effector in combination with the weight of the photovoltaic panel. During path planning, the path curvature radius is dynamically adjusted according to the load torque to ensure that the torque of each joint is always below the safety threshold. At the same time, the distance between adjacent nodes is adjusted in real time based on the phase gradient change rate, reducing the distance when the gradient changes violently to improve control accuracy, and increasing the distance when the gradient is gentle to improve motion efficiency. The closed-loop feedback system continuously monitors the attitude of the robotic arm and the motion state of the chassis, and dynamically optimizes the path curvature and speed parameters to ensure smooth motion under high load.

[0039] Then, the optical field wave equation is discretized into a four-dimensional grid by the bidirectional spatio-temporal grid discretization algorithm. Based on the dynamic forbidden zone data and the historical fracture probability, the phase fracture risk value on the future path is calculated in real time. If the risk value exceeds the preset threshold, the pre-compensation mechanism is triggered, and compensation parameters are injected by adjusting the piezoelectric control signal of the micro-nano phase modulation layer to reshape the current optical field distribution map to offset the future fracture risk. At the same time, the wave equation solver generates a microsecond-level motion trajectory correction instruction to control the chassis and the robotic arm to jointly adjust the motion direction and speed to bypass the fracture area and ensure that the actual trajectory is consistent with the global optimal path.

[0040] After installation, the phase-sensitive camera re-collects the actual phase field data and performs a pixel-by-pixel differential comparison with the holographic phase field model to calculate the root mean square value of the phase difference. If the deviation exceeds the tolerance threshold, the error source is traced through the time-reversed optical field prediction equation to locate the phase gradient mapping error or obstacle recognition deviation, and the initial phase distribution of the wave equation and the motion tangent plane constraint parameters are adjusted. The reinforcement learning algorithm iteratively optimizes the weight coefficients of the phase gradient tracking algorithm based on the historical error data set, adjusts the gradient weight and tolerance allocation ratio of the path search, and gradually reduces the phase deviation of subsequent tasks.

[0041] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A path planning method for a vision-based intelligent installation robot of a photovoltaic panel, characterized in that, Including: S1: By means of a phase-sensitive camera deployed on the robotic manipulator and the mobile chassis, the light intensity and phase information of the photovoltaic panel storage area and the installation area are collected in real time, and the 6-degree-of-freedom pose data and time stamp of the robot are embedded into the light intensity and phase information to generate a holographic data stream with spatio-temporal tags, where an end effector is equipped at the end of the robotic manipulator; a micro-nano phase modulation layer is provided at the front end of the phase-sensitive camera, and the modulation layer dynamically adjusts the incident light phase offset by receiving a piezoelectric control signal; S2: Based on the holographic data stream, the bidirectional spatio-temporal grid discretization algorithm is used in the wave equation solver to solve the light field wave equation to generate a continuous phase field, and an optical field distribution map is constructed. Then, combined with the robot motion tangent plane, dimensionality reduction mapping is performed, and at the same time, future phase distortion is predicted and deduced through time-reversed light field to generate pre-compensation parameters; S3: The continuous phase field and the pre-compensation parameters are fused to construct a holographic phase field model. The phase information is converted into height values through a preset phase difference and incident angle conversion relationship, a digital elevation model of the installation area is constructed, and the three-dimensional contour of the obstacle is identified based on the diffraction ripples and mutation bands in the holographic phase field, and the identified obstacle area is marked as a dynamic restricted area; S4: Set the installation target position of the photovoltaic panel and the phase gradient tolerance parameter, initialize the path planning algorithm in combination with the holographic phase field model, generate a global optimal path along the optimal direction of the phase gradient and transmit it to the phase gradient tracking algorithm. The phase gradient tracking algorithm controls the robot to move along the phase gradient direction through a closed-loop feedback mechanism, and the phase gradient tolerance parameter is the allowable phase deviation threshold; S5: Control the robot to move along the light field energy flow direction through the phase gradient tracking algorithm, and dynamically optimize the path curvature and motion speed according to the size and weight of the photovoltaic panel; S6: Based on the time-reversed light field prediction result, the phase break risk on the future path is detected in real time, the phase modulation parameter is dynamically adjusted to reshape the optical field distribution map, and the robot motion trajectory is corrected by generating a microsecond-level motion trajectory correction instruction; S7: After the installation is completed, the actual phase field data is collected through the phase-sensitive camera and subjected to differential analysis with the holographic phase field model. If the deviation exceeds the set phase gradient tolerance threshold, trigger the time-reversed light field to re-plan the path and iteratively optimize the phase gradient mapping parameter and the obstacle contour recognition parameter of the holographic phase field model through the reinforcement learning algorithm.

2. The path planning method of a vision-based intelligent installation robot for photovoltaic panels according to claim 1, characterized in that: The S1 step includes the following operations: S11: Deploy multiple phase-sensitive cameras on the robotic manipulator and the mobile chassis to cover the 360° field of view of the photovoltaic panel storage area and the installation area. The front end of the photosensitive element of the phase-sensitive camera is integrated with a micro-nano phase modulation layer composed of a piezoelectric ceramic array, which can dynamically adjust the incident light phase offset according to the piezoelectric control signal and is connected to the robot IMU / odometer through a nanosecond-level synchronization bus; S12: By means of the micro-nano phase modulation layer integrated at the front end of the camera photosensitive element, dynamically adjust the phase offset of the incident light according to the pre-compensation parameters, and simultaneously record the light intensity and the phase offset, and embed the 6-degree-of-freedom pose data and the nanosecond-level time stamp provided by the robot IMU / odometer; S13: Use the entropy coding algorithm to perform differential compression on the spatio-temporal tags, reduce data transmission redundancy, and transmit the compressed holographic data stream to the processing unit through a dedicated light field data channel; The processing unit is the operating platform for the wave equation solver and the phase gradient tracking algorithm.

3. The path planning method of a vision-based intelligent installation robot for photovoltaic panels according to claim 2, wherein: The S2 step includes the following operations: S21: Input the light intensity and phase information in the holographic data stream into the wave equation solver, discretize the light field wave equation through the two-way spatio-temporal grid discretization algorithm, project the high-dimensional phase field data onto the two-dimensional motion tangent plane through the principal component analysis algorithm to complete the dimensionality reduction mapping, and project the discretized data onto the robot motion tangent plane; S22: Based on the path planning target end point, predict and deduce the phase distortion distribution through time-reversed light field prediction, generate pre-compensation parameters including the position of the fracture zone, the compensation amount, and the phase modulation parameters, and accelerate the solution process through GPU parallel computing. The phase modulation parameters are used to dynamically adjust the phase offset of the incident light through the micro-nano phase modulation layer; S23: Input the pre-compensation parameters into the micro-nano phase modulation layer of the phase-sensitive camera, adjust the current phase offset according to the pre-compensation parameters through the piezoelectric control signal, and cancel the interference of future light field phase fractures on the current data acquisition through time-reversed light field prediction propagation.

4. A path planning method for a vision-based intelligent photovoltaic panel installation robot according to claim 1, characterized in that: The S3 step includes the following operations: S31: Fuse the continuous phase field and the pre-compensation parameters through the weighted superposition algorithm to generate a holographic phase field model including phase gradient, obstacle contour, and terrain height information, and extract the phase gradient direction as the motion reference of the phase gradient tracking algorithm; S32: Mark the terrain undulation based on the steep increase region of the phase gradient, solve the height value based on the preset conversion relationship between the phase difference and the incident angle, and construct a digital elevation model; S33: Detect the circular diffraction ripple and the phase mutation band in the phase field, invert the three-dimensional contour and material properties of the obstacle by analyzing the wavelength offset of the circular diffraction ripple and the refractive index change characteristics of the phase mutation band, mark the obstacle area as a dynamic no-go zone, update the path avoidance strategy in real time, and encapsulate the no-go zone coordinates, shape parameters, and danger level data included in the dynamic no-go zone as structured data.

5. A path planning method for a vision-based intelligent photovoltaic panel installation robot according to claim 1, characterized in that: The S4 step includes the following operations: S41: Set the installation target coordinates and the phase gradient tolerance threshold through the remote monitoring center, and transmit the photovoltaic panel size and weight parameters to the path planning algorithm to dynamically constrain the path curvature and the motion speed; S42: Generate a globally optimal path in the holographic phase field model along the steepest descent direction of the phase gradient, avoiding terrain undulation and dynamic no-go zones; S43: Decompose the globally optimal path into discrete phase gradient nodes, and each node includes a target phase value, a motion speed, and an allowed deviation range.

6. The path planning method of a vision-based intelligent installation robot for photovoltaic panels according to claim 1, characterized in that: The S5 step includes the following operations: S51: Control the robot to move along the phase gradient direction through the phase gradient tracking algorithm. The end effector of the robot arm compares the phase gradient tolerance between the current phase value and the target node phase value in real time, and adjusts the robot arm posture through the closed-loop feedback mechanism of the phase gradient tracking algorithm; S52: Adjust the path curvature radius according to the weight of the photovoltaic panel and the robot arm dynamics model, so that the load torque of the robot arm does not exceed the preset safety torque threshold; The robot arm dynamics model is a predefined mathematical model that includes the mass, inertia, joint torque constraints, and kinematic parameters of the robot arm, and is used to calculate the motion trajectory and load torque of the end effector of the robot arm; S53: In a motion scenario, the phase gradient tracking algorithm monitors the phase gradient change rate in real time and dynamically adjusts the spacing of the phase gradient nodes at a preset ratio to improve the control accuracy.

7. A path planning method for a vision-based intelligent photovoltaic panel installation robot according to claim 1, characterized in that: The S6 step includes the following operations: S61: Discretize the optical field wave equation into a four-dimensional grid including the forward time axis and the backward time axis through the bidirectional spatio-temporal grid discretization algorithm, and calculate the phase break probability on the future path in real time based on the dynamic forbidden zone; S62: If the phase break probability exceeds the break probability threshold, inject pre-compensation parameters by adjusting the piezoelectric control signal of the micro-nano phase modulation layer to reshape the optical field distribution map to cancel future phase breaks; S63: Generate a motion trajectory correction instruction through the wave equation solver, and control the robot to collaboratively adjust the motion trajectory to avoid obstacles, so that it is consistent with the global optimal path.

8. A path planning method for a vision-based intelligent photovoltaic panel installation robot according to claim 1, characterized in that: The S7 step includes the following operations: S71: Collect the actual phase field data of the installed photovoltaic panel area through a phase-sensitive camera, calculate the root mean square value of the phase difference between it and the holographic phase field model, and store it in the historical error data; S72: If the phase difference exceeds the phase gradient tolerance threshold, deduce the error source through the time-reversed optical field prediction, and adjust the boundary conditions of the optical field wave equation to correct the phase gradient mapping relationship and obstacle contour recognition parameters of the holographic phase field model; Iteratively update the weight coefficient of the phase gradient tracking algorithm through the reinforcement learning algorithm.

Images