Humanoid robot path planning system based on obstacle avoidance algorithm

Abstract

The invention discloses a humanoid robot path planning system based on an obstacle avoidance algorithm, and the system comprises the steps: S1, obtaining multi-source environment data, and generating a multi-dimensional dynamic environment representation; s2, realizing real-time avoidance of dynamic obstacles through local path re-planning; s3, a path multi-objective optimization module is communicated with a self-adaptive path adjustment module, and an optimal path with global adaptability is generated through a reinforcement learning mechanism and an iterative optimization process; s4, a multi-modal data fusion module which constructs a unified environment feature model through a feature extraction and fusion algorithm; s5, based on an initial path generated by a path multi-objective optimization module, performing dynamic correction on the path by using a nonlinear programming technology, and updating spatial distribution of path points in real time; and S6, quickly searching the global path space through a topological optimization algorithm, and determining a global optimal path. The method has the advantages of strong real-time performance, high adaptability, strong path optimization capability and accurate environment perception.

Description

Technical Field

[0001] The present invention relates to the technical field of path planning, and in particular to a humanoid robot path planning system based on an obstacle avoidance algorithm. Background Art

[0002] With the rapid development of artificial intelligence technology and robotics technology, humanoid robots are increasingly being used in industries, services, and rescue. Path planning is one of the important technologies for humanoid robots to achieve autonomous mobility, and its performance has a direct impact on the efficiency and safety of robot operation. Especially in complex dynamic environments, robots need to find the optimal path while avoiding obstacles to achieve efficient task execution. However, existing technologies still have many limitations in robot path planning and cannot fully meet the needs of efficient obstacle avoidance and path optimization in dynamic environments.

[0003] In the prior art, traditional humanoid robot path planning methods are usually based on static algorithm design, mainly relying on preset rules or simple sensor feedback. These methods often show low flexibility and robustness when facing complex and changing dynamic environments. Specifically, the prior art has the following main defects:

[0004] 1. Lack of real-time perception and adaptive capabilities: The obstacle avoidance algorithms in existing technologies are usually unable to effectively handle the real-time changes of dynamic obstacles, lack the real-time perception of complex environments and the ability to adaptively adjust paths, which makes the robot prone to path planning failure or collision risks.

[0005] 2. Insufficient path optimization capabilities: Traditional methods are mostly based on simple shortest path algorithms, which lack comprehensive consideration of multiple objective factors such as path length, energy consumption, safety and time, making it difficult to ensure the practicality and global optimality of path planning results.

[0006] 3. Insufficient multimodal data fusion: The existing technology has weak ability to fuse data from multiple sensors such as lidar, visual sensors and inertial measurement units, making it difficult to build a unified environmental model, resulting in inaccurate perception and understanding of complex scenes.

[0007] 4. Low efficiency of dynamic path correction: Existing technologies for dynamic path correction often rely on simple linear programming models, lack nonlinear optimization capabilities, and cannot quickly respond to changes in dynamic obstacles.

[0008] 5. Insufficient coordination between global and local paths: In traditional path planning methods, the planning of global and local paths is often carried out independently, lacking a coordination mechanism, resulting in the path planning results being unable to fully adapt to the dynamic changes of the environment.

[0009] Therefore, how to provide a humanoid robot path planning system based on an obstacle avoidance algorithm is a problem that technical personnel in this field urgently need to solve. Summary of the invention

[0010] One purpose of the present invention is to propose a humanoid robot path planning system based on an obstacle avoidance algorithm. The present invention comprehensively utilizes real-time perception of dynamic environments, adaptive path adjustment, multi-objective optimization deep learning models, multimodal data fusion, nonlinear path correction, and global and local path collaborative planning technology, and describes in detail the specific implementation methods of the robot's efficient obstacle avoidance and optimal path generation in a dynamic environment. It has the advantages of strong real-time performance, high adaptability, strong path optimization capability, and accurate environmental perception.

[0011] A humanoid robot path planning system based on an obstacle avoidance algorithm according to an embodiment of the present invention is characterized by comprising:

[0012] S1, dynamic environment information acquisition module, acquires multi-source environment data through lidar, visual sensor and inertial measurement unit, generates multi-dimensional dynamic environment representation, and perceives the spatial position, motion trajectory and relative relationship between obstacles and the robot in real time;

[0013] S2, the adaptive path adjustment module is connected to the dynamic environment information acquisition module, processes the dynamic environment data based on the adaptive obstacle avoidance algorithm, adjusts the robot path according to the dynamic changes of obstacles, and realizes real-time avoidance of dynamic obstacles through local path replanning;

[0014] S3, the path multi-objective optimization module is connected with the adaptive path adjustment module, which includes a pre-trained multi-objective deep learning model. The multi-objective deep learning model comprehensively considers the path length, energy consumption distribution, safety constraints and time optimization, and generates an optimal path with global adaptability through a reinforcement learning mechanism and iterative optimization process;

[0015] S4, multimodal data fusion module, processes the lidar point cloud data, visual image data and inertial measurement unit data collected by the dynamic environment information acquisition module, builds a unified environmental feature model through feature extraction and fusion algorithms, optimizes the perception accuracy and environmental understanding ability of complex scenes, and provides accurate input for path planning;

[0016] S5, path dynamic correction module, based on the initial path generated by the path multi-objective optimization module, uses nonlinear programming technology to dynamically correct the path, update the spatial distribution of path points in real time, and adapt to dynamic obstacle changes and robot motion characteristics;

[0017] S6, the topological global path planning module, builds a path search framework based on the environmental topological structure modeling method, quickly searches the global path space through the topological optimization algorithm, determines the global optimal path, and collaborates with the adaptive path adjustment module to achieve dynamic interaction and comprehensive coordination between local path optimization and global path optimization.

[0018] Optionally, the S1 specifically includes:

[0019] S11. The laser radar data acquisition module uses the laser radar device to obtain the three-dimensional point cloud data in the environment in real time. Based on the spatial position and shape characteristics of the point cloud data, a dynamic point cloud data set is formed through time marking. Each point cloud data point is represented by L i :

[0020] L i =(x i ,y i ,z i ,t i ),i=1,2,…,N L ;

[0021] Among them, L i Represents a point cloud data point, x i ,y i 、z i are the three-dimensional coordinates, t i is the acquisition timestamp, N L is the total number of point cloud data points;

[0022] S12, the visual information processing module obtains the two-dimensional image sequence in the environment through the visual sensor, and extracts the category, movement direction, speed and spatial position characteristics of the obstacle by combining the target detection and tracking algorithm to generate the obstacle feature matrix V j :

[0023] V j ={T j ,P j ,S j ,D j},j=1,2,…,N V ;

[0024] Among them, T j Represents the category label of the obstacle, P j =(x j ,y j ) is the two-dimensional plane position, S j is the speed value, D j is the direction of movement, N V is the total number of obstacles identified;

[0025] S13, the inertial measurement unit data acquisition module obtains the robot's own motion state in real time, including linear velocity, angular velocity and acceleration characteristics, and forms the robot's motion state vector R k :

[0026]

[0027] in, and is the linear velocity component of the robot on the two-dimensional plane, ω k represents the angular velocity, a k represents acceleration;

[0028] S14, the multimodal data fusion module integrates the multi-source data provided by the lidar data acquisition module, the visual information processing module and the inertial measurement unit data acquisition module based on the time synchronization algorithm and the space alignment technology, eliminates the time deviation and space coordinate error between sensors, and generates a comprehensive dynamic environment feature representation;

[0029] S15. The dynamic environment modeling module combines the dynamic environment feature representation generated by the multimodal data fusion module, uses the topological modeling method to construct a dynamic environment topology map, and generates a raster map to provide a real-time environment description.

[0030] Optionally, the S2 specifically includes:

[0031] S21, the dynamic environment data processing module is connected to the dynamic environment information acquisition module to receive real-time environment data, including the laser radar point cloud data set L i , obstacle feature matrix V j and the robot motion state vector R k , forming a unified input data stream through time alignment and spatial normalization operations;

[0032] S22, obstacle motion prediction module based on obstacle feature matrix V j and the robot motion state vector R k , combining the improved kinematic model and the temporal deep learning network to predict the future motion trajectory of the obstacle and generate the trajectory prediction matrix T j :

[0033]

[0034] in, is the position coordinate of obstacle j at time t, Δt is the time interval, t 0 is the current time, t n To predict the end time;

[0035] S23, the adaptive local path planning module receives the obstacle trajectory prediction matrix Tj And the dynamic point cloud dataset L i , using the adaptive obstacle avoidance algorithm based on reinforcement learning and combining the path smoothness constraint to generate the local planning path P loc :

[0036] P loc ={(x loc ,y loc )|loc=1,2,…,N loc};

[0037] Among them, (x loc ,y loc ) represents the two-dimensional coordinates of the local path point, N loc is the total number of local path points;

[0038] S24, the dynamic path optimization module calculates the local planning path P loc Perform optimization processing, using nonlinear programming algorithms combined with dynamic obstacle characteristics to adjust the spatial distribution of the path to avoid conflicts between the path and the predicted obstacle trajectory;

[0039] S25, the path real-time update module merges the dynamically optimized local path with the global path, dynamically adjusts the path node connection order through the topology optimization algorithm, and forms the final motion path P that adapts to real-time environmental changes. final And send the path to the execution module to drive the robot to move.

[0040] Optionally, the S3 specifically includes:

[0041] S31, the path optimization target definition module receives the local path data and the global path data generated by the adaptive path adjustment module, and defines the target set of path optimization, including the path length target L opt , Energy consumption target E opt , safety constraint target S opt and time optimization target T opt , comprehensive function F of path optimization objective obj It is expressed as:

[0042] F obj =λ 1 L opt +λ 2 E opt +λ 3 S opt +λ 4 T opt ;

[0043] Among them, λ 1 , 2 , 3 , 4is the weight coefficient, satisfying λ 1 +λ 2 +λ 3 +λ 4 =1;

[0044] S32, multi-objective deep learning optimization model module uses a deep learning framework based on generative adversarial networks to process the path optimization objective function F through a pre-trained model obj And environmental feature data, generate a set of candidate paths:

[0045] P cand = {P k |k=1,2,…,N cand};

[0046] Among them, P k represents candidate path k, N cand is the total number of candidate paths;

[0047] S33: The path evaluation module receives the candidate path set P. cand , using the path optimization objective function F obj Comprehensively evaluate each candidate path, sort the candidate paths according to the multi-objective scores, and select the path with the highest comprehensive score P best As an evaluation outcome path;

[0048] S34, the path dynamic optimization module evaluates the path P output by the evaluation module best Perform dynamic optimization, dynamically adjust path parameters through reinforcement learning mechanism and real-time feedback from the environment, and generate the final optimized path P final :

[0049] P final ={(x f ,y f )|f=1,2,…,N final};

[0050] Among them, (x f ,y f ) represents the two-dimensional coordinate point of the final optimization path, N final is the total number of path points;

[0051] S35, the path output module will finally optimize the path P final The output is sent to the path update module and connected to the path execution module to provide optimized path data for robot movement.

[0052] Optionally, the S4 specifically includes:

[0053] S41, the multi-source data preprocessing module receives the laser radar point cloud data collected by the dynamic environment information collection module L, visual image data Λ V and inertial measurement unit data Λ I , for point cloud data Λ L Perform denoising, voxelization and key point extraction on the visual image data V Perform geometric correction, edge detection and texture feature extraction on the inertial measurement unit data Λ I Perform time synchronization, coordinate system conversion and abnormal data elimination to generate pre-processed feature data;

[0054] S42, the feature extraction module extracts specific features from the preprocessed data and constructs a spatial geometric feature set Φ of the point cloud data L , texture and edge feature set of visual image data Φ V and the motion feature set Φ of the inertial measurement unit data I :

[0055]

[0056] in, Represents the feature point i of the point cloud data, represents the feature point j of the visual data, k represents the motion feature point k of the inertial measurement unit data;

[0057] S43, the multimodal feature fusion module combines the geometric feature set Φ based on the feature alignment algorithm and weighted fusion method. L , texture feature set Φ V and motion feature set Φ I Fusion generates a unified environmental feature model Φ E :

[0058] Φ E =f fusion (Φ L ,Φ V ,Φ I );

[0059] Among them, f fusion represents the feature fusion function, which combines weighting strategy with correlation matching for multimodal optimization;

[0060] S44, environmental modeling module is based on the unified environmental feature model Φ E , construct dynamic environment description, including grid map Γ G And the topological structure diagram Γ T , forming a set of environment descriptions:

[0061] Γ E ={Γ G ,Γ T};

[0062] Among them, Γ E Represents the spatial and connectivity description of the environment;

[0063] S45, the data output module sets the environment description Γ E The output is sent to the path planning module to provide real-time and consistent input data for path optimization and obstacle avoidance algorithms, and continuously update the dynamic changes of the environment.

[0064] Optionally, the S5 specifically includes:

[0065] S51, the initial path receiving module receives the initial path data set UI from the path multi-objective optimization module init The initial path data is in the form of discrete path points Pi init It is expressed as:

[0066]

[0067] in, Represents the path point π i The three-dimensional coordinates of N π Indicates the total number of waypoints;

[0068] S52, the environment change detection module receives the dynamic environment description set Γ from the multimodal data fusion module env , analyze the distribution and motion characteristics of dynamic obstacles in the environment, and generate a set of environmental change parameters Δ Γ :

[0069]

[0070] in, Represents a dynamic obstacle γ j spatial location, Indicates its speed, Indicates the direction of movement, Represents its influence radius, N γ Indicates the number of dynamic obstacles;

[0071] S53, nonlinear path correction module combines initial path data Π init and the environmental change parameter set Δ Γ , construct a nonlinear optimization model, dynamically adjust the spatial position of the path points to generate a corrected path Π corr , optimize the objective function F corr for:

[0072]

[0073] in, Represents the corrected path point π′ i , It represents the weight priority of the path point, and the second item constrains the safe distance between the path and the dynamic obstacle;

[0074] S54, the path smoothing module corrects the path Π corr Optimize and smooth the path Π smooth :

[0075]

[0076] in, represents the coordinates of the smoothed path points, Represents the total number of smooth path points;

[0077] S55, the path update and output module smoothes the path Π smooth The output is sent to the path update module, which dynamically updates the path data and connects with the robot control module to execute the corrected path in real time.

[0078] Optionally, the S6 specifically includes:

[0079] S61, the environment topology modeling module receives the dynamic environment description data Γ from the multimodal data fusion module env , generate the environment topology map T through the topology modeling algorithm env , the environment topology graph consists of a node set N env and edge set E env constitute:

[0080] T env =(N env ,E env );

[0081] Among them, N env ={n envi |i=1,2,…,N env} represents the key node set in the environment, represents the set of connection paths between nodes, N env and M env are the number of nodes and edges respectively;

[0082] S62, the global path search framework module is based on the environment topology map T env , generate the candidate global path set Π through the improved heuristic search algorithm global :

[0083]

[0084] in, represents the candidate global path, P globalis the total number of candidate paths. The global path search framework dynamically adjusts the candidate path set according to real-time environmental changes;

[0085] S63: The global path optimization module optimizes the candidate global path set Π global Optimize and construct the optimization objective function F based on path length, energy consumption, safety and time constraints global , optimize the objective function F global for:

[0086] F global =λ 1 L global +λ 2 E global +λ 3 S global +λ 4 T global ;

[0087] Among them, λ 1 , 2 , 3 , 4 is the target weight coefficient, satisfying λ 1 +λ 2 +λ 3 +λ 4 =1, the global path optimization module selects the global optimal path π by solving the optimal solution globalopt ;

[0088] S64, local and global path coordination module combined with global optimal path π globalopt and the local path set Π generated by the adaptive path adjustment module local , generate the comprehensive path Π through the path fusion function final :

[0089] Π final =f merge (Π local ,π globalopt );

[0090] Among them, f merge represents the path fusion function, which makes the local path and the global path continuous and consistent;

[0091] S65, the path output module converts the comprehensive path Π final The output is sent to the path update module and connected to the robot control module to provide the final path data for robot motion control.

[0092] The beneficial effects of the present invention are:

[0093] (1) The present invention combines real-time perception of dynamic environments, adaptive path adjustment, multi-objective optimization deep learning models, and multimodal data fusion technology to achieve the robot's accurate perception of complex dynamic environments and intelligent obstacle avoidance, enabling the system to adapt to changes in obstacles in real time and significantly improving the flexibility and robustness of path planning, especially in highly dynamic and complex scenarios.

[0094] (2) The present invention uses nonlinear programming technology and real-time path correction mechanism to quickly adjust the dynamic changes of the path, and combines the global and local path collaborative planning algorithms to achieve comprehensive coordination of path optimization. This not only ensures the global optimality of the path planning results, but also improves the adaptability and real-time performance of the local path, thereby achieving efficient dynamic path adjustment.

[0095] (3) The present invention constructs an environmental topology structure and combines it with a topology optimization algorithm to quickly search the global path space. At the same time, it works in conjunction with an adaptive path adjustment module to dynamically optimize the interaction between global and local paths, effectively improving the system's path planning efficiency and execution accuracy, and reducing the calculation time and execution error of path planning.

[0096] (4) The present invention utilizes multimodal data processing technology that integrates lidar, visual sensors, and inertial measurement units to provide multidimensional dynamic modeling capabilities for the environment, enabling the robot to more comprehensively understand the spatial distribution and obstacle characteristics of complex scenes, thereby providing more accurate environmental input data for path planning and reducing the possibility of path planning failure.

[0097] (5) The present invention integrates a variety of advanced technologies such as deep learning, nonlinear optimization and topological structure search, and provides a comprehensive solution with strong real-time performance, high environmental adaptability and intelligent path planning. It can effectively solve the problems of insufficient flexibility and poor dynamic adaptability in path planning in the prior art, and provides technical support for the practical application of humanoid robots in multiple fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0098] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0099] Figure 1 This is the overall framework diagram of a humanoid robot path planning system based on an obstacle avoidance algorithm proposed in the present invention. DETAILED DESCRIPTION

[0100] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0101] refer to Figure 1 , a humanoid robot path planning system based on obstacle avoidance algorithm, comprising:

[0102] S1, dynamic environment information acquisition module, acquires multi-source environment data through lidar, visual sensor and inertial measurement unit, generates multi-dimensional dynamic environment representation, and perceives the spatial position, motion trajectory and relative relationship between obstacles and the robot in real time;

[0103] In this implementation, S1 specifically includes:

[0104] S11. The laser radar data acquisition module uses the laser radar device to obtain the three-dimensional point cloud data in the environment in real time. Based on the spatial position and shape characteristics of the point cloud data, a dynamic point cloud data set is formed through time marking. Each point cloud data point is represented by L i :

[0105] L i =(x i ,y i ,z i ,t i ),i=1,2,…,N L ;

[0106] Among them, L i Represents a point cloud data point, x i ,y i 、z i are the three-dimensional coordinates, t i is the acquisition timestamp, N L is the total number of point cloud data points;

[0107] S12, the visual information processing module obtains the two-dimensional image sequence in the environment through the visual sensor, and extracts the category, movement direction, speed and spatial position characteristics of the obstacle by combining the target detection and tracking algorithm to generate the obstacle feature matrix V j :

[0108] V j ={T j ,P j ,S j ,D j},j=1,2,…,N V ;

[0109] Among them, T j Represents the category label of the obstacle, P j =(x j ,y j ) is the two-dimensional plane position, S j is the speed value, D j is the direction of movement, N Vis the total number of obstacles identified;

[0110] S13, the inertial measurement unit data acquisition module obtains the robot's own motion state in real time, including linear velocity, angular velocity and acceleration characteristics, and forms the robot's motion state vector R k :

[0111]

[0112] in, and is the linear velocity component of the robot on the two-dimensional plane, ω k represents the angular velocity, a k represents acceleration;

[0113] S14, the multimodal data fusion module integrates the multi-source data provided by the lidar data acquisition module, the visual information processing module and the inertial measurement unit data acquisition module based on the time synchronization algorithm and the space alignment technology, eliminates the time deviation and space coordinate error between sensors, and generates a comprehensive dynamic environment feature representation;

[0114] S15. The dynamic environment modeling module combines the dynamic environment feature representation generated by the multimodal data fusion module, uses the topological modeling method to construct a dynamic environment topology map, and generates a raster map to provide a real-time environment description.

[0115] S2, the adaptive path adjustment module is connected to the dynamic environment information acquisition module, processes the dynamic environment data based on the adaptive obstacle avoidance algorithm, adjusts the robot path according to the dynamic changes of obstacles, and realizes real-time avoidance of dynamic obstacles through local path replanning;

[0116] In this implementation, S2 specifically includes:

[0117] S21, the dynamic environment data processing module is connected to the dynamic environment information acquisition module to receive real-time environment data, including the laser radar point cloud data set L i , obstacle feature matrix V j and the robot motion state vector R k , forming a unified input data stream through time alignment and spatial normalization operations;

[0118] S22, obstacle motion prediction module based on obstacle feature matrix V j and the robot motion state vector R k , combining the improved kinematic model and the temporal deep learning network to predict the future motion trajectory of the obstacle and generate the trajectory prediction matrix T j :

[0119]

[0120] in, is the position coordinate of obstacle j at time t, Δt is the time interval, t 0 is the current time, t n To predict the end time;

[0121] S23, the adaptive local path planning module receives the obstacle trajectory prediction matrix T j And the dynamic point cloud dataset L i , using the adaptive obstacle avoidance algorithm based on reinforcement learning and combining the path smoothness constraint to generate the local planning path P loc :

[0122] P loc ={(x loc ,y loc )|loc=1,2,…,N loc};

[0123] Among them, (x loc ,y loc ) represents the two-dimensional coordinates of the local path point, N loc is the total number of local path points;

[0124] S24, the dynamic path optimization module calculates the local planning path P loc Perform optimization processing, using nonlinear programming algorithms combined with dynamic obstacle characteristics to adjust the spatial distribution of the path to avoid conflicts between the path and the predicted obstacle trajectory;

[0125] S25, the path real-time update module merges the dynamically optimized local path with the global path, dynamically adjusts the path node connection order through the topology optimization algorithm, and forms the final motion path P that adapts to real-time environmental changes. final And send the path to the execution module to drive the robot to move.

[0126] S3, the path multi-objective optimization module is connected with the adaptive path adjustment module, which includes a pre-trained multi-objective deep learning model. The multi-objective deep learning model comprehensively considers the path length, energy consumption distribution, safety constraints and time optimization, and generates an optimal path with global adaptability through a reinforcement learning mechanism and iterative optimization process;

[0127] In this implementation, S3 specifically includes:

[0128] S31, the path optimization target definition module receives the local path data and the global path data generated by the adaptive path adjustment module, and defines the target set of path optimization, including the path length target L opt , Energy consumption target E opt , safety constraint target S opt and time optimization target Topt , comprehensive function F of path optimization objective obj It is expressed as:

[0129] F obj =λ 1 L opt +λ 2 E opt +λ 3 S opt +λ 4 T opt ;

[0130] Among them, λ 1 , 2 , 3 , 4 is the weight coefficient, satisfying λ 1 +λ 2 +λ 3 +λ 4 =1;

[0131] S32, multi-objective deep learning optimization model module uses a deep learning framework based on generative adversarial networks to process the path optimization objective function F through a pre-trained model obj And environmental feature data, generate a set of candidate paths:

[0132] P cand = {P k |k=1,2,…,N cand};

[0133] Among them, P k represents candidate path k, N cand is the total number of candidate paths;

[0134] S33: The path evaluation module receives the candidate path set P. cand , using the path optimization objective function F obj Comprehensively evaluate each candidate path, sort the candidate paths according to the multi-objective scores, and select the path with the highest comprehensive score P best As an evaluation outcome path;

[0135] S34, the path dynamic optimization module evaluates the path P output by the evaluation module best Perform dynamic optimization, dynamically adjust path parameters through reinforcement learning mechanism and real-time feedback from the environment, and generate the final optimized path P final :

[0136] P final ={(x f ,y f )|f=1,2,…,N final};

[0137] Among them, (x f ,y f ) represents the two-dimensional coordinate point of the final optimization path, N final is the total number of path points;

[0138] S35, the path output module will finally optimize the path P final The output is sent to the path update module and connected to the path execution module to provide optimized path data for robot movement.

[0139] S4, multimodal data fusion module, processes the lidar point cloud data, visual image data and inertial measurement unit data collected by the dynamic environment information acquisition module, builds a unified environmental feature model through feature extraction and fusion algorithms, optimizes the perception accuracy and environmental understanding ability of complex scenes, and provides accurate input for path planning;

[0140] In this implementation, S4 specifically includes:

[0141] S41, the multi-source data preprocessing module receives the laser radar point cloud data collected by the dynamic environment information collection module L , visual image data Λ V and inertial measurement unit data Λ I , for point cloud data Λ L Perform denoising, voxelization and key point extraction on the visual image data V Perform geometric correction, edge detection and texture feature extraction on the inertial measurement unit data Λ I Perform time synchronization, coordinate system conversion and abnormal data elimination to generate pre-processed feature data;

[0142] S42, the feature extraction module extracts specific features from the preprocessed data and constructs a spatial geometric feature set Φ of the point cloud data L , texture and edge feature set of visual image data Φ V and the motion feature set Φ of the inertial measurement unit data I :

[0143]

[0144] in, Represents the feature point i of the point cloud data, represents the feature point j of the visual data, k represents the motion feature point k of the inertial measurement unit data;

[0145] S43, the multimodal feature fusion module combines the geometric feature set Φ based on the feature alignment algorithm and weighted fusion method. L , texture feature set Φ V and motion feature set ΦI Fusion generates a unified environmental feature model Φ E :

[0146] Φ E =f fusion (Φ L ,Φ V ,Φ I );

[0147] Among them, f fusion represents the feature fusion function, which combines weighting strategy with correlation matching for multimodal optimization;

[0148] S44, environmental modeling module is based on the unified environmental feature model Φ E , construct dynamic environment description, including grid map Γ G And the topological structure diagram Γ T , forming a set of environment descriptions:

[0149] Γ E ={Γ G ,Γ T};

[0150] Among them, Γ E Represents the spatial and connectivity description of the environment;

[0151] S45, the data output module sets the environment description Γ E The output is sent to the path planning module to provide real-time and consistent input data for path optimization and obstacle avoidance algorithms, and continuously update the dynamic changes of the environment.

[0152] S5, path dynamic correction module, based on the initial path generated by the path multi-objective optimization module, uses nonlinear programming technology to dynamically correct the path, update the spatial distribution of path points in real time, and adapt to dynamic obstacle changes and robot motion characteristics;

[0153] In this implementation, S5 specifically includes:

[0154] S51, the initial path receiving module receives the initial path data set UI from the path multi-objective optimization module init The initial path data is in the form of discrete path points Pi init It is expressed as:

[0155]

[0156] in, Represents the path point π i The three-dimensional coordinates of N π Indicates the total number of waypoints;

[0157] S52, the environment change detection module receives the dynamic environment description set Γ from the multimodal data fusion module env , analyze the distribution and motion characteristics of dynamic obstacles in the environment, and generate a set of environmental change parameters Δ Γ :

[0158]

[0159] in, Represents a dynamic obstacle γ j spatial location, Indicates its speed, Indicates the direction of movement, Represents its influence radius, N γ Indicates the number of dynamic obstacles;

[0160] S53, nonlinear path correction module combines initial path data Π init and the environmental change parameter set Δ Γ , construct a nonlinear optimization model, dynamically adjust the spatial position of the path points to generate a corrected path Π corr , optimize the objective function F corr for:

[0161]

[0162] in, Represents the corrected path point π′ i , It represents the weight priority of the path point, and the second item constrains the safe distance between the path and the dynamic obstacle;

[0163] S54, the path smoothing module corrects the path Π corr Perform optimization processing to generate a smooth path Π smooth :

[0164]

[0165] in, represents the coordinates of the smoothed path points, Represents the total number of smooth path points;

[0166] S55, the path update and output module smoothes the path Π smooth The output is sent to the path update module, which dynamically updates the path data and connects with the robot control module to execute the corrected path in real time.

[0167] S6, the topological global path planning module, builds a path search framework based on the environmental topological structure modeling method, quickly searches the global path space through the topological optimization algorithm, determines the global optimal path, and collaborates with the adaptive path adjustment module to achieve dynamic interaction and comprehensive coordination between local path optimization and global path optimization.

[0168] In this implementation, S6 specifically includes:

[0169] S61, the environment topology modeling module receives the dynamic environment description data Γ from the multimodal data fusion module env , generate the environment topology map T through the topology modeling algorithm env , the environment topology graph consists of a node set N env and edge set E env constitute:

[0170] T env =(N env ,E env );

[0171] in, represents the set of key nodes in the environment, represents the set of connection paths between nodes, N env and M env are the number of nodes and edges respectively;

[0172] S62, the global path search framework module is based on the environment topology map T env , generate the candidate global path set Π through the improved heuristic search algorithm global :

[0173] Π global ={π globalk |k=1,2,…,P global};

[0174] Among them, π globalk represents the candidate global path, P global is the total number of candidate paths. The global path search framework dynamically adjusts the candidate path set according to real-time environmental changes;

[0175] S63: The global path optimization module optimizes the candidate global path set Π global Optimize and construct the optimization objective function F based on path length, energy consumption, safety and time constraints global , optimize the objective function F global for:

[0176] F global =λ 1 L global +λ 2 Eglobal +λ 3 S global +λ 4 T global ;

[0177] Among them, λ 1 , 2 , 3 , 4 is the target weight coefficient, satisfying λ 1 +λ 2 +λ 3 +λ 4 =1, the global path optimization module selects the global optimal path π by solving the optimal solution globalopt ;

[0178] S64, local and global path coordination module combined with global optimal path π globalopt and the local path set Π generated by the adaptive path adjustment module local , generate the comprehensive path Π through the path fusion function final :

[0179] Π final =f merge (Π local ,π globalopt );

[0180] Among them, f merge represents the path fusion function, which makes the local path and the global path continuous and consistent;

[0181] S65, the path output module converts the comprehensive path Π final The output is sent to the path update module and connected to the robot control module to provide the final path data for robot motion control.

[0182] Embodiment 1:

[0183] In order to verify the feasibility and effectiveness of the present invention, the humanoid robot path planning system and method based on the obstacle avoidance algorithm of the present invention are applied in the dynamic environment of a large logistics storage center. The internal environment of the storage center is complex, including various types of dynamic obstacles (such as forklifts, transport robots and staff), and path planning and obstacle avoidance problems have become key challenges to achieve efficient operation of robots. Traditional path planning systems are difficult to adapt to the dynamic changes of the storage environment due to their lack of flexibility and real-time performance, resulting in frequent path conflicts, delays and even stagnation of robots. For this reason, the system of the present invention is deployed and tested in this environment.

[0184] During deployment, the robot is equipped with a variety of sensors, including lidar, visual cameras, and inertial measurement units, to perceive the dynamic changes of the surrounding environment in real time. The dynamic environment information acquisition module is responsible for collecting sensor data and inputting it into the multimodal data fusion module for processing to build a unified environmental feature model. Based on the environmental feature model, the adaptive path adjustment module and the topological global path planning module work together to dynamically generate an adaptive optimal path, solving the problem of insufficient flexibility of path planning in complex environments in traditional methods.

[0185] The test scenario is the peak period of the logistics center, simulating the key issues of robot obstacle avoidance and path planning in a dynamic environment. In this scenario, obstacles in the storage area move randomly, and the robot needs to complete the cargo transportation task in a dynamic environment while avoiding collisions with obstacles. The system uses a nonlinear path correction module to optimize the path in real time, and ensures the global optimality and local adaptability of path planning through local and global path collaborative planning modules. To evaluate the system performance, the experiment recorded the key data of the robot in the storage environment before and after deployment, and the data summary is shown in Table 1.

[0186] Table 1 Comparison of path planning performance of logistics center robots

[0187]

[0188] It can be seen from the data in Table 1 that the system of the present invention significantly improves the performance of robot path planning after deployment. In terms of task completion time, the average time for each task is shortened from 12.3 minutes to 8.1 minutes, and the efficiency is improved by more than 34%. The obstacle avoidance success rate is increased from 84.2% to 98.7%, which greatly reduces the path interruption problem caused by obstacles. At the same time, the path planning failure rate is reduced from 15.6% to 1.3%, indicating that the robustness of the system in a dynamic environment is greatly improved. The response time is reduced from 540 milliseconds to 150 milliseconds, allowing the system to adapt to dynamic environmental changes more quickly. The number of tasks per unit time increased from 5.1 times to 7.8 times, and the daily operation efficiency of the robot was significantly improved. In addition, the robot collision incident was completely eliminated after the system was deployed, indicating that the safety of the system in path planning is effectively guaranteed.

[0189] During the specific test, a forklift suddenly moved into the robot's path in a certain shelf area. The adaptive path adjustment module detected the dynamic changes of the obstacle in real time and quickly adjusted the local path. At the same time, the topological global path planning module re-optimized the global path to ensure that the robot safely bypassed the obstacle and completed the task. The system uses the nonlinear path correction module to smoothly optimize the path, avoid sudden stops or excessive detours of the robot, and improve the smoothness and efficiency of the transportation task.

[0190] It can be seen from the above experimental results that the humanoid robot path planning system and method based on the obstacle avoidance algorithm of the present invention effectively solves the problems of insufficient flexibility and real-time performance of traditional path planning methods in dynamic environments. It not only significantly improves the robot's operating efficiency and obstacle avoidance success rate, but also enhances the safety and robustness of path planning through a collaborative planning mechanism, providing an efficient and intelligent solution for robot autonomous navigation in complex dynamic environments.

[0191] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A humanoid robot path planning system based on obstacle avoidance algorithm, characterized in that: include: S1, dynamic environment information acquisition module, acquires multi-source environment data through lidar, visual sensor and inertial measurement unit, generates multi-dimensional dynamic environment representation, and perceives the spatial position, motion trajectory and relative relationship between obstacles and the robot in real time; S2, the adaptive path adjustment module is connected to the dynamic environment information acquisition module, processes the dynamic environment data based on the adaptive obstacle avoidance algorithm, adjusts the robot path according to the dynamic changes of obstacles, and realizes real-time avoidance of dynamic obstacles through local path replanning; S3, the path multi-objective optimization module is connected with the adaptive path adjustment module, which includes a pre-trained multi-objective deep learning model. The multi-objective deep learning model comprehensively considers the path length, energy consumption distribution, safety constraints and time optimization, and generates an optimal path with global adaptability through a reinforcement learning mechanism and iterative optimization process; S4, multimodal data fusion module, processes the lidar point cloud data, visual image data and inertial measurement unit data collected by the dynamic environment information acquisition module, builds a unified environmental feature model through feature extraction and fusion algorithms, optimizes the perception accuracy and environmental understanding ability of complex scenes, and provides accurate input for path planning; S5, path dynamic correction module, based on the initial path generated by the path multi-objective optimization module, uses nonlinear programming technology to dynamically correct the path, update the spatial distribution of path points in real time, and adapt to dynamic obstacle changes and robot motion characteristics; S6, the topological global path planning module, builds a path search framework based on the environmental topological structure modeling method, quickly searches the global path space through the topological optimization algorithm, determines the global optimal path, and collaborates with the adaptive path adjustment module to achieve dynamic interaction and comprehensive coordination between local path optimization and global path optimization.

2. A humanoid robot path planning system based on obstacle avoidance algorithm according to claim 1, characterized in that: The S1 specifically includes: S11. The laser radar data acquisition module uses the laser radar device to obtain the three-dimensional point cloud data in the environment in real time. Based on the spatial position and shape characteristics of the point cloud data, a dynamic point cloud data set is formed through time marking. Each point cloud data point is represented by L i : L i =(x i ,y i ,z i ,t i ),i=1,2,…,N L ; Among them, L i Represents a point cloud data point, x i ,y i 、z i are the three-dimensional coordinates, t i is the acquisition timestamp, N L is the total number of point cloud data points; S12, the visual information processing module obtains the two-dimensional image sequence in the environment through the visual sensor, and extracts the category, movement direction, speed and spatial position characteristics of the obstacle by combining the target detection and tracking algorithm to generate the obstacle feature matrix V j : V j ={T j ,P j ,S j ,D j },j=1,2,…,N V ; Among them, T j Represents the category label of the obstacle, P j =(x j ,y j ) is the two-dimensional plane position, S j is the speed value, D j is the direction of movement, N V is the total number of obstacles identified; S13, the inertial measurement unit data acquisition module obtains the robot's own motion state in real time, including linear velocity, angular velocity and acceleration characteristics, and forms the robot's motion state vector R k : in, and is the linear velocity component of the robot on the two-dimensional plane, ω k represents the angular velocity, a k represents acceleration; S14, the multimodal data fusion module integrates the multi-source data provided by the lidar data acquisition module, the visual information processing module and the inertial measurement unit data acquisition module based on the time synchronization algorithm and the space alignment technology, eliminates the time deviation and space coordinate error between sensors, and generates a comprehensive dynamic environment feature representation; S15. The dynamic environment modeling module combines the dynamic environment feature representation generated by the multimodal data fusion module, uses the topological modeling method to construct a dynamic environment topology map, and generates a raster map to provide a real-time environment description.

3. A humanoid robot path planning system based on obstacle avoidance algorithm according to claim 1, characterized in that: The S2 specifically includes: S21, the dynamic environment data processing module is connected to the dynamic environment information acquisition module to receive real-time environment data, including the laser radar point cloud data set L i , obstacle feature matrix V j and the robot motion state vector R k , forming a unified input data stream through time alignment and spatial normalization operations; S22, obstacle motion prediction module based on obstacle feature matrix V j and the robot motion state vector R k , combining the improved kinematic model and the temporal deep learning network to predict the future motion trajectory of the obstacle and generate the trajectory prediction matrix T j : in, is the position coordinate of obstacle j at time t, Δt is the time interval, t0 is the current time, t n To predict the end time; S23, the adaptive local path planning module receives the obstacle trajectory prediction matrix T j And the dynamic point cloud dataset L i , using the adaptive obstacle avoidance algorithm based on reinforcement learning and combining the path smoothness constraint to generate the local planning path P loc : P loc {(x loc ,y loc )∣loc=1,2,…,N loc }; Among them, (x loc ,y loc ) represents the two-dimensional coordinates of the local path point, N loc is the total number of local path points; S24, the dynamic path optimization module calculates the local planning path P loc Perform optimization processing, using nonlinear programming algorithms combined with dynamic obstacle characteristics to adjust the spatial distribution of the path to avoid conflicts between the path and the predicted obstacle trajectory; S25, the path real-time update module merges the dynamically optimized local path with the global path, dynamically adjusts the path node connection order through the topology optimization algorithm, and forms the final motion path P that adapts to real-time environmental changes. final And send the path to the execution module to drive the robot to move.

4. A humanoid robot path planning system based on obstacle avoidance algorithm according to claim 1, characterized in that: The S3 specifically includes: S31, the path optimization target definition module receives the local path data and the global path data generated by the adaptive path adjustment module, and defines the target set of path optimization, including the path length target L opt , Energy consumption target E opt , safety constraint target S opt and time optimization target T opt , comprehensive function F of path optimization objective obj It is expressed as: F obj =λ1L opt +λ2E opt +λ3S opt +λ4T opt ; Among them, λ1, λ2, λ3, and λ4 are weight coefficients, satisfying λ1+λ2+λ3+λ4=1; S32, multi-objective deep learning optimization model module uses a deep learning framework based on generative adversarial networks to process the path optimization objective function F through a pre-trained model obj And environmental feature data, generate a set of candidate paths: P cand ={P k ∣k=1,2,…,N cand }; Among them, P k represents candidate path k, N cand is the total number of candidate paths; S33: The path evaluation module receives the candidate path set P. cand , using the path optimization objective function F obj Comprehensively evaluate each candidate path, sort the candidate paths according to the multi-objective scores, and select the path with the highest comprehensive score P best As an evaluation outcome path; S34, the path dynamic optimization module evaluates the path P output by the evaluation module best Perform dynamic optimization, dynamically adjust path parameters through reinforcement learning mechanism and real-time feedback from the environment, and generate the final optimized path P final : P final {(x f ,y f )∣f=1,2,…,N final }; Among them, (x f ,y f ) represents the two-dimensional coordinate point of the final optimization path, N final is the total number of path points; S35, the path output module will finally optimize the path P final The output is sent to the path update module and connected to the path execution module to provide optimized path data for robot movement.

5. The humanoid robot path planning system based on obstacle avoidance algorithm according to claim 1, characterized in that: The S4 specifically includes: S41, the multi-source data preprocessing module receives the laser radar point cloud data collected by the dynamic environment information collection module L , visual image data Λ V and inertial measurement unit data Λ I , for point cloud data Λ L Perform denoising, voxelization and key point extraction on the visual image data V Perform geometric correction, edge detection and texture feature extraction on the inertial measurement unit data Λ I Perform time synchronization, coordinate system conversion and abnormal data elimination to generate pre-processed feature data; S42, the feature extraction module extracts specific features from the preprocessed data and constructs a spatial geometric feature set Φ of the point cloud data L , texture and edge feature set of visual image data Φ V and the motion feature set Φ of the inertial measurement unit data I : in, Represents the feature point i of the point cloud data, represents the feature point j of the visual data, k represents the motion feature point k of the inertial measurement unit data; S43, the multimodal feature fusion module combines the geometric feature set Φ based on the feature alignment algorithm and weighted fusion method. L , texture feature set Φ V and motion feature set Φ I Fusion generates a unified environmental feature model Φ E : F E =f fusion (F L ,F V ,F I ); Among them, f fusion represents the feature fusion function, which combines weighting strategy with correlation matching for multimodal optimization; S44, environmental modeling module is based on the unified environmental feature model Φ E , construct dynamic environment description, including grid map Γ G And the topological structure diagram Γ T , forming a set of environment descriptions: C E ={Γ G ,C T }; Among them, Γ E Represents the spatial and connectivity description of the environment; S45, the data output module sets the environment description Γ E The output is sent to the path planning module to provide real-time and consistent input data for path optimization and obstacle avoidance algorithms, and continuously update the dynamic changes of the environment.

6. A humanoid robot path planning system based on obstacle avoidance algorithm according to claim 1, characterized in that: The S5 specifically includes: S51, the initial path receiving module receives the initial path data set UI from the path multi-objective optimization module init The initial path data is in the form of discrete path points Pi init It is expressed as: in, Represents the path point π i The three-dimensional coordinates of N π Indicates the total number of waypoints; S52, the environment change detection module receives the dynamic environment description set Γ from the multimodal data fusion module env , analyze the distribution and motion characteristics of dynamic obstacles in the environment, and generate a set of environmental change parameters Δ Γ : in, Represents a dynamic obstacle γ j spatial location, Indicates its speed, Indicates the direction of movement, Represents its influence radius, N γ Indicates the number of dynamic obstacles; S53, nonlinear path correction module combines initial path data Π init and the environmental change parameter set Δ Γ , construct a nonlinear optimization model, dynamically adjust the spatial position of the path points to generate a corrected path Π corr , optimize the objective function F corr for: in, Represents the corrected path point π′ i , It represents the weight priority of the path point, and the second item constrains the safe distance between the path and the dynamic obstacle; S54, the path smoothing module corrects the path Π corr Perform optimization processing to generate a smooth path Π smooth : in, represents the coordinates of the smoothed path points, Represents the total number of smooth path points; S55, the path update and output module smoothes the path Π smooth The output is sent to the path update module, which dynamically updates the path data and connects with the robot control module to execute the corrected path in real time.

7. A humanoid robot path planning system based on obstacle avoidance algorithm according to claim 1, characterized in that: The S6 specifically includes: S61, the environment topology modeling module receives the dynamic environment description data Γ from the multimodal data fusion module env , generate the environment topology map T through the topology modeling algorithm env , the environment topology graph consists of a node set N env and edge set E env constitute: T env =(N env ,E env ); in, represents the set of key nodes in the environment, represents the set of connection paths between nodes, N env and M env are the number of nodes and edges respectively; S62, the global path search framework module is based on the environment topology map T env , generate the candidate global path set Π through the improved heuristic search algorithm global : in, represents the candidate global path, P global is the total number of candidate paths. The global path search framework dynamically adjusts the candidate path set according to real-time environmental changes; S63: The global path optimization module optimizes the candidate global path set Π global Optimize and construct the optimization objective function F based on path length, energy consumption, safety and time constraints global , optimize the objective function F global for: F global =λ1L global +λ2E global +λ3S global +λ4T global ; Among them, λ1, λ2, λ3, and λ4 are target weight coefficients, satisfying λ1+λ2+λ3+λ4=1. The global path optimization module selects the global optimal path by solving the optimal solution. S64, local and global path coordination module combined with global optimal path and the local path set Π generated by the adaptive path adjustment module local , generate the comprehensive path Π through the path fusion function final : Among them, f merge represents the path fusion function, which makes the local path and the global path continuous and consistent; S65, the path output module converts the comprehensive path Π final The output is sent to the path update module and connected to the robot control module to provide the final path data for robot motion control.

Images