Intelligent path planning method and system based on complex environment

Abstract

The invention relates to the technical field of robot intelligent control, in particular to an intelligent path planning method and system based on a complex environment. Comprising the steps that through heterogeneous laser radar combination (air-ground global obstacle perception is achieved, a layered map (a static global map and a dynamic local map) is constructed, airspace safety corridor constraints are introduced based on an improved RRT * algorithm, and cooperative obstacle avoidance of double mechanical arms, a three-axis platform and an AGV is ensured. Real-time local path optimization is realized in combination with a dynamic window method (DWA), the path risk is monitored through a safety index (SI), and millisecond-level re-planning or emergency stop is triggered. The method has the advantages that the airspace obstacle detection height reaches 3 m, the path planning success rate is increased to 98.7%, the re-planning response time is smaller than or equal to 180 ms, the collision risk is remarkably reduced, and the method is suitable for autonomous navigation of complex scenes such as electric power inspection and warehouse logistics.

Description

Technical Field

[0001] This invention relates to the field of robot intelligent control technology, and in particular to an intelligent path planning method and system based on complex environments. Background Technology

[0002] The core objective of robot path planning is to search for a safe, collision-free path from the starting point to the ending point in an environment with obstacles, while satisfying specific optimization criteria (shortest distance, lowest energy consumption). Traditional algorithms (such as artificial potential field method, DWA, RRT) perform well in simple scenarios, but have significant limitations in complex environments: insufficient airspace obstacle perception; traditional methods rely on AGV chassis LiDAR, which can only detect ground obstacles (such as electrical cabinet bases, tools), and cannot identify high-level obstacles (such as open high electrical cabinet doors, suspended pipelines), leading to collision risks for the robotic arm or three-axis platform. Poor adaptability to dynamic environments; static maps are unable to handle temporary obstacles, and real-time replanning response speed is insufficient, easily causing path failure. Lack of multi-sensor data fusion; single sensors are susceptible to noise interference and cannot build a global environment model. Meanwhile, in the power distribution room environment, electrical cabinets are densely distributed, passages are narrow, and there are high-level instruments. Existing AGVs rely only on ground perception, which cannot meet the requirements for safe airspace passage. The robotic arm needs to avoid high-level obstacles during operation; dynamic obstacles require real-time response, and temporary obstacles require millisecond-level path updates. Summary of the Invention

[0003] The purpose of this invention is to address the shortcomings of existing technologies by proposing an intelligent path planning method and system based on complex environments. By acquiring and calculating the space range within which the robot can move freely, the method dynamically plans the robot's path, preventing the robot from affecting the actions of other people or objects during operation, thus making the robot's operation more efficient and intelligent.

[0004] To achieve the above objectives, the present invention adopts the following technical solution:

[0005] An intelligent path planning method based on complex environments includes the following steps:

[0006] Step 1: Collect point cloud data of obstacles on the ground and in the air using a multi-line lidar mounted on the AGV chassis and an airspace lidar mounted on the top of the three-axis platform, respectively.

[0007] Step 2: Perform time synchronization and spatial calibration on multi-source radar data to construct a layered map;

[0008] Step 3: Generate a global initial path based on the improved RRT* algorithm, with the objective function being:

[0009]

[0010] in, Indicates the path length. Indicates the steering angle. Indicates a safe distance. γ represents the weighting coefficients used to adjust the relative importance of different cost terms in the overall objective function J. Here, α is the weighting coefficient controlling path length optimization; increasing α will encourage the algorithm to choose shorter paths. β is the weighting coefficient controlling path smoothness optimization; increasing β will encourage the algorithm to choose smoother paths with gentler turns. γ is the weighting coefficient controlling path safety optimization; increasing γ will encourage the algorithm to choose paths farther from obstacles.

[0011] Step 4: During the movement, optimize the local path in real time using the Dynamic Window Method (DWA) and calculate the Safety Index (SI):

[0012]

[0013] in, Indicates a safe distance. Indicates the current distance. This indicates the robot's speed; if SI < a set threshold, the robot will immediately stop and send a safety alarm signal.

[0014] Preferably, in step 1, the collected data is first filtered for outliers to remove noise, insects, and other influences. Based on the IMU data, the point cloud distortion caused by the robot's own motion is compensated.

[0015]

[0016] in, Represents the rotation measured by the IMU. Represents the translation measured by the IMU;

[0017] Point cloud registration is performed using the ICP algorithm. First, the normal features and FPFH of the point cloud are extracted. Then, coarse registration is performed by selecting initial corresponding points based on RANSAC and calculating the initial transformation T_init. Finally, high-precision registration is performed by executing weighted ICP and optimizing the objective function.

[0018]

[0019] in, Represents a data item. This indicates the weights based on the distance between point pairs. This represents the point to be registered, the i-th point in the source point cloud. Indicates the target point cloud and The corresponding nearest neighbor is determined by nearest neighbor search, where T represents the rigid body transformation matrix; Represents the regularization term. This represents the regularization coefficient, balancing the weights of the data items and the regularization term. Constrain the properties of the transformation matrix T to prevent overfitting or non-physical transformations.

[0020] Preferably, in step 2, constructing the layered map includes:

[0021] A static global map is generated from an initial scan, producing a 3D raster map containing fixed obstacles.

[0022] Dynamic local map, which integrates point cloud data from the current frame in real time, marks temporary obstacles and updates them to the hybrid map.

[0023] Preferably, the static global map uses a TSDF model to generate a 3D raster map with a resolution of 5cm, and the dynamic local map uses probabilistic raster to represent dynamic obstacles, whose probability values ​​decay exponentially over time.

[0024] Preferably, in step 3, the improved RRT* algorithm introduces a spatial safety corridor constraint, specifically restricting the cooperative obstacle avoidance area between the robotic arm's operating space and the AGV's movement path, with the safety corridor width ≥ 0.2m.

[0025] By adopting the above technical solution: robot navigation relies on the LiDAR 4 on the AGV chassis 3. Because the AGV chassis 3 is low, the LiDAR 4 mainly detects ground obstacles, making it difficult to detect airborne obstacles. A comprehensive perception architecture is constructed, including surface perception and airborne perception. Surface perception is achieved by deploying a 16-line LiDAR (360° horizontal scanning) on ​​the AGV chassis, forming a ring-shaped detection zone to accurately detect low-altitude obstacles such as electrical cabinet bases and dropped tools. The AGV chassis installation diagram is shown below. Figure 4 As shown, the airspace perception system uses a 32-line lidar mounted on the top of a three-axis platform to construct a 0.5-3m high-altitude scanning network, specifically designed to detect potential hazards in traditional blind spots such as open cabinet doors and suspended pipelines. Figure 5 This diagram shows the installation location of the lidar on the three-axis platform.

[0026] The dual-radar network achieves data fusion through a spatiotemporal synchronization hub. At the hardware level, it adopts the IEEE 1588 PTP protocol to achieve microsecond-level clock synchronization. At the spatial level, it establishes a dynamic coordinate transformation matrix through hand-eye calibration, unifying point cloud data to the robot's base coordinate system to form a three-dimensional environment model without blind spots.

[0027] A layered map fusion mechanism is established, fusing a static global map with dynamic local maps. The static global map is generated during the initialization phase using a robot that rotates and scans in place, employing the TSDF algorithm to create a 3D raster map that permanently marks fixed obstacles such as walls and electrical cabinets. The dynamic local map is generated in real-time by fusing multi-radar point cloud flows, using a probabilistic raster model to represent temporary obstacles (such as mobile carts and personnel). The probability of an obstacle's existence decays exponentially; if an obstacle persists for an extended period, it is converted to a static element. The two maps are updated collaboratively using priority coverage rules, with dynamic layer data continuously overwriting the static layer. However, the static map automatically reverts to its original state when a dynamic obstacle disappears. This mechanism ensures real-time environmental awareness while preventing temporary interference from disrupting the map structure.

[0028] During path planning calculations, global path planning utilizes an improved RRT* algorithm. This algorithm introduces a spatial safety corridor constraint on top of the classic RRT* algorithm to address the challenge of collaborative obstacle avoidance between the robotic arm and the AGV. The intersection area between the robotic arm's working envelope and the AGV's movement path is defined as the critical spatial domain. Path sampling mandates that nodes be at least 0.2m from the nearest spatial obstacle (adjustable parameter). The objective function incorporates three optimization criteria: shortest path length (α weight), smoothest turning (β weight), and maximum safe distance (γ weight). The weight coefficients are dynamically adjusted based on the scenario's risk level, optimizing the path cost function in real-time to ensure a reliable and safe route for the robot to reach its designated location.

[0029] Guided by the global path, the local optimization adopts the DWA algorithm, which uses real-time sampling of velocity space to deal with sudden obstacles. Linear velocity and angular velocity generate hundreds of candidate trajectories within a limited window. The local optimal path is selected by comprehensively evaluating key performance factors such as progress efficiency (velocity term), motion smoothness (inverse of angular velocity term), and obstacle avoidance safety (obstacle distance term).

[0030] The innovative Safety Index (SI) quantifies path risk, with core parameters including the current distance to the nearest obstacle. Robot movement speed Preset safe distance The calculation logic is to take the minimum ratio of the theoretical time for the robot to reach the nearest obstacle to the safety threshold. In actual operation, this parameter is responded to in stages (the threshold can be adjusted again based on experience): SI>0.8 means normal passage, 0.5≤SI≤0.8 means reduced speed, and SI<0.5 means emergency braking and triggering of audible and visual alarms.

[0031] This invention also provides an intelligent path planning system based on complex environments, comprising:

[0032] Multi-line LiDAR module: including a 16-line LiDAR on the AGV chassis and a 32-line LiDAR on the top of the three-axis platform;

[0033] Map building module: used for generating and dynamically updating layered maps;

[0034] Path planning module: Integrates improved RRT* algorithm and dynamic window method, supporting global planning and local obstacle avoidance;

[0035] Safety monitoring module: Calculates safety index in real time and triggers emergency response.

[0036] By adopting the above technical solution, a layered heterogeneous communication architecture was constructed in the system for practical application. Through multi-protocol collaboration and priority scheduling mechanisms, the high-speed flow of perception data and the accurate execution of decision-making instructions across the entire domain are ensured. Its core lies in establishing three logically isolated data channels, with each layer employing customized communication strategies for different business needs.

[0037] The real-time control layer employs an industrial-grade synchronous real-time protocol to construct a hard real-time channel with millisecond-level cycles, dedicated to transmitting motion control commands and safety emergency signals. This layer achieves precise clock synchronization through time-sensitive networking technology and introduces a frame preemption mechanism to ensure that high-priority commands (such as emergency braking) can immediately interrupt low-priority data transmission, forming a closed-loop link where control commands directly reach the execution terminal.

[0038] The sensor data layer achieves device-level interconnection based on a distributed bus protocol, periodically transmitting lidar point cloud and attitude sensing data. Its core innovation lies in its master-slave clock automatic calibration system, ensuring all sensor nodes maintain microsecond-level time alignment with the master controller. Data encapsulation employs a structured unit design, with each frame completely carrying the point cloud information of a single scan, and optimized topology reducing wiring complexity. This layer incorporates a dynamic bandwidth allocation strategy, automatically prioritizing the transmission of airspace radar data when a sudden obstacle is detected.

[0039] The management information layer employs a publish / subscribe model to construct an asynchronous communication network, dedicated to non-real-time data transmission such as environment map updates, system status, and maintenance instructions. It primarily revolves around three core aspects: intelligent routing and compression, achieving targeted data distribution based on a dynamic topic mechanism, allowing subscribing nodes to acquire information on demand, and implementing feature differential compression and spatial block encoding on continuous data streams to significantly reduce transmission load; reliable transmission assurance, establishing a multi-level temporary storage architecture, buffering real-time data in memory queues, persisting offline tasks in a local library, and ensuring the delivery of critical instructions through transaction confirmation and breakpoint resumption mechanisms; and a zero-trust security system, with two-way authentication upon device access, dynamic key rotation to encrypt data streams, and a behavior monitoring engine to defend against abnormal access in real time.

[0040] The three-layer data channel achieves organic collaboration through an intelligent internet gateway: a protocol conversion middleware is used between the control layer and the management layer to achieve bidirectional conversion of real-time commands and status information; the data replication latency is avoided between the sensing layer and the control layer through a memory pass-through architecture. A typical workflow is as follows: spatial perception data is transmitted via a high-speed bus to facilitate real-time fusion and updating of the dynamic map module; map updates enable the path planner to generate obstacle avoidance commands; and the control layer issues and executes these commands in milliseconds, forming a closed-loop response chain of perception-decision-execution.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] 1. This invention systematically solves the core challenges of robot collaborative obstacle avoidance and dynamic response in complex environments by constructing a technical system of full-domain three-dimensional perception, hierarchical intelligent decision-making, and real-time safety control. It innovatively proposes a heterogeneous radar collaborative scanning architecture, breaking through the limitations of traditional ground perception; and establishes a dynamic-static coupled map modeling mechanism to achieve dynamic evolution of environmental cognition.

[0043] 2. This invention integrates airspace safety constraints and local optimization algorithms at the path planning layer to form a collaborative decision-making chain of global guidance and instantaneous obstacle avoidance. The communication architecture design adheres to the principles of "data hierarchical, dedicated channels, and protocol adaptation," ensuring the reliable transmission of critical data through a three-layer heterogeneous network.

[0044] 3. This invention not only significantly improves the navigation safety of robots in dense obstacle scenarios, but also provides a scalable technical paradigm for autonomous systems in fields such as intelligent manufacturing and energy inspection, promoting the paradigm leap of mobile robots from single-dimensional obstacle avoidance to full-domain intelligence. Attached Figure Description

[0045] Figure 1 This is an overall flowchart of the present invention;

[0046] Figure 2 This is a flowchart of the path planning process of the present invention;

[0047] Figure 3 This is a schematic diagram of the overall installation of the robot, the three-axis platform, and the AGV chassis in this invention;

[0048] Figure 4 This is a schematic diagram of the installation of the lidar and the AGV chassis in this invention;

[0049] Figure 5 This is a schematic diagram of the installation of the three-axis platform and the lidar in this invention.

[0050] In the diagram: 1. Dual collaborative arms; 2. Three-axis platform; 3. AGV chassis; 4. LiDAR. Detailed Implementation

[0051] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings, so that those skilled in the art can better understand the advantages and features of the present invention, thereby making a clearer definition of the scope of protection of the present invention. The embodiments described in this invention are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0052] The composite robot system of this invention mainly includes core components such as an AGV chassis 3, a three-axis platform 2, a dual collaborative arm 1, and a lidar 4. Figure 3 As shown. The AGV chassis 3 serves as a mobile platform, providing autonomous navigation and carrying capabilities; the three-axis platform 2 is fixed to the chassis, serving as the mounting base for the dual arms and performing operations on the electrical cabinet; the dual cooperative arms 1 are mounted on the three-axis platform 2 to perform operational tasks; the lidar 4 is mounted on the bottom of the AGV chassis 3 and is used as a sensor for sensing obstacles on the ground and in the airspace.

[0053] An intelligent path planning method based on complex environments includes the following steps:

[0054] Step 1: Collect point cloud data of obstacles on the ground and in the air using a multi-line lidar mounted on the AGV chassis and an airspace lidar mounted on the top of the three-axis platform, respectively.

[0055] Step 2: Perform time synchronization and spatial calibration on multi-source radar data to construct a layered map;

[0056] Step 3: Generate a global initial path based on the improved RRT* algorithm, with the objective function being:

[0057]

[0058] in, Indicates the path length. Indicates the steering angle. Here, α represents the safety distance, β represents the weighting coefficient for optimizing the control path length, β represents the weighting coefficient for optimizing the control path smoothness, and γ represents the weighting coefficient for optimizing the control path safety. These weighting coefficients can be dynamically adjusted based on planning priorities. During the journey along the initial path, if a dynamic obstacle is detected, the local optimization module is triggered, and the DWA algorithm is used to form an obstacle avoidance trajectory. The DWA evaluation function is as follows:

[0059]

[0060] in, To evaluate the overall score, Indicates the minimum distance to the obstacle. This represents angular velocity, specifically the angular velocity of the robot's rotation around its own center. This represents linear velocity, which is the linear velocity at which the robot moves in the current direction.

[0061] Step 4: During the movement, optimize the local path in real time using the Dynamic Window Method (DWA) and calculate the Safety Index (SI):

[0062]

[0063] in, Indicates a safe distance. Indicates the current distance. This indicates the robot's speed; if SI < a set threshold, the robot will immediately stop and send a safety alarm signal.

[0064] In the global map construction phase, the first step is to ensure the spatiotemporal synchronization of each sensor. This is achieved by synchronizing the clocks of each radar using PTP (Precise Time Protocol) with an error of ≤1ms. The conversion relationship between each radar and the robot's base coordinate system is then established using a calibration board and hand-eye calibration method.

[0065] Precise Time Protocol (PTP) constructs a master-slave clock tree, with the master controller periodically broadcasting time synchronization messages. Each lidar records the message arrival time using hardware timestamps, and combined with a path delay compensation algorithm, achieves microsecond-level clock alignment at the protocol stack level. Its core lies in using bidirectional message exchange to measure network transmission latency, dynamically calibrating slave device clock drift, and ensuring strict synchronization of multi-liquidity sampling times. A sensor-robot coordinate transformation chain is established using hand-eye calibration technology. The robotic arm drives a checkerboard calibration board to complete multi-pose movements, simultaneously collecting the pose relationship from the robotic arm's end effector to the calibration board in each pose (calculated through a kinematic model) and the relative pose of the lidar to the calibration board (detected through image corner points). A pose transformation equation for "sensor-calibration board-robotic arm-base" is constructed, and the optimal coordinate transformation matrix is ​​solved using the least squares method to eliminate spatial errors caused by installation position deviations.

[0066] Specifically, in step 1, the collected data is first filtered for outliers to remove noise, insects, and other influences. Based on the IMU data, the point cloud distortion caused by the robot's own motion is compensated.

[0067]

[0068] in, Represents the rotation measured by the IMU. Represents the translation measured by the IMU;

[0069] Point cloud registration is performed using the ICP algorithm. First, the normal features and FPFH of the point cloud are extracted. Then, coarse registration is performed by selecting initial corresponding points based on RANSAC and calculating the initial transformation T_init. Finally, high-precision registration is performed by executing weighted ICP and optimizing the objective function.

[0070]

[0071] in, Represents a data item. This indicates the weights based on the distance between point pairs. This represents the point to be registered, the i-th point in the source point cloud. Indicates the target point cloud and The corresponding nearest neighbor is determined by nearest neighbor search, where T represents the rigid body transformation matrix; Represents the regularization term. This represents the regularization coefficient, balancing the weights of the data items and the regularization term. Constrain the properties of the transformation matrix T to prevent overfitting or non-physical transformations.

[0072] Outlier filtering is based on statistical distribution characteristics to filter out dynamic noise: by calculating the distance distribution of the point cloud neighborhood, using 3 times the standard deviation as a threshold, discrete points whose distance mean exceeds the threshold (such as flying insects and dust reflection points) are removed, while the continuous surface point cloud is retained, thereby maintaining the integrity of the environmental structure.

[0073] Motion distortion compensation addresses point cloud deformation caused by robot movement: Angular velocity and linear acceleration data are acquired in real-time using an IMU, decomposing the motion within a single scan cycle into a continuous sequence of micro-displacements. Then, pose changes are calculated via reverse interpolation based on the precise timestamp of each laser point. Finally, the original point cloud is inversely projected onto the coordinate system at the start of the scan. This process eliminates point cloud stretching or compression distortion caused by robot movement and turning, restoring the geometry of the real environment.

[0074] The point cloud registration process is as follows: First, point cloud normal vectors are calculated based on local surface curvature to characterize the geometric features of the environment. Then, an FPFH descriptor is generated, and a 33-dimensional FPFH descriptor vector is generated by integrating features such as point spacing and normal angle. Next, the RANSAC algorithm is used for coarse matching, randomly selecting three pairs of points with similar features, solving the initial transformation matrix, and verifying the number of interior points. The transformation relationship corresponding to the largest set of interior points is selected through iterative optimization. Finally, weighted fine registration is performed, which is a refined registration based on the initial transformation—dynamically allocating point pair weights based on the similarity of point spacing and normals to reduce the impact of mismatches. At the same time, a variable stiffness regularization strategy is adopted, which weakens the rigidity constraint in the early stage of registration to allow for large-scale deformation, and strengthens the orthogonality constraint in the later stage of iteration to avoid non-physical distortion. When implementing the multi-scale convergence strategy, the initial stage accelerates the calculation by downsampling the point cloud, and the final stage uses the full-resolution point cloud to optimize details, thereby achieving accurate alignment of the static map and providing an environmental model with centimeter-level accuracy for path planning.

[0075] Specifically, in step 2, constructing the layered map includes:

[0076] A static global map is generated from an initial scan, producing a 3D raster map containing fixed obstacles.

[0077] Dynamic local map, which integrates point cloud data from the current frame in real time, marks temporary obstacles and updates them to the hybrid map.

[0078] Specifically, the static global map uses a TSDF model to generate a 3D raster map with a resolution of 5cm, while the dynamic local map uses probabilistic raster to represent dynamic obstacles, whose probability values ​​decay exponentially over time.

[0079] Specifically, in step 3, the improved RRT* algorithm introduces a spatial safety corridor constraint, which specifically restricts the cooperative obstacle avoidance area between the robotic arm's operating space and the AGV's movement path, with a safety corridor width ≥ 0.2m.

[0080] In the global path planning phase, the improved RRT* algorithm enhances path reliability through optimized sampling strategies and dynamic weight adjustment. For critical areas in complex environments (such as narrow passages and areas with dense obstacles), guided sampling is employed to increase effective path nodes. The objective function comprehensively considers path length, turning smoothness, and safety distance, and the weight coefficients can be dynamically adjusted according to environmental risks; for example, when high-altitude obstacles are detected, the weight of the safety distance is automatically increased. When using the DWA algorithm for local optimization, the velocity sampling window is subdivided according to the robot's load state: a higher linear and angular velocity range is allowed when unloaded, while the velocity threshold is automatically reduced when the robot arm is under load to avoid collision risks caused by motion inertia.

[0081] An intelligent path planning system based on complex environments includes:

[0082] Multi-line LiDAR module: including a 16-line LiDAR on the AGV chassis and a 32-line LiDAR on the top of the three-axis platform;

[0083] Map building module: used for generating and dynamically updating layered maps;

[0084] Path planning module: Integrates improved RRT* algorithm and dynamic window method, supporting global planning and local obstacle avoidance;

[0085] Safety monitoring module: Calculates safety index in real time and triggers emergency response.

[0086] The safety monitoring module uses a Safety Index (SI) to dynamically assess risks. Its calculation logic combines the current obstacle distance, robot speed, and preset safety thresholds. When the SI is in the medium-risk range, the system automatically increases the scanning frequency of the airspace lidar to predict changes in obstacles ahead. If the SI is below the safety threshold, the emergency braking mechanism is executed according to priority: first, the robot arm's power source is cut off, then the chassis braking device is activated, and simultaneously, an audible and visual alarm is triggered. The safety monitoring system employs a dual-redundancy design: the main unit drives safety responses in real time, while backup data from the slave unit is used for fault backtracking. Furthermore, the safety threshold can be adaptively adjusted according to scene characteristics (such as personnel density), enhancing the safety of human-machine collaboration.

[0087] In this embodiment, firstly, the power distribution room is scanned globally using all the lidars on the composite robot. The information acquired by different lidars is fused to obtain a global static map. In subsequent tasks, when faced with path planning, the composite robot prioritizes calculating the optimal path in the static map. If no feasible path exists, it sends a distress signal, waiting for manual obstacle removal. If a feasible path exists, it sends a signal to the mobile chassis and moves according to the planned path. During movement, the lidars dynamically scan the entire area to ensure the reliability and safety of the path, avoiding collisions with dynamic obstacles or airspace obstacles not shown in the static global map. If a new obstacle is detected during movement, it is marked and modified in the static map, and a report is sent to the operator requesting obstacle removal. The route is then replanned until the target location is successfully reached.

[0088] In summary, this invention constructs a technical system of global perception, hierarchical decision-making, and real-time control. It achieves three-dimensional obstacle detection in both ground and airspace through a combination of heterogeneous LiDAR systems, and utilizes a fusion mechanism of static global maps and dynamic local maps to achieve accurate construction and real-time updates of the environmental model. At the path planning level, the RRT* algorithm and DWA algorithm are improved to work collaboratively, combined with dynamic weight adjustment and speed window subdivision, enhancing the efficiency and safety of path planning in complex scenarios. The safety detection module achieves millisecond-level risk warning and emergency response through a hierarchical response mechanism and dual redundancy design. This solution systematically solves the limitations of traditional path planning in global obstacle perception and dynamic environment adaptation, providing a safe and efficient autonomous navigation solution for mobile robots in fields such as power line inspection and warehousing logistics.

[0089] The descriptions and practices disclosed in this invention are readily apparent and understandable to those skilled in the art, and various modifications and refinements can be made without departing from the principles of this invention. Therefore, any modifications or improvements made without departing from the spirit of this invention should also be considered within the scope of protection of this invention.

Claims

1. An intelligent path planning method based on complex environments, characterized in that, The steps include: Step 1: Collect point cloud data of obstacles on the ground and in the air using a multi-line lidar mounted on the AGV chassis and an airspace lidar mounted on the top of the three-axis platform, respectively. Step 2: Perform time synchronization and spatial calibration on multi-source radar data to construct a layered map; Step 3: Generate a global initial path based on the improved RRT* algorithm, with the objective function being: in, Indicates the path length. Indicates the steering angle. The safe distance is represented by α, which is the weighting coefficient for optimizing the control path length, β is the weighting coefficient for optimizing the control path smoothness, and γ is the weighting coefficient for optimizing the control path safety. Step 4: During the movement, optimize the local path in real time using the Dynamic Window Method (DWA) and calculate the safety index SI. in, Indicates a safe distance. Indicates the current distance. This indicates the robot's speed; if SI < a set threshold, the robot will immediately stop and send a safety alarm signal.

2. The intelligent path planning method based on complex environments according to claim 1, characterized in that, In step 1, the collected data is first filtered for outliers to remove noise and insect interference. Based on the IMU data, the point cloud distortion caused by the robot's own motion is compensated. in, Represents the rotation measured by the IMU. Represents the translation measured by the IMU; Point cloud registration is performed using the ICP algorithm. First, the normal features and FPFH of the point cloud are extracted. Then, coarse registration is performed by selecting initial corresponding points based on RANSAC and calculating the initial transformation T_init. Finally, high-precision registration is performed by executing weighted ICP and optimizing the objective function. in, Represents a data item. This indicates the weights based on the distance between point pairs. This represents the point to be registered, the i-th point in the source point cloud. Indicates the target point cloud and The corresponding nearest neighbor is determined by nearest neighbor search, where T represents the rigid body transformation matrix; Represents the regularization term. This represents the regularization coefficient, balancing the weights of the data items and the regularization term. Constrain the properties of the transformation matrix T to prevent overfitting or non-physical transformations.

3. The intelligent path planning method based on complex environments according to claim 1, characterized in that, Step 2, constructing the layered map includes: A static global map is generated from an initial scan, producing a 3D raster map containing fixed obstacles. Dynamic local map, which integrates point cloud data from the current frame in real time, marks temporary obstacles and updates them to the hybrid map.

4. The intelligent path planning method based on complex environments according to claim 3, characterized in that, The static global map uses a TSDF model to generate a 3D raster map with a resolution of 5cm. The dynamic local map uses probabilistic raster to represent dynamic obstacles, and the probability value decays exponentially over time.

5. The intelligent path planning method based on complex environments according to claim 1, characterized in that, In step 3, the improved RRT* algorithm introduces a spatial safety corridor constraint, specifically restricting the cooperative obstacle avoidance area between the robotic arm's operating space and the AGV's movement path, with a safety corridor width ≥ 0.2m.

6. An intelligent path planning system based on complex environments, characterized in that, include: Multi-line LiDAR module: including a 16-line LiDAR on the AGV chassis and a 32-line LiDAR on the top of the three-axis platform; Map building module: used for generating and dynamically updating layered maps; Path planning module: Integrates improved RRT* algorithm and dynamic window method, supporting global planning and local obstacle avoidance; Safety monitoring module: Calculates safety index in real time and triggers emergency response.

Images