A fusion path planning method and system for an intelligent elderly care robot

Abstract

The application relates to the technical field of intelligent old-age care robots, in particular to a fusion path planning method and system for an intelligent old-age care robot. The method comprises the following steps: constructing a two-dimensional continuous space map containing static and dynamic obstacles; performing global path planning through a dynamic probability target bias strategy and a path post-processing mechanism by using an improved BI-RRT-APF algorithm, generating a smooth path and discretizing the smooth path into a path point sequence; adopting an improved DWA algorithm, taking a current path point as a local target, and realizing local dynamic obstacle avoidance through an accurate circular arc motion model and an adaptive weight evaluation function; realizing deep fusion of global and local planning through a path point by path point tracking mechanism, and realizing autonomous escape from a trap in combination with a trap escape module. The application can realize whole path planning from a starting point to a target point, and improve the passing capacity of the intelligent old-age care robot in a complex environment.

Description

Technical Field

[0001] This invention relates to the field of control technology for elderly care robots, and more specifically, to a fusion path planning method and system for intelligent elderly care robots. Background Technology

[0002] It undertakes services such as companionship, delivery of items, and emergency calls. The elderly care environment is characterized by dense obstacles, narrow spaces, and many dynamic obstacles, which places high demands on the path planning and dynamic obstacle avoidance capabilities of mobile robots.

[0003] Currently, mobile robot path planning mainly employs a combination of global and local planning methods. Typical global path planning algorithms include the A algorithm, Dijkstra's algorithm, Rapid Extended Random Tree (RRT) algorithm, and Artificial Potential Field (APF) method. These methods exhibit high planning stability in static environments but show poor adaptability in dynamic and unstructured environments. Classical local path planning algorithms include the Dynamic Window (DWA) method and the D algorithm.

[0004] However, traditional global planning algorithms have significant limitations in elderly care environments. Algorithms such as RRT, APF, and BI-RRT-APF suffer from problems like high blind node expansion, large path redundancy, and susceptibility to local minima, making them difficult to adapt to complex layouts in elderly care scenarios, such as narrow doorways and densely packed furniture. They also lack real-time obstacle avoidance capabilities for dynamic obstacles (elderly people, pets). Furthermore, the traditional Dynamic Window (DWA) method in elderly care environments suffers from the following problems: the robot travels around the outside of dense obstacle areas, increasing the total distance traveled; the evaluation function is prone to failure when encountering complex combinations of similarly located obstacles; endpoint oscillations are common when approaching the target point, hindering accurate positioning; and the robot is prone to getting stuck and stalling in scenarios surrounded by obstacles such as chair legs and trolleys, with difficulty in autonomous recovery. In addition, existing fusion schemes of global and local planning are mostly loosely coupled structures of "global path + local tracking," with insufficient information feedback between the global and local levels; the handling of dynamic obstacles is often limited to simple predictions at the local planning layer, lacking a global-level collaborative obstacle avoidance mechanism; and the safety and comfort considerations in elderly care environments are insufficient.

[0005] Therefore, it is necessary to propose a fusion path planning method and system for intelligent elderly care robots. By improving the deep integration of global and local planning algorithms, and combining dynamic probabilistic target bias strategy, path post-processing mechanism, precise docking mechanism and jamming escape module, safe, efficient and accurate path planning and dynamic obstacle avoidance can be achieved in complex elderly care environments. This will improve the robot's ability to navigate and its service quality in narrow passages, dense obstacles and dynamic obstacles. Summary of the Invention

[0006] In view of this, the present invention proposes a fusion path planning method and system for intelligent elderly care robots, aiming to solve the problems of blind node expansion, high path redundancy, lack of real-time dynamic obstacle avoidance capability in the global planning algorithm, excessive detour distance, easy failure of evaluation function, difficulty in accurate docking and lack of autonomous escape capability in the local planning algorithm, and the fact that the fusion of global and local planning is mostly a loosely coupled structure with insufficient information feedback and lack of global collaborative obstacle avoidance mechanism.

[0007] This invention proposes a fusion path planning method for intelligent elderly care robots, comprising: Obtain a two-dimensional spatial map corresponding to the working environment of the elderly care robot, mark static obstacle areas and dynamic obstacle areas in the two-dimensional spatial map, and set the robot's starting position, target position and robot motion state parameters; Global path planning is performed based on an improved bidirectional fast expanding random tree combined with an artificial potential field algorithm; among which, A bidirectional search tree is established based on the robot's starting position and target position. During the path expansion process, a dynamic target bias strategy is adopted to adjust the target position sampling probability according to the current search iteration state and expand the search tree based on random sampling points. After obtaining the initial path, the initial path is optimized. The presence of obstacle collisions is determined by detecting the straight-line connection relationship between path nodes. When there is no collision, the intermediate path nodes between the connecting nodes are deleted. The optimized path is then interpolated according to the preset path point interval to generate a global path point sequence. Local path planning is based on the improved dynamic window method, which takes the current path point to be tracked as the local target and establishes a dynamic velocity sampling window according to the robot's current motion state and motion constraints. Multiple speed control parameters are generated within the dynamic speed sampling window. The motion trajectory corresponding to each speed control parameter is predicted based on the robot motion model. The predicted motion trajectory is evaluated based on the trajectory evaluation function, and the corresponding speed control parameter is selected as the local motion control quantity. The global path point sequence is fused with the local motion control variables. The robot is controlled to move according to the path point tracking order. When the robot's current position reaches the position range corresponding to the current path point, the next path point is updated as the local target. The local path planning process is executed cyclically until the robot reaches the target position.

[0008] Furthermore, when constructing a two-dimensional spatial map of the working environment of the elderly care robot, it includes: Point cloud data of the robot's surrounding environment is collected using a lidar sensor, and the point cloud data is filtered to remove outliers and noise points, resulting in processed environmental point cloud data. The processed environmental point cloud data is mapped to a two-dimensional coordinate space and divided into multiple grid units according to a preset grid size. A two-dimensional occupied grid map is generated based on the obstacle occupancy information in each grid unit. The two-dimensional occupied grid map includes obstacle areas and passable areas. The system acquires environmental image data based on visual sensors or depth cameras, performs target detection processing on the environmental image data, identifies dynamic obstacles in the environment, and obtains the category information, location information, and motion state information of the dynamic obstacles. The two-dimensional occupied grid map and the dynamic obstacle information are converted to a unified coordinate system and the map data is fused. The dynamic obstacle information is then loaded into the two-dimensional occupied grid map to generate an environmental map containing both static and dynamic obstacle information.

[0009] Furthermore, when using the improved bidirectional fast expanding random tree combined with the artificial potential field algorithm for global path planning, it includes: A bidirectional search tree is built based on the robot's starting position and target position, and the maximum number of iterations for path search and the current number of iterations are set. During the path expansion process, the target bias probability is determined based on the current iteration number and the maximum iteration number. The target bias probability decreases according to a preset change rule as the iteration number increases, and a minimum limit value for the target bias probability is set. During each path expansion iteration, a random number within a preset range is generated, and the random number is compared with the current target bias probability; When the random number is less than the current target bias probability, the target position is used as the current extended sampling point; when the random number is greater than or equal to the current target bias probability, an extended sampling point is generated according to the random sampling method, and the search tree is updated based on the extended sampling point. After the search tree generates an initial path from the starting position to the target position, the initial path is optimized.

[0010] Furthermore, when performing path optimization processing on the initial path, it includes: Obtain multiple path nodes in the initial path, select the candidate node farthest from the current node starting from the starting node, and determine whether the connecting line segment between the current node and the candidate node collides with an environmental obstacle, wherein: When the connecting line segment does not collide with an obstacle, delete the intermediate path node located between the connecting line segments and retain the candidate node; When the connecting line segment collides with an obstacle, a new candidate node is selected for connection determination. The path length is calculated based on the optimized path nodes, and the optimized path is resampled according to the preset path point interval to generate a path point sequence with equal spacing.

[0011] Furthermore, when performing local path planning based on the improved dynamic window method, it includes: Acquire the robot's current pose and motion state information, and establish a trajectory prediction model based on the differential drive motion model; Predict the robot's motion state at the next moment based on candidate linear velocity and angular velocity parameters; Specifically, when the robot's angular velocity is not zero, the position coordinates and attitude angle at the next moment are calculated according to the current linear velocity, angular velocity, and the robot's current attitude angle, following a circular motion pattern; when the robot's angular velocity is zero, the position coordinates and attitude angle at the next moment are calculated according to the current linear velocity and the robot's current attitude angle, following a linear motion pattern. A dynamic velocity window is constructed based on robot motion constraint parameters, wherein the motion constraint parameters include maximum linear acceleration, maximum angular acceleration, maximum linear velocity, and maximum angular velocity. The range of values ​​for linear velocity and angular velocity is limited based on the motion constraint parameters, and multiple candidate velocity combinations are generated within the limited range. The robot's motion trajectory is predicted based on each candidate velocity combination, and the motion trajectory is evaluated based on the trajectory evaluation function. The corresponding velocity control parameter is then selected as the local motion control quantity.

[0012] Furthermore, when evaluating the motion trajectory based on the trajectory evaluation function and selecting the corresponding velocity control parameter as the local motion control quantity, the process includes: During local path planning, the distance between the robot's current position and the target point is obtained, and the robot's maximum allowable linear velocity is adjusted based on this distance, wherein: When the distance between the robot's current position and the target point is greater than the preset deceleration distance, the preset maximum linear velocity is used as the speed limit value; When the distance between the robot's current position and the target point is less than or equal to the preset deceleration distance, the maximum allowable linear speed is reduced according to the distance change relationship, and the dynamic speed window is updated according to the adjusted speed limit value.

[0013] Furthermore, when fusing the global pathpoint sequence with the local motion control variables, the process includes: The global path generated by the improved bidirectional fast expanding random tree combined with the artificial potential field algorithm is obtained, and the global path is discretized according to the preset path sampling rules to generate a sequence of path points with sequential relationship. Within the robot motion control cycle, the current path point to be tracked is determined from the path point sequence based on the robot's current position, and the current path point to be tracked is used as a local target in the dynamic window method local path planning; The distance between the robot's current position and the current path point to be tracked is obtained in real time, and the distance is compared with a preset path point switching threshold. When the distance between the robot's current position and the current path point to be tracked is less than or equal to the path point switching threshold, the next path point in the path point sequence is updated as the new local target, and local path planning continues. Once all path points in the path point sequence have been tracked, the final target position is used as the local planning target, and target position docking control is executed based on the positional relationship between the robot's current position and the final target position.

[0014] Furthermore, it also includes stuck state detection and escape control, specifically including: During the robot's path tracking process, the robot's position information is acquired at preset time intervals, and the robot's corresponding movement distance is calculated based on the position changes within consecutive time periods. The distance traveled is compared with a preset stagnation threshold; When the moving distance is less than the stall threshold, the robot is determined to be stuck and the current path tracking control is paused. The robot's current posture information is acquired, and a steering control command different from the current direction of motion is generated to control the robot to perform steering motion. After performing the turning motion, the robot continues to monitor the distance it has moved. When the robot resumes its movement, it exits the stuck state and resumes path point tracking control.

[0015] Furthermore, the robot's current posture information is acquired, and a steering control command different from the current direction of motion is generated. When controlling the robot to perform steering motion, this includes: The robot motion state detection time interval and the stagnation distance threshold are set, wherein the motion state detection time interval is set to 3 to 5 seconds, and the stagnation distance threshold is set to 0.1 to 0.2 meters; Acquire robot position change information within each motion state detection time interval and calculate the corresponding robot movement distance; When the moving distance is less than the stopping distance threshold, the robot is determined to be stuck. When the robot is stuck, a steering control command is generated based on the robot's current attitude angle to control the robot to perform a stationary rotation action, wherein the rotation angle of the stationary rotation action is set to 90 degrees to 180 degrees. After completing the rotation, control the robot to move in the direction of the rotation and reacquire the robot's movement distance; When the robot's movement distance is still less than the stall distance threshold, different steering control commands are regenerated, and the rotation action and movement state detection process are repeated until the robot's movement distance reaches or exceeds the stall distance threshold, at which point it exits the stuck state.

[0016] Compared with existing technologies, the advantages of this invention are as follows: By employing a dynamic probabilistic target bias strategy, the global planning algorithm enhances convergence speed in the early stages of the search and maintains a certain level of exploration capability in the later stages, effectively improving the planning efficiency and success rate of global paths in complex indoor environments; by removing redundant nodes through greedy pruning and equidistant interpolation post-processing mechanisms, the path becomes shorter and straighter, ensuring the uniformity of path point density during DWA tracking; by eliminating the accumulated error of the traditional Euler approximation through a precise circular arc motion model, the trajectory prediction accuracy is improved; and by employing a distance-adaptive deceleration strategy and a precise docking mechanism, the robot can dynamically adjust its speed according to the target distance and achieve millimeter-level precise docking when approaching the target, reducing the arrival error compared to traditional DWA. The distance was reduced from 2.950m to 0.000m; the weight of each evaluation index was dynamically adjusted according to the distance through an adaptive weight evaluation function, so that the robot could prioritize different motion targets at different stages; through a path point tracking mechanism that deeply integrates global and local paths, the optimality of the global path was preserved, and the robot also had the ability to respond in real time to dynamic environments; by monitoring the stagnation status in real time and performing turning and escape actions through the jamming escape module, the number of times the robot jammed and stopped was effectively reduced. Simulation experiments showed that the number of jamming escapes was reduced by 53.8% to 79.3%, and the path tracking rate was as high as 94% or more, which improved the intelligent elderly care robot's ability to pass through narrow passages, dense obstacles and dynamic obstacles coexisting environments and the quality of service.

[0017] On the other hand, this application also provides a fusion path planning system for intelligent elderly care robots, including: The data acquisition module is configured to acquire a two-dimensional spatial map corresponding to the working environment of the elderly care robot, mark static obstacle areas and dynamic obstacle areas in the two-dimensional spatial map, and set the robot's starting position, target position and robot motion state parameters. The analysis module, electrically connected to the acquisition module, is configured to perform global path planning based on an improved bidirectional fast expanding random tree combined with an artificial potential field algorithm; wherein, The analysis module is also configured to build a bidirectional search tree based on the robot's starting position and target position, adopt a dynamic target bias strategy during path expansion, adjust the target position sampling probability according to the current search iteration state, and expand the search tree based on random sampling points; The control module is electrically connected to the analysis module. The control module is configured to perform path optimization processing on the initial path after obtaining the initial path. It determines whether there is an obstacle collision by detecting the straight connection relationship between the path nodes. When there is no collision, the intermediate path nodes between the connecting nodes are deleted. The optimized path is then interpolated according to the preset path point interval to generate a global path point sequence. The control module is also configured to perform local path planning based on an improved dynamic window method, taking the current path point to be tracked as the local target, and establishing a dynamic velocity sampling window based on the robot's current motion state and motion constraints. The control module is also configured to generate multiple speed control parameters within the dynamic speed sampling window, predict the motion trajectory corresponding to each speed control parameter based on the robot motion model, evaluate the predicted motion trajectory based on the trajectory evaluation function, and select the corresponding speed control parameter as the local motion control quantity. The control module is also configured to fuse the global path point sequence with local motion control quantities, control the robot's movement according to the path point tracking order, and update the next path point as the local target when the robot's current position reaches the position range corresponding to the current path point. The local path planning process is executed cyclically until the robot reaches the target position.

[0018] It is understood that the fusion path planning method and system for intelligent elderly care robots in the above embodiments of this application have the same beneficial effects, and will not be described again. Attached Figure Description

[0019] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A flowchart illustrating a fusion path planning method for intelligent elderly care robots, provided as an embodiment of the present invention; Figure 2 A flowchart illustrating a fusion path planning method for intelligent elderly care robots provided in an embodiment of the present invention; Figure 3 This is a functional block diagram of a fusion path planning system for intelligent elderly care robots provided in an embodiment of the present invention. Detailed Implementation

[0020] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0021] like Figures 1-2 As shown in some embodiments of this application, this embodiment provides a fusion path planning method for intelligent elderly care robots, including: Step S100: Construct a two-dimensional continuous spatial map of the working environment of the elderly care robot.

[0022] Specifically, constructing a two-dimensional continuous spatial map includes: collecting environmental point cloud data using a LiDAR sensor; filtering the point cloud data to remove noise and measurement errors caused by dynamic obstacles; rasterizing the filtered point cloud data to generate a two-dimensional occupancy grid map containing passable and obstacle areas; identifying dynamic obstacles using a visual sensor or depth camera to obtain information on the type, location, and speed of the dynamic obstacles; and fusing the static obstacle map with the dynamic obstacle information to construct a complete environmental map containing both static and dynamic obstacles.

[0023] Understandably, collecting environmental point cloud data through LiDAR sensors enables large-scale, high-precision perception of the elderly care environment; point cloud filtering effectively removes noise from dynamic obstacles (such as walking elderly people and pets) and sensor measurement errors, ensuring the accuracy of the static map; rasterization divides the continuous two-dimensional space into discrete grid cells, with each grid storing occupancy status information, facilitating subsequent path planning algorithm search and calculation; visual sensors or depth cameras can identify the types of dynamic obstacles (such as people, wheelchairs, and pets), their real-time positions, and movement speeds, providing data support for dynamic obstacle avoidance in the local planning layer; the fusion of static obstacle maps and dynamic obstacle information constructs a complete environmental perception model, laying the foundation for collaborative work between global and local planning.

[0024] For example, in this embodiment, the lidar sensor is a two-dimensional laser scanner, installed at the center of the top of the robot, with a scanning angle range of 360°, an angular resolution of 0.5°, a maximum detection distance of 8m, and a sampling frequency of 10Hz; the point cloud filtering uses a statistical outlier removal algorithm, calculating the neighborhood of each point. The average distance between the nearest neighbors is used to identify points that exceed two standard deviations from the global average. Points exceeding this average distance are considered noise and removed. The grid map resolution is set to 0.05m, meaning each grid cell represents a 5cm x 5cm space. A grid value of 0 indicates passage, and a value of 100 indicates an obstacle. The vision sensor uses a depth camera mounted below the front of the robot, with a detection angle of 90° horizontally and 60° vertically. It can identify dynamic obstacles including people, wheelchairs, canes, and pets, with a position estimation accuracy better than 0.1m and a velocity estimation accuracy better than 0.05m / s. Dynamic obstacle information is stored in a list format, with each obstacle containing a category identifier and location coordinates. Velocity vector And timestamps, used to predict the future trajectory of obstacles.

[0025] Step S200: Perform global path planning using the improved BI-RRT-APF algorithm.

[0026] Specifically, when using the improved BI-RRT-APF algorithm for global path planning, it includes two parts: a dynamic probabilistic target bias strategy and a path post-processing mechanism; assuming the maximum number of iterations is... The current iteration number is Define the target bias probability As shown in equation (1): In the formula: The initial bias probability is set to 0.3, which gives the tree a greater chance to expand directly toward the target point in the early stages of the search, thus accelerating the approach to the target area. This is the final bias probability, with a value of 0.2; The lower bound for the probability is set to 0.05 to prevent the bias probability from being too small and causing the search to completely lose its target guidance in the later stages. In each iteration, uniformly distributed random numbers are first generated. ,like If the target point is selected, it will be used as the random sampling point for this expansion; otherwise, regular random sampling will be performed.

[0027] Understandably, the dynamic probabilistic target bias strategy increases the probability of the target point being selected in the early stages of the search, enabling the expanding tree to grow rapidly toward the target region and significantly shortening the search time in the initial phase. As the number of iterations increases, the target bias probability gradually decreases and approaches a certain value. This allows the search to maintain a certain degree of random exploration capability in the later stages, avoiding getting trapped in local minima; probability lower bound The settings ensure that even in the later stages of the search, a minimum level of target guidance is retained, preventing the search from degenerating into a blind random walk. Compared with the traditional BI-RRT-APF algorithm with fixed bias probabilities, the dynamic bias strategy of this invention can achieve a better balance between convergence speed and search breadth.

[0028] The path post-processing mechanism consists of two parts: greedy pruning and equidistant interpolation. The goal of greedy pruning is to replace the winding parts of the original path with the longest possible straight segments while maintaining a collision-free path, thereby reducing the number of nodes, shortening the total path length, and making the path more linear. Let the initial path be... ,in For the path nodes to be connected sequentially, the pruning process starts from the starting point and attempts to connect them with the farthest node as the target. Specifically, it includes the following steps: Initializing the pruned path. And set the index of the current processing node. From the last node Start traversing downwards to If line segment If there is no collision with any obstacles (using discrete sampling detection with a sampling step size of 0.05m), then... join in ,renew And exit the inner loop; if all loops have been traversed If no directly reachable node is found after that, then move the next node of the current node to the next node. join in and order Repeat the above steps until... At this point, the pruning path includes all the necessary turning points.

[0029] Understandably, the greedy pruning algorithm starts from the starting point and attempts to skip intermediate nodes and directly connect to the farthest reachable node, thus replacing the winding path with a straight line; the time complexity of this algorithm is O(n log n). However, it has high execution efficiency when the number of path nodes is small; discretized sampling detection determines whether a line segment collides with an obstacle by sampling points at equal intervals on the line and detecting whether each sampling point is located within the obstacle grid. The sampling step size of 0.05m achieves a balance between detection accuracy and computational efficiency; the number of path nodes after pruning is significantly reduced, the total path length is shortened, and the path tends to be straight, providing a better global path for subsequent local planning.

[0030] Isometric interpolation: Although the number of nodes on the pruned path is significantly reduced, the distances between adjacent nodes remain uneven. If the pruned path is directly passed to the DWA local planner, when the robot is far from the current target point, the local planner may blindly move in one direction and ignore nearby obstacles. When nodes are too close, frequent switching of the target point will cause control oscillations. Let the total length of the pruned path be... As shown in equation (2): In the formula: For pruning path nodes, The number of nodes is calculated based on the total path length and the preset interpolation interval. The pruned path is resampled into a sequence of equally spaced path points.

[0031] It is understandable that Equation (2) calculates the total length of the pruned path and is used to determine the parameters of the equidistant interpolation. The equidistant interpolation resamples the uneven path points after pruning into a sequence of path points with constant spacing, so that the distance between adjacent path points remains consistent. The preset interpolation interval is usually set to 0.3m to 0.5m, which can ensure sufficient resolution of the path to meet the passage requirements of narrow passages, and will not be too dense to increase the computational burden. The equidistant interpolation process ensures that the DWA local planner can move smoothly point by point when tracking the global path, avoiding control oscillations and frequent switching of target points caused by sudden changes in the spacing between path points.

[0032] For example, in this embodiment, the parameters of the improved BI-RRT-APF algorithm are set as follows: , , , The gravitational gain coefficient of the artificial potential field method is set as follows: The repulsion gain coefficient is set to The effective distance is set to The sampling and detection step size for greedy pruning is set to 0.05m; the preset interval for equidistant interpolation is set to 0.4m; when the starting point is... The target point is The traditional BI-RRT-APF algorithm plans a path with 42 nodes and a total length of 7.85m, with a search time of 1.23s. After adopting the dynamic probability target bias strategy of this invention, the search time is reduced to 0.87s, and the efficiency is improved by about 29%. After greedy pruning, the number of path nodes is reduced to 18, and the total path length is shortened to 7.12m. After equidistant interpolation, 18 uniform path points with a spacing of 0.4m are generated, reducing the number of target point switching times during DWA tracking to 17, and significantly reducing control oscillations.

[0033] Step S300: Use the improved DWA algorithm for local path planning.

[0034] Specifically, when using the improved DWA algorithm for local path planning, it includes: an accurate circular motion model, dynamic window construction and adaptive speed sampling, an accurate docking mechanism, and an adaptive weight evaluation function.

[0035] Precise circular motion model: Predicting robot trajectory using precise discretized motion equations from a differential drive model; when angular velocity... At that time, the pose is updated according to the circular motion formula. when At that time, the pose is updated according to the linear motion formula. Eliminate the accumulated error of the traditional Euler approximation and improve trajectory prediction accuracy.

[0036] Understandably, the differential drive model assumes that the speed difference between the robot's left and right wheels generates angular velocity motion, and is the most commonly used kinematic model for indoor mobile robots. The traditional DWA algorithm uses the Euler approximation method to update the pose, which assumes that the robot moves along a straight line in a short time. This approximation will produce significant cumulative errors when the speed is large or the prediction time is long. This invention uses an accurate circular arc motion equation to calculate the accurate trajectory of the robot moving along the circular arc when the angular velocity is not zero, thus eliminating the systematic error of the Euler approximation. For the straight line motion case (i.e., the angular velocity is zero), the standard straight line motion formula is used to update the pose. The accurate motion model significantly reduces the deviation between the predicted trajectory and the actual motion trajectory, thereby improving the accuracy of the evaluation function selection.

[0037] Dynamic window construction and adaptive velocity sampling: considering the robot's maximum linear acceleration Maximum angular acceleration Maximum linear velocity and maximum angular velocity Limitations, constructing a dynamic speed window Introducing a distance-adaptive deceleration strategy: based on the distance from the robot to the target point... Dynamically adjust the upper limit of the maximum linear velocity. in ;when At this time, the maximum speed decreases linearly with distance to prevent the robot from overshooting or overshooting.

[0038] Understandably, the dynamic velocity window determines the achievable velocity space for the next moment based on the robot's current state and physical constraints; considering maximum linear acceleration and angular acceleration constraints, it ensures that the sampled velocity combinations are within the reach of the robot's dynamics, avoiding sampling invalid velocities that cannot be executed; the distance-adaptive deceleration strategy enables the robot to automatically reduce its maximum speed limit as it approaches the target point, achieving deceleration and stopping behavior similar to human walking; when When the maximum speed decreases linearly to zero with distance, this gradual deceleration method is smoother than the emergency stop method of traditional DWA, which helps to improve the sense of security and comfort of elderly users.

[0039] Precise docking mechanism: when the robot is far from the target point At this time, it enters precise docking mode, abandons speed sampling, and adopts proportional control law; specifically, it includes: first calculating the heading error. ,like ,but , ;otherwise , Increase overshoot protection, if ,but ;when At that time, the speed returns to zero, completing a precise stop.

[0040] Understandably, the precise docking mechanism switches to a deterministic control law when the robot approaches the target point, avoiding the randomness and uncertainty caused by speed sampling. When the heading error is small, the robot moves directly towards the target point, steadily approaching at a low speed. When the heading error is large, the robot first adjusts its heading angle to align the front of the robot with the target point before moving forward. The overshoot protection mechanism detects if the current speed might cause the robot to overshoot the target point and immediately reduces the speed to a safe level. Finally, a docking distance threshold is set. It achieves millimeter-level precise positioning, meeting the needs of high-precision service scenarios such as intelligent elderly care robots delivering items and medicines.

[0041] Adaptive weight evaluation function: Evaluation function ,in Scoring is given based on the course. Score based on the distance to the target. Score points for speed. Scoring points for obstacles; based on Dynamically adjust weights: when hour, ;when When the weight is adjusted to By adjusting the weights, the robot prioritizes path efficiency when it is far from the target and docking accuracy when it is close to the target.

[0042] Understandably, the adaptive weighted evaluation function dynamically adjusts the relative importance of each evaluation index based on the distance between the robot and the target, causing the local planner to exhibit different behavioral characteristics at different stages. When the distance is far, the target distance score has the highest weight, guiding the robot to choose the path with the shortest total distance. When the distance is short, the heading score weight increases significantly, prompting the robot to prioritize adjusting its posture to align with the target point, preparing for precise docking. The speed score weight remains relatively stable, ensuring that the robot always prioritizes higher speeds to improve movement efficiency. The obstacle score weight is constant, ensuring that safe obstacle avoidance is always the primary constraint.

[0043] For example, in this embodiment, the robot parameters are set as follows: maximum linear velocity. Maximum angular velocity Maximum linear acceleration Maximum angular acceleration Velocity sampling resolution: 8 samples in the linear velocity direction and 11 samples in the angular velocity direction; Prediction time Discrete time step of the trajectory ;when When using long-distance weights Robot Proceeding at a relatively high speed; when When the weights switch to The robot's speed decreased to The accuracy of heading adjustment is improved; when the robot arrives Upon reaching the designated position, the precise docking mechanism is triggered, and the speed smoothly decreases to zero, ultimately resulting in a docking error of [missing value]. It meets the requirements for high-precision docking.

[0044] Step S400: Deeply integrate global path planning and local path planning to achieve collaborative navigation.

[0045] Specifically, the deep integration of global path planning and local path planning includes: discretizing the smooth path generated by the improved BI-RRT-APF algorithm into a series of path points; using the current path point to be tracked as the local target in each control cycle; automatically switching to the next path point as the new local target when the robot enters within 0.5m of the current path point; and switching the local target to the final target point after traversing all path points, triggering the precise docking stage. The key to the fusion algorithm is that the improved DWA no longer uses the final target point as the only local target, but instead sequentially tracks the smooth path points generated by the improved BI-RRT-APF, thus preserving the optimality of the global path while possessing real-time response capabilities in dynamic environments.

[0046] Understandably, the deep fusion mechanism transforms the traditional loosely coupled structure of global planning and local planning into a tightly coupled structure, enabling bidirectional information flow. By using the path point sequence provided by global planning as a sub-objective of local planning, local planning always has a clear short-term goal. This avoids the path optimization problem caused by local planning directly facing the final goal in traditional methods, while maintaining the real-time obstacle avoidance capability of local planning. The 0.5m target switching threshold is reasonably set, ensuring that the robot fully tracks the direction of the current path point without causing excessively frequent target switching due to an excessively large threshold. The point-to-point tracking mechanism enables the robot to flexibly cope with dynamic obstacles in complex environments. When a path point is temporarily obstructed, DWA can choose to detour to the next path point instead of globally replanning.

[0047] Stuck and escape module: Real-time monitoring of the robot's movement distance within a preset time interval. When the detected movement distance is less than a preset stagnation threshold, the robot is determined to be stuck and will perform a turning escape action, choosing a new direction to try to get out of trouble.

[0048] Understandably, the stuck escape module provides a solution for scenarios where robots get stuck, such as "dead ends" or "obstacles surrounding" common in elderly care environments. The preset time interval is usually set to 3 to 5 seconds, and the stagnation threshold is set to 0.1m to 0.2m depending on the robot size. When the robot is detected to have moved less than the threshold for an extended period, it is determined to be stuck. Possible causes include getting stuck in a chair leg array or being surrounded by moved obstacles. The turning escape action usually includes: rotating in place by a certain angle (such as 90° or 180°) and then trying to move in the new direction. If the robot still cannot escape after the first turn, it can try different directions multiple times until it successfully escapes. This module effectively improves the robot's ability to autonomously escape from complex environments and reduces the need for human intervention.

[0049] For example, in this embodiment, the total number of path points is 18, and the distance between adjacent path points is 0.4m; the target switching threshold is set to 0.5m, that is, when the robot is less than 0.5m away from the current path point, it automatically switches to the next path point; the stagnation detection time interval for the stuck escape module is set to 4s, and the stagnation threshold is set to 0.15m; the robot starts from the starting point After setting off, and changing waypoints 18 times, the final destination is reached. Nearby; when the robot is blocked by a suddenly appearing obstacle near the 12th waypoint, DWA plans a detour trajectory in real time, successfully avoids the obstacle, and continues to track subsequent waypoints; when the robot unfortunately gets stuck in a narrow passage, the stuck escape module detects the stagnation state, performs a 90° turn, and successfully escapes the stagnation, continuing to complete the tracking of the remaining path; throughout the entire navigation process, the path tracking rate reaches over 94%, and the precise docking error is controlled within 0.05m.

[0050] Step S500: Drive the robot chassis to move according to the linear velocity and angular velocity commands output by the fusion control module.

[0051] Specifically, the motion execution module drives the left and right wheel motors of the robot's differential chassis through the underlying drive controller based on the linear velocity and angular velocity commands output by the fusion control module to achieve the desired motion trajectory; the fusion control module coordinates the work of the global planning module and the local planning module, maintains the path point index, and realizes point-to-point tracking and target switching.

[0052] Understandably, the fusion control module is the central hub of the entire path planning system, responsible for coordinating the timing and data interaction of various functional modules; the path point index records the currently tracked path point number, and the index automatically increments as the robot reaches each path point; when the local planning detects dynamic obstacles, it maintains tracking of the global path while avoiding obstacles, realizing cooperative navigation in dynamic environments; the motion execution module receives linear velocity and angular velocity commands, and drives the left and right wheel motors through PID closed-loop control, so that the robot's actual motion trajectory is consistent with the planned trajectory.

[0053] In the above embodiments, a dynamic probabilistic target bias strategy enhances the convergence speed of the global planning algorithm in the early stages of the search and maintains a certain level of exploration capability in the later stages, effectively improving the planning efficiency and success rate of global paths in complex indoor environments. Greedy pruning and equidistant interpolation post-processing mechanisms remove redundant nodes, making the path shorter and straighter, ensuring the uniformity of path point density during DWA tracking. A precise circular arc motion model eliminates the accumulated error of the traditional Euler approximation, improving trajectory prediction accuracy. A distance-adaptive deceleration strategy and a precise docking mechanism enable the robot to dynamically adjust its speed according to the target distance and achieve millimeter-level precise docking when approaching the target, reducing the arrival error from 2.950 in traditional DWA. The m value was reduced to 0.000m; the weight of each evaluation index was dynamically adjusted according to the distance through an adaptive weight evaluation function, so that the robot could prioritize different motion targets at different stages; the path point tracking mechanism with deep fusion of global and local data not only preserved the optimality of the global path, but also had the ability to respond in real time to dynamic environments; the stuck escape module monitored the stagnation status in real time and executed turning and escape actions, which effectively reduced the number of times the robot got stuck. Simulation experiments showed that the number of stuck escapes was reduced by 53.8% to 79.3%, and the path tracking rate was as high as 94% or more, which improved the intelligent elderly care robot's ability to pass through narrow passages, dense obstacles and dynamic obstacles coexisting environments and the quality of service.

[0054] In another preferred embodiment based on the above embodiments, such as Figure 3 As shown, this embodiment provides a fusion path planning system for intelligent elderly care robots, including: The data acquisition module is configured to acquire a two-dimensional spatial map corresponding to the working environment of the elderly care robot, mark static obstacle areas and dynamic obstacle areas in the two-dimensional spatial map, and set the robot's starting position, target position and robot motion state parameters. The analysis module, electrically connected to the acquisition module, is configured to perform global path planning based on an improved bidirectional fast expanding random tree combined with an artificial potential field algorithm; wherein, The analysis module is also configured to build a bidirectional search tree based on the robot's starting position and target position, adopt a dynamic target bias strategy during path expansion, adjust the target position sampling probability according to the current search iteration state, and expand the search tree based on random sampling points; The control module is electrically connected to the analysis module. The control module is configured to perform path optimization processing on the initial path after obtaining the initial path, and to determine whether there is an obstacle collision by detecting the straight connection relationship between the path nodes. When there is no collision, the intermediate path nodes between the connecting nodes are deleted, and the optimized path is interpolated according to the preset path point interval to generate a global path point sequence. The control module is also configured to perform local path planning based on an improved dynamic window method, taking the current path point to be tracked as the local target, and establishing a dynamic velocity sampling window based on the robot's current motion state and motion constraints. The control module is also configured to generate multiple speed control parameters within the dynamic speed sampling window, predict the motion trajectory corresponding to each speed control parameter based on the robot motion model, evaluate the predicted motion trajectory based on the trajectory evaluation function, and select the corresponding speed control parameter as the local motion control quantity. The control module is also configured to fuse the global path point sequence with local motion control quantities, control the robot's movement according to the path point tracking order, and update the next path point as the local target when the robot's current position reaches the position range corresponding to the current path point. The local path planning process is executed cyclically until the robot reaches the target position.

[0055] It is understood that the fusion path planning method and system for intelligent elderly care robots in the above embodiments of this application have the same beneficial effects, and will not be described again.

[0056] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program goods. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program goods embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0057] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program goods according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0058] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0059] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0060] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A fusion path planning method for intelligent elderly care robots, characterized in that, include: Obtain a two-dimensional spatial map corresponding to the working environment of the elderly care robot, mark static obstacle areas and dynamic obstacle areas in the two-dimensional spatial map, and set the robot's starting position, target position and robot motion state parameters; Global path planning is performed based on an improved bidirectional fast expanding random tree combined with an artificial potential field algorithm; among which, A bidirectional search tree is established based on the robot's starting position and target position. During the path expansion process, a dynamic target bias strategy is adopted to adjust the target position sampling probability according to the current search iteration state and expand the search tree based on random sampling points. After obtaining the initial path, the initial path is optimized. The presence of obstacle collisions is determined by detecting the straight-line connection relationship between path nodes. When there is no collision, the intermediate path nodes between the connecting nodes are deleted. The optimized path is then interpolated according to the preset path point interval to generate a global path point sequence. Local path planning is based on the improved dynamic window method, which takes the current path point to be tracked as the local target and establishes a dynamic velocity sampling window according to the robot's current motion state and motion constraints. Multiple speed control parameters are generated within the dynamic speed sampling window. The motion trajectory corresponding to each speed control parameter is predicted based on the robot motion model. The predicted motion trajectory is evaluated based on the trajectory evaluation function, and the corresponding speed control parameter is selected as the local motion control quantity. The global path point sequence is fused with the local motion control variables. The robot is controlled to move according to the path point tracking order. When the robot's current position reaches the position range corresponding to the current path point, the next path point is updated as the local target. The local path planning process is executed cyclically until the robot reaches the target position.

2. The fusion path planning method for intelligent elderly care robots as described in claim 1, characterized in that, When constructing a two-dimensional spatial map of the working environment of elderly care robots, the following should be included: Point cloud data of the robot's surrounding environment is collected using a lidar sensor, and the point cloud data is filtered to remove outliers and noise points, resulting in processed environmental point cloud data. The processed environmental point cloud data is mapped to a two-dimensional coordinate space and divided into multiple grid units according to a preset grid size. A two-dimensional occupied grid map is generated based on the obstacle occupancy information in each grid unit. The two-dimensional occupied grid map includes obstacle areas and passable areas. The system acquires environmental image data based on visual sensors or depth cameras, performs target detection processing on the environmental image data, identifies dynamic obstacles in the environment, and obtains the category information, location information, and motion state information of the dynamic obstacles. The two-dimensional occupied grid map and the dynamic obstacle information are converted to a unified coordinate system and the map data is fused. The dynamic obstacle information is then loaded into the two-dimensional occupied grid map to generate an environmental map containing both static and dynamic obstacle information.

3. The fusion path planning method for intelligent elderly care robots as described in claim 2, characterized in that, When using an improved bidirectional fast expanding random tree combined with an artificial potential field algorithm for global path planning, the following steps are included: A bidirectional search tree is built based on the robot's starting position and target position, and the maximum number of iterations for path search and the current number of iterations are set. During the path expansion process, the target bias probability is determined based on the current iteration number and the maximum iteration number. The target bias probability decreases according to a preset change rule as the iteration number increases, and a minimum limit value for the target bias probability is set. During each path expansion iteration, a random number within a preset range is generated, and the random number is compared with the current target bias probability; When the random number is less than the current target bias probability, the target position is used as the current extended sampling point; when the random number is greater than or equal to the current target bias probability, an extended sampling point is generated according to the random sampling method, and the search tree is updated based on the extended sampling point. After the search tree generates an initial path from the starting position to the target position, the initial path is optimized.

4. The fusion path planning method for intelligent elderly care robots as described in claim 3, characterized in that, When performing path optimization on the initial path, the process includes: Obtain multiple path nodes in the initial path, select the candidate node farthest from the current node starting from the starting node, and determine whether the connecting line segment between the current node and the candidate node collides with an environmental obstacle, wherein: When the connecting line segment does not collide with an obstacle, delete the intermediate path node located between the connecting line segments and retain the candidate node; When the connecting line segment collides with an obstacle, a new candidate node is selected for connection determination. The path length is calculated based on the optimized path nodes, and the optimized path is resampled according to the preset path point interval to generate a path point sequence with equal spacing.

5. The fusion path planning method for intelligent elderly care robots as described in claim 3, characterized in that, When performing local path planning based on the improved dynamic window method, it includes: Acquire the robot's current pose and motion state information, and establish a trajectory prediction model based on the differential drive motion model; Predict the robot's motion state at the next moment based on candidate linear velocity and angular velocity parameters; Specifically, when the robot's angular velocity is not zero, the position coordinates and attitude angle at the next moment are calculated according to the current linear velocity, angular velocity, and the robot's current attitude angle, following a circular motion pattern; when the robot's angular velocity is zero, the position coordinates and attitude angle at the next moment are calculated according to the current linear velocity and the robot's current attitude angle, following a linear motion pattern. A dynamic velocity window is constructed based on robot motion constraint parameters, wherein the motion constraint parameters include maximum linear acceleration, maximum angular acceleration, maximum linear velocity, and maximum angular velocity. The range of values ​​for linear velocity and angular velocity is limited based on the motion constraint parameters, and multiple candidate velocity combinations are generated within the limited range. The robot's motion trajectory is predicted based on each candidate velocity combination, and the motion trajectory is evaluated based on the trajectory evaluation function. The corresponding velocity control parameter is then selected as the local motion control quantity.

6. The fusion path planning method for intelligent elderly care robots as described in claim 5, characterized in that, When evaluating the motion trajectory based on the trajectory evaluation function and selecting the corresponding velocity control parameter as the local motion control quantity, the following are included: During local path planning, the distance between the robot's current position and the target point is obtained, and the robot's maximum allowable linear velocity is adjusted based on this distance, wherein: When the distance between the robot's current position and the target point is greater than the preset deceleration distance, the preset maximum linear velocity is used as the speed limit value; When the distance between the robot's current position and the target point is less than or equal to the preset deceleration distance, the maximum allowable linear speed is reduced according to the distance change relationship, and the dynamic speed window is updated according to the adjusted speed limit value.

7. The fusion path planning method for intelligent elderly care robots as described in claim 6, characterized in that, When fusing the global path point sequence with local motion control variables, the following is included: The global path generated by the improved bidirectional fast expanding random tree combined with the artificial potential field algorithm is obtained, and the global path is discretized according to the preset path sampling rules to generate a sequence of path points with sequential relationship. Within the robot motion control cycle, the current path point to be tracked is determined from the path point sequence based on the robot's current position, and the current path point to be tracked is used as a local target in the dynamic window method local path planning; The distance between the robot's current position and the current path point to be tracked is obtained in real time, and the distance is compared with a preset path point switching threshold. When the distance between the robot's current position and the current path point to be tracked is less than or equal to the path point switching threshold, the next path point in the path point sequence is updated as the new local target, and local path planning continues. Once all path points in the path point sequence have been tracked, the final target position is used as the local planning target, and target position docking control is executed based on the positional relationship between the robot's current position and the final target position.

8. The fusion path planning method for intelligent elderly care robots as described in claim 7, characterized in that, It also includes stuck state detection and escape control, specifically including: During the robot's path tracking process, the robot's position information is acquired at preset time intervals, and the robot's corresponding movement distance is calculated based on the position changes within consecutive time periods. The distance traveled is compared with a preset stagnation threshold; When the moving distance is less than the stall threshold, the robot is determined to be stuck and the current path tracking control is paused. The robot's current posture information is acquired, and a steering control command different from the current direction of motion is generated to control the robot to perform steering motion. After performing the turning motion, the robot continues to monitor the distance it has moved. When the robot resumes its movement, it exits the stuck state and resumes path point tracking control.

9. The fusion path planning method for intelligent elderly care robots as described in claim 8, characterized in that, Acquire the robot's current posture information and generate steering control commands that differ from the current direction of motion. When controlling the robot to perform steering movements, this includes: The robot motion state detection time interval and the stagnation distance threshold are set, wherein the motion state detection time interval is set to 3 to 5 seconds, and the stagnation distance threshold is set to 0.1 to 0.2 meters; Acquire robot position change information within each motion state detection time interval and calculate the corresponding robot movement distance; When the moving distance is less than the stopping distance threshold, the robot is determined to be stuck. When the robot is stuck, a steering control command is generated based on the robot's current attitude angle to control the robot to perform a stationary rotation action, wherein the rotation angle of the stationary rotation action is set to 90 degrees to 180 degrees. After completing the rotation, control the robot to move in the direction of the rotation and reacquire the robot's movement distance; When the robot's movement distance is still less than the stall distance threshold, different steering control commands are regenerated, and the rotation action and movement state detection process are repeated until the robot's movement distance reaches or exceeds the stall distance threshold, at which point it exits the stuck state.

10. A fusion path planning system for intelligent elderly care robots, applicable to the fusion path planning method for intelligent elderly care robots described in claims 1-9, characterized in that, include: The data acquisition module is configured to acquire a two-dimensional spatial map corresponding to the working environment of the elderly care robot, mark static obstacle areas and dynamic obstacle areas in the two-dimensional spatial map, and set the robot's starting position, target position and robot motion state parameters. The analysis module, electrically connected to the acquisition module, is configured to perform global path planning based on an improved bidirectional fast expanding random tree combined with an artificial potential field algorithm; wherein, The analysis module is also configured to build a bidirectional search tree based on the robot's starting position and target position, adopt a dynamic target bias strategy during path expansion, adjust the target position sampling probability according to the current search iteration state, and expand the search tree based on random sampling points; The control module is electrically connected to the analysis module. The control module is configured to perform path optimization processing on the initial path after obtaining the initial path. It determines whether there is an obstacle collision by detecting the straight connection relationship between the path nodes. When there is no collision, the intermediate path nodes between the connecting nodes are deleted. The optimized path is then interpolated according to the preset path point interval to generate a global path point sequence. The control module is also configured to perform local path planning based on an improved dynamic window method, taking the current path point to be tracked as the local target, and establishing a dynamic velocity sampling window based on the robot's current motion state and motion constraints. The control module is also configured to generate multiple speed control parameters within the dynamic speed sampling window, predict the motion trajectory corresponding to each speed control parameter based on the robot motion model, evaluate the predicted motion trajectory based on the trajectory evaluation function, and select the corresponding speed control parameter as the local motion control quantity. The control module is also configured to fuse the global path point sequence with local motion control quantities, control the robot's movement according to the path point tracking order, and update the next path point as the local target when the robot's current position reaches the position range corresponding to the current path point. The local path planning process is executed cyclically until the robot reaches the target position.

Images