Multi-infrared fusion bed surface cleaning robot dynamic path planning system

Abstract

The invention discloses a multi-infrared fusion bed surface cleaning robot dynamic path planning system, relates to the technical field of intelligent bed surface cleaning and mobile robot path planning, and is used for solving the problems of low coverage efficiency, insufficient anti-falling safety and difficulty in flexible obstacle avoidance of a bed surface cleaning robot in a high bed complex bed surface scene. The bed surface cleaning robot constructs local grid environment representation and bed edge high-risk marks under a body coordinate system of the bed surface cleaning robot through calibration of a multi-path infrared obstacle avoidance sensor and a falling sensor and establishment of a unified time reference, and the local grid environment representation and the bed edge high-risk marks are fused with a static working grid map. Edge exposure intensity, flexible disturbance intensity and support stability are introduced to construct a multi-dimensional risk field and dynamic risk budget, linear velocity and angular velocity reference values are screened, and a rolling brush and a dust collector are driven to perform cleaning and sweeping.

Description

Technical Field

[0001] This invention relates to the field of intelligent bed cleaning and mobile robot path planning technology, and more specifically, to a dynamic path planning system for a bed cleaning robot using multi-infrared fusion. Background Technology

[0002] With the development of smart homes, bed hygiene and dust mite control have gradually gained attention, and robotic vacuum cleaners designed for floor environments are relatively mature. However, bed scenarios such as raised beds, narrow beds, thick mattresses, and folded bedding differ significantly from ordinary floors: one side is flush against the wall, while the other side is suspended; the mattress has a soft structure with localized sagging and bulging areas; the edges of multiple layers of bedding naturally droop, forming flexible overhangs; and power outlets and bent, drooping charging cables are often found near bedside tables, creating a combination of rigid obstacles, flexible obstacles, and areas with significant drops on the bed surface. Existing bed cleaning devices mostly employ a single-path mode or simple anti-fall strategies, relying solely on drop sensors with fixed thresholds and control logic that slows down upon approaching obstacles. This makes it difficult to balance safety and efficiency in scenarios with complex bed edges and flexible overhangs. On the one hand, when bed cleaning robots make repeated tentative movements near the edge of the bed, existing methods lack the cumulative measurement of "long-term edge exposure risk," which can easily lead to the robot gradually approaching the high-risk area of ​​the bed edge under the cumulative effect of multiple cycles, posing a risk of falling. On the other hand, to avoid tangling with blanket corners and cables, a simple retreat strategy relying on the nearest obstacle distance can easily cause the robot to repeatedly circle around the bed corners and move at extremely low speeds for extended periods, resulting in uncleaned blind spots and overly conservative approaches. In addition, traditional path planning methods are often based on static obstacle models or single cost functions. In the dynamic environment of a bed surface with frequent mattress deformation and flexible swaying, they lack a comprehensive measurement and decision-making mechanism for multi-dimensional risks such as edge exposure, flexible disturbances, and support stability. It is difficult to finely adjust linear and angular velocities, and difficult to achieve high-coverage cleaning under limited power and time constraints.

[0003] To address the above problems, this invention proposes a solution. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a dynamic path planning system for a bed cleaning robot based on multi-infrared fusion, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] In a preferred embodiment, it includes: a whole machine configuration initialization module, a multi-infrared drop sensing modeling module, a dynamic path planning control module, a cleaning execution closed-loop update module, and signal connections between the modules;

[0007] The overall configuration initialization module is used to complete the structural installation and electrical connection of the overall components of the bed cleaning robot after it is powered on. The main control board is configured with parameters for multiple infrared obstacle avoidance sensors and multiple drop sensors.

[0008] The multi-infrared fall perception modeling module is used to synchronously collect data from multiple infrared obstacle avoidance sensors and multiple fall sensors through the main control board, and generate a local environment representation with obstacle occupancy probability and bed edge risk marking;

[0009] The dynamic path planning control module is used to construct a multi-dimensional risk field by maintaining edge exposure intensity index, flexible disturbance intensity index and support stability index for grid cells through the main control board. It performs integrated analysis on the risks, sweeping coverage and sweeping benefits of candidate linear velocity and angular velocity, and eliminates candidate velocities with excessive risks by combining the dynamically updated risk budget and determines the reference values ​​of linear velocity and angular velocity.

[0010] The cleaning execution closed-loop update module is used by the main control board to perform differential drive and closed-loop update based on the reference values ​​of linear velocity and angular velocity, and to link the roller brush, vacuuming and ultraviolet disinfection lamp to complete the cleaning.

[0011] In a preferred embodiment, after being powered on, the bed cleaning robot completes the structural installation and electrical connection of the shell, front and side multi-channel infrared obstacle avoidance sensors, bottom drop sensors, roller brush and dust collection unit, ultraviolet disinfection lamp, left and right drive wheels, lithium battery and voice speaker components. The main control board records the installation position, installation posture and bed edge height determination parameters for each multi-channel infrared obstacle avoidance sensor and each pair of drop sensors, and loads kinematic constraints such as wheel radius, wheel track, maximum running speed and minimum turning radius.

[0012] In a preferred embodiment, after the power self-test and link handshake are normal, the sensor sampling period and the start-up constraints of the actuator are set to generate parameterized power-on initialization results.

[0013] In a preferred embodiment, the main control board synchronously collects ranging information from multiple infrared obstacle avoidance sensors and reflection height information from multiple fall sensors according to the control cycle. The ranging results from the multiple infrared obstacle avoidance sensors are linearly calibrated and smoothed, and combined with the installation posture, they are converted into obstacle observation points in the body coordinate system of the bed cleaning robot. The obstacle observation points are written into the discrete grid environment model around the bed cleaning robot, and the obstacle occupancy probability of each grid is recursively updated according to obstacle observation and idle observation. At the same time, based on the bed edge detection results of multiple fall sensors in multiple consecutive control cycles, the corresponding grids are marked as impassable bed edge areas. The main control board outputs a local environment representation containing obstacle occupancy probability and bed edge risk marking under a unified time reference.

[0014] In a preferred embodiment, the main control board establishes a working grid map over the entire bed surface based on the local environment representation output by the multi-infrared fall perception modeling module, and divides the area into passable and inaccessible grids according to the mattress outline and the area outside the bed. It then overlays obstacle occupancy probability and bed edge risk markers to construct a comprehensive grid map.

[0015] On the integrated grid map, the main control board uses a heuristic search path planning algorithm with obstacle penalty and bed edge penalty to generate a reference trajectory from the current position of the bed cleaning robot to the target cleaning grid that satisfies the passage constraint and minimum turning constraint, and replans the reference trajectory in a local area when it is affected by new obstacles or new bed edge risks.

[0016] Meanwhile, based on dynamic window velocity sampling and short-time trajectory prediction, the main control board selects reference values ​​of linear velocity and angular velocity that meet obstacle avoidance and kinematic constraints from candidate velocities that satisfy velocity and acceleration constraints, and drives the bed cleaning robot to complete the cleaning along the reference trajectory.

[0017] In a preferred embodiment, based on the comprehensive grid map output by the multi-infrared fall perception modeling module, a multi-dimensional risk field is constructed to fit the scenario of raised beds, flexible bedding, and drooping cables by additionally maintaining edge exposure intensity, flexibility disturbance intensity, and support stability indices for each passable grid cell. Within each control cycle, the bed cleaning robot performs forward trajectory prediction for each set of candidate linear and angular velocities based on dynamic window velocity sampling. The three types of risk indices corresponding to the grid cells traversed by the predicted trajectory are accumulated over time to obtain the comprehensive risk accumulation of the candidate speed within the current prediction time window. The cleaning execution closed-loop update estimates the number of cleaning coverage units and speed utilization rate that the candidate speed can add within the prediction time window, and calculates the corresponding cleaning benefits.

[0018] In a preferred embodiment, the main control board dynamically updates the risk budget based on the dwell time of the bed cleaning robot in the safe area, the bed edge buffer zone, and the flexible high-risk area. All candidate speeds with a total cumulative risk exceeding the current risk budget are eliminated. Only the optimal set of linear and angular velocities from the remaining candidate speeds is selected as the speed reference values ​​for the current control cycle in descending order of cleaning benefits. Under the premise of ensuring bed edge safety and avoiding cable entanglement, the cleaning speed and coverage efficiency are adaptively balanced.

[0019] In a preferred embodiment, the main control board calculates the target rotational speed of the left and right drive wheels based on the linear velocity and angular velocity reference values ​​given by the dynamic path planning control module, and performs closed-loop control under maximum speed and acceleration / deceleration constraints to continuously estimate the pose of the bed cleaning robot.

[0020] In a preferred embodiment, the roller brush speed, suction power, and UV disinfection lamp working status are adjusted in conjunction with the current linear speed, remaining battery energy, and cleaning progress. The environmental grid and cleaning coverage marks are continuously updated using multiple infrared obstacle avoidance sensors and drop sensors, and the cleaning coverage rate is calculated. When the coverage rate reaches the set target or the battery power drops to a safe threshold, an exit path is generated, and the bed cleaning robot is controlled to exit the bed in an orderly manner and complete the cleaning task.

[0021] The technical effects and advantages of the multi-infrared fusion dynamic path planning system for bed cleaning robots of the present invention are as follows:

[0022] This invention optimizes both hardware and algorithms for raised bed scenarios through the collaborative design of a whole-machine configuration initialization module, a multi-infrared fall perception modeling module, a dynamic path planning and control module, and a cleaning execution closed-loop update module. At the hardware level, parameterized calibration of multiple infrared obstacle avoidance sensors and bottom fall sensors, along with a unified time reference configuration, enables the bed cleaning robot to obtain refined distance, height, and bed edge information in its body coordinate system. This is combined with a gridded environment representation to construct a unified expression of traversable and high-risk areas. At the algorithm level, based on traditional A* path search and dynamic window velocity sampling, three types of indicators are introduced: edge exposure intensity, flexible disturbance intensity, and support stability. This forms a multi-dimensional risk field that conforms to the characteristics of the bed edge, overhanging corners, and drooping cables. A risk budget B(k) constrains the comprehensive risk of candidate velocity pairs, thereby optimizing the motion control quantity that balances coverage increment and speed utilization while ensuring fall prevention and avoiding entanglement. The cleaning execution closed-loop update module updates the integrated grid map M(k) and cleaning coverage matrix in real time, so that path planning and cleaning progress form a closed loop, which significantly improves the full coverage capability, operational safety and environmental adaptability in complex bed environment, reduces uncleaned dead spots and improves user experience. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of a dynamic path planning system module for a bed cleaning robot using multi-infrared fusion according to the present invention.

[0024] Figure 2 This is a timing diagram of a dynamic path planning system for a bed cleaning robot using multi-infrared fusion according to the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] Example: This invention discloses a dynamic path planning system for a bed cleaning robot based on multi-infrared fusion, such as... Figure 1 As shown, it includes: a whole machine configuration initialization module, a multi-infrared drop sensing modeling module, a dynamic path planning and control module, a cleaning execution closed-loop update module, and signal connections between the modules.

[0027] like Figure 2 As shown, in the overall configuration initialization module, the bed cleaning robot first completes the power-on initialization of the overall structure configuration and parameterization. The bed cleaning robot includes a shell, multi-channel infrared obstacle avoidance sensors located at the front and sides of the shell, a drop sensor located at the bottom front edge of the shell, a roller brush, roller brush motor, ultraviolet disinfection lamp, suction port, suction motor, and dust cup air duct assembly located in the front half of the shell, left and right drive wheels and their drive motors, a lithium battery, and a rear air duct located in the rear half of the shell, and a transparent dust cup and speaker located at the top of the shell. It also includes a main control board installed inside the shell. All the above components are electrically connected to the signal and power terminals of the main control board via wiring harnesses, enabling the main control board to collect data from the sensors and perform unified scheduling and control of the left and right drive wheels, roller brush motor, suction motor, ultraviolet disinfection lamp, and speaker.

[0028] During the structural configuration phase, the front and side infrared obstacle avoidance sensors are fixed to the front and side edges of the housing in a predetermined arc or U-shaped array. Each infrared obstacle avoidance sensor corresponds to a unique installation angle and installation coordinate point. The main control board pre-records the installation position vector and installation attitude angle for each infrared obstacle avoidance sensor in non-volatile memory, establishing a one-to-one mapping relationship between the sensor index number and the installation position vector and installation attitude angle, ensuring that the raw distance values ​​output by each sensor can be converted into the polar coordinate information of the obstacle in the body coordinate system of the bed cleaning robot.

[0029] Several pairs of downward-looking infrared transceivers are arranged along the bottom edge of the housing as drop sensors. The detection areas of each pair of transceivers overlap and cover the bed surface area below the leading edge of the bed cleaning robot. The main control board also records the installation position index and corresponding bed edge height determination parameters for each pair of drop sensors, which is used to distinguish between the central area and the edge area of ​​the bed surface during operation.

[0030] In terms of execution unit configuration, the roller brush is mounted in the opening area of ​​the front half of the housing via a roller brush motor. The outer periphery of the roller brush is equipped with striking blades or bristles. The output shaft of the roller brush motor is fixed to the roller brush via a key or spline connection, allowing the roller brush to rotate at high speed around its own axis to strike the bed surface. The suction port is located behind the roller brush or adjacent to its axis. The air inlet of the suction motor is connected to the suction port via a dust cup duct, and the air outlet of the suction motor is connected to a rear duct, which in turn connects to the exhaust port at the rear of the housing. The dust cup is installed inside the housing near the air inlet of the suction motor. A removable filter is installed inside the dust cup. The filter's pore size and material were selected during the prototype stage through comparative tests of mite removal efficiency and wind resistance loss, ensuring airflow while intercepting dust and mite particles stirred up from the bed surface. An ultraviolet disinfection lamp is located in the bottom area near the roller brush, its radiation coverage corresponding to the effective cleaning width of the roller brush. The power supply of the ultraviolet disinfection lamp is connected to the main control board via a dedicated drive circuit.

[0031] Regarding the drive unit configuration, the left and right drive wheels are respectively mounted on the left and right sides of the rear half of the housing, and are connected to the main control board via drive motors. The main control board stores the wheel radius r, wheel track L, and maximum allowable linear velocity of the drive wheels in non-volatile memory. and maximum angular velocity Based on kinematic parameters, a mapping relationship between the angular velocities of the left and right drive wheels and the linear velocity v and angular velocity ω of the bed cleaning robot is established using a differential drive model. Specifically, during the control cycle... Inside, the main control board records the angular velocities of the left and right drive wheels as... and The linear velocity and angular velocity of the bed cleaning robot in the bed surface plane satisfy the following constraints: v=(r / 2)·( + ); ω=(r / L)·( - );

[0032] Where v represents the instantaneous linear velocity of the bed cleaning robot within the bed surface, ω represents the instantaneous angular velocity of the bed cleaning robot about an axis perpendicular to the bed surface, r represents the radius of the left and right drive wheels, and L represents the center distance between the left and right drive wheels. During power-on initialization, the main control board uses the above relationships to... and The allowed value range is calculated in reverse to ensure that |v|≤ 0 during the entire operation. With |ω|≤ Speed ​​safety constraints; and at the same time according to and Further derive the minimum turning radius And record the minimum turning radius as one of the constraints in the dynamic path planning stage;

[0033] During the power and interface initialization phase, the lithium battery is installed in the battery compartment at the rear of the casing and connected to the main control board, motor drive circuit, sensor circuit, and ultraviolet disinfection lamp drive circuit via the power management circuit. When the user places the bed cleaning robot on the bed and presses the power button on the main control board, the lithium battery begins to power the entire machine. After the main control board completes its reset, it first performs a power self-test, determining whether the current battery level is higher than the preset cleaning task power threshold by collecting voltage and current information output by the power management circuit. .

[0034] If the available energy corresponding to the remaining power is detected Less than The main control board will then announce a low battery warning via its voice speaker and keep all execution components off; if Greater than or equal to Then the main control board continues to perform connectivity detection between the sensor link and the execution link.

[0035] During the connectivity test, the main control board sequentially sends handshake frames to each infrared obstacle avoidance sensor, drop sensor, roller brush motor, vacuum cleaner motor, drive wheel motor, ultraviolet disinfection lamp drive circuit, and voice speaker, and reads the response flags returned by each component. Only when all components return a normal flag within the specified time window will the main control board mark the system status as hardware ready.

[0036] After completing the hardware readiness check, the main control board loads the calibration parameters of the multiple infrared obstacle avoidance sensors and the height threshold parameters of the drop sensor from the non-volatile memory. Among them, the calibration parameters of the multiple infrared obstacle avoidance sensors include the zero-point deviation, linear correction coefficient, and attitude transformation parameters between each infrared obstacle avoidance sensor and the body coordinate system of the bed cleaning robot.

[0037] Subsequently, during the power-on initialization phase, the main control board constructs a lookup table or calibration coefficient set based on these parameters to ensure that the original ranging value output by any infrared obstacle avoidance sensor is consistent with the original ranging value. At runtime, it can be done via d=a· The linear correction relationship +b is converted into an approximate true distance d, where a is the linear scaling factor obtained through calibration, and b is the zero-point offset correction amount obtained through calibration. The height threshold of the drop sensor is preset based on the typical bed surface height difference. When the equivalent reflection height calculated by a certain downward-looking infrared drop sensor during operation is lower than... At that time, the main control board will determine the area covered by the sensor as the edge of the bed or the step area.

[0038] Based on the aforementioned hardware structure and calibration parameters, the main control board establishes a unified time base and configures the data sampling period Δt of the multiple infrared sensors and drop sensors during the final stage of power-on initialization, setting Δt as the subsequent dynamic path planning control period. Integer frequency division or equivalent values ​​are used to align sensor sampling with motion control in time. Simultaneously, the main control board allocates an independent data buffer and timestamp queue for each sensor channel to record data at each discrete time k. The sensor ranging results and corresponding attitude parameters are used; at the same time, the initial acceleration slope and maximum duty cycle are set for the control output channels of the left and right drive wheels, roller brush motor and vacuum motor to ensure that the bed cleaning robot can start smoothly under the premise of meeting speed constraints and power constraints when the user issues a cleaning start command.

[0039] After completing the above parameterized power-on initialization, the main control board broadcasts standby information through the voice speaker, guiding the user to select the full-coverage dynamic path planning cleaning mode or the local enhanced cleaning mode. After receiving the mode selection and start command, the system status is switched to prepare for multi-infrared data acquisition and fusion modeling.

[0040] In the multi-infrared fall perception modeling module, after the overall configuration initialization module completes the structural configuration and parameterized power-on initialization of the bed cleaning robot, the main control board initiates the data acquisition and fusion modeling process of the multi-infrared obstacle avoidance sensor and fall sensor under a unified time reference. This is used to construct a local environment representation in the body coordinate system of the bed cleaning robot that can be used for dynamic path planning. Specifically, the main control board sets the control cycle to... In each discrete time k· The front and side sections are triggered according to the sampling period Δt set in the overall configuration initialization module. Road infrared obstacle avoidance sensor and bottom The road drop sensors operate synchronously, ensuring that each sensor acquires distance or height information within the same time window. For the i-th infrared obstacle avoidance sensor, the main control board first reads its raw ranging value output during the k-th control cycle via an analog-to-digital converter. Then, the calibration relationship established in the overall system configuration initialization module is called to... Linear conversion to calibration distance That is, adopt The transformation relationship, where For the linear scaling factor of the i-th sensor, The above is the zero-point offset correction amount for this sensor. and The data has been written into the non-volatile memory of the main control board during the prototype calibration stage.

[0041] To mitigate the impact of single-sampling noise and surface reflection fluctuations on the ranging results, the main control board performs a conversion of the distance for each channel. Further exponential smoothing filtering is performed. Specifically, the main control board maintains a smoothing distance for each infrared obstacle avoidance sensor. The update is performed in the k-th control cycle according to the following recursive relationship: ; in, λ represents the smoothed distance of the i-th sensor in the previous control cycle, and λ is a smoothing coefficient between 0 and 1. When λ is close to 1, the weight of historical measurements is emphasized, and when λ is close to 0, the weight of the current measurement is emphasized.

[0042] Furthermore, during the prototype stage, the main control board selected the value range of λ based on typical bed surface materials and sensor noise characteristics, and stored λ as an operating parameter in non-volatile memory. After obtaining the smoothing distance... Subsequently, the main control board further limited the outlier values, when Less than the preset minimum detection distance When, set it as ,when Greater than the preset maximum detection distance When, set it as Or it can be considered as barrier-free measurement; Next, based on the installation orientation of each infrared obstacle avoidance sensor recorded in the overall configuration initialization module, the main control board establishes a local geometric model in the body coordinate system of the bed cleaning robot. For the i-th infrared obstacle avoidance sensor, its installation azimuth angle is denoted as in the body coordinate system of the bed cleaning robot. Then, during the k-th control cycle, the coordinates of the obstacle observation point corresponding to the sensor on the bed coordinate plane can be approximately expressed as: ; ; in, This represents the coordinate components along the forward direction in the body coordinate system of the bed cleaning robot. This represents the coordinate components along the lateral direction. The main control board uses the geometric center of the bed cleaning robot or the midpoint of the left and right drive wheel axes as the origin of the coordinate system, with the forward direction as the positive x-axis and the leftward direction as the positive y-axis. Through the above coordinate transformation, the multi-channel infrared ranging results are uniformly mapped to the same coordinate system, thereby forming a set of obstacle observation points within the current control cycle. .

[0043] Synchronized with the infrared obstacle avoidance sensor, bottom The drop sensor on each path also performs a reflection height detection under the control of the main control board during the k-th control cycle. For the j-th drop sensor, the main control board reads its original reflection signal from the analog-to-digital converter circuit and calculates the equivalent height. and the height threshold set in the overall configuration initialization module. Comparison. When Greater than or equal to At that time, the main control board marks the area covered by the drop sensor as the presence of the bed surface. Less than At that time, the area is marked as a bed edge or step.

[0044] To enhance the robustness of drop detection, the main control board introduces binary logic variables for drop status. Agreement This indicates that the edge of the bed has been detected. This indicates that no bed edge was detected, and statistics were compiled over several consecutive control cycles. The time series, only when A forced deceleration or stop command is triggered only when the number of consecutive cycles is 1 within a preset number of cycles, thereby avoiding unnecessary shutdowns caused by false detections in a single cycle.

[0045] Furthermore, to enable dynamic path planning to utilize sensor data with limited computing resources, the main control board constructs a discretized local grid environment model around the bed cleaning robot. Specifically, the main control board defines a rectangular region in the bed cleaning robot's body coordinate system, with the geometric center of the bed cleaning robot as the origin, covering a certain range in front and to the left and right. This region is divided into several grid cells according to fixed spatial resolutions Δx and Δy, with (m,n) representing the grid cell in the m-th row and n-th column, and an obstacle occupancy probability is maintained for each grid cell. After the overall configuration initialization module is completed, the main control board sets the initial occupancy probability of all grid cells to... ,in The prior idle probability is less than 0.5.

[0046] In each control cycle Inside, the main control board first... , For each infrared observation point, find the grid cell it belongs to. , Then, based on the current ranging results, the obstacle occupancy probability of the grid cell is updated. For ease of calculation, the main control board uses a recursive update model in log-likelihood form, expressing the obstacle occupancy state of the grid cell using the log-likelihood value. This indicates that, during the initialization phase, according to calculate In the k-th control cycle, when the smoothed distance of the i-th infrared obstacle avoidance sensor is detected... Within the effective detection range , And coordinates , Falling into the grid ( , When this happens, the main control board updates the log-likelihood of the grid according to the following relationship: ; raster index , This represents the log increment of positive evidence for a single obstacle observation. This represents the logarithmic increment of idle evidence along the sensor beam path. The main control board is selected based on the reliability of the infrared obstacle avoidance sensor in detecting typical bed surface obstacles during prototype testing. and The numerical value, and will be along the beam from the bed cleaning robot body to ( , The log-likelihood of the grid cells traversed between them decreases. This is to reflect the fact that these grids are more likely to be empty space. After updating all infrared observation points, the main control board... The relation transforms the log-likelihood into a barrier occupancy probability between 0 and 1, where The closer the value is to 1, the more likely the grid cell is to be occupied by an obstacle. The closer to 0, the more likely the grid is to be free space.

[0047] In parallel with the obstacle occupancy probability update, the main control board will update the output of the bottom drop sensor. Mapped onto one or more rows of grid cells near the leading edge of the bed cleaning robot. For grid cells located below the coverage area of ​​the drop sensor, if within several consecutive control cycles... If the value is determined to be 1, the main control board has a probability of occupying that grid cell. The value is forcibly increased to a high value close to 1, while an impassable and fall-prevention-priority risk marker is recorded for the area in an additional edge risk marker array.

[0048] At the same time, at the end of each control cycle, the main control board will display the grid occupancy probability matrix for the current moment. Together with the edge risk label array, it is encapsulated into a local environment representation E(k), and E(k) is associated with and stored with the pose estimation result of the bed cleaning robot itself; In the dynamic path planning control module, after the multi-infrared fall perception modeling module completes the data acquisition and fusion modeling of multiple infrared obstacle avoidance sensors and fall sensors, and constructs a local grid environment representation E(k) in the body coordinate system of the bed cleaning robot, the main control board performs control in each control cycle. The process of dynamic path planning and obstacle avoidance control quantity generation based on multi-infrared fusion results is initiated internally to obtain reference values ​​of linear velocity and angular velocity that meet the bed surface constraints and avoid obstacles and bed edges. , ; Specifically, when the user selects the full-coverage dynamic path planning cleaning mode in the overall configuration initialization module, the main control board first establishes a static working grid map on the entire area of ​​the bed surface. , The grid resolution is consistent with Δx and Δy used in the multi-infrared fall sensing modeling module to construct the local environment representation E(k). Based on the mattress outline dimensions and the boundary positions of the head and foot of the bed, the main control board maps the passable area of ​​the bed surface into a set of passable grid cells. At the same time, areas clearly located outside the bed are marked as inaccessible sets. .

[0049] Subsequently, during actual operation, the main control board will use the local occupancy probability matrix obtained from the multi-infrared drop sensing modeling module. And edge risk marker array aligned and mapped to the current pose of the bed cleaning robot Above, the static map is superimposed with the local dynamic environment E(k). Incremental updates are performed to form a comprehensive grid map M(k) for the current moment. In this comprehensive map, static furniture, bed legs, and other fixed obstacles, as well as dynamic obstacles and bed edges obtained based on multi-infrared fusion and fall sensing, are all marked with a uniform occupancy probability or risk label. To achieve systematic coverage of the bed surface area and avoid high-risk areas, the main control board constructs a reference path using a heuristic search path planning algorithm on the comprehensive grid map M(k). Specifically, the main control board denotes the grid cell where the bed surface cleaning robot is located as the starting grid s, the target coverage grid cell of the current planning round as the ending grid g, and uses the A* algorithm to search for feasible paths from s to g. For any intermediate grid node n in the path search, the main control board defines its cost function F(n) as: F(n) = G(n) + H(n) + C(n); Where G(n) represents the cumulative path cost from the starting grid s to grid n along the current candidate path, H(n) represents the heuristically estimated cost from grid n to the ending grid g, and C(n) represents the additional penalty term related to the obstacle occupancy probability and edge risk of grid n. G(n) can be accumulated by the Euclidean distance between adjacent grids, that is, for each grid... The main control board moves to the grid n with a step size Δs. The form of recursive update, where Here, H(n) is the path length weighting coefficient. H(n) can be estimated using the Euclidean distance between grid n and the endpoint grid g, i.e. ,in , Let n be the coordinates of the grid. , Let g be the coordinates of the grid. Heuristic weights. The additional penalty term C(n) is related to the occupancy probability in the multi-infrared fall sensing modeling module. and edge risk marking Related. Within each control cycle, the main control board defines the additional penalty term for grid n as: ; in, This represents the obstacle occupancy probability of grid n at the current time. This is a marker variable indicating whether the grid belongs to a high-risk area at the edge of the bed. This indicates that the area is a high-risk zone around the bed. This indicates that the area around the bed is not a high-risk area; The penalty weight is assigned to obstacles. The weight for bed edge risk penalty. When the grid n Close to 1 or When the value is equal to 1, the corresponding C(n) is significantly increased, thus causing the A* search process to avoid the grid or only choose to pass through the grid when there are no other feasible paths.

[0050] Subsequently, when performing an A* search, the main control board only allows the path to be expanded to meet the occupancy probability. Less than the preset passage limit Furthermore, within the grid set that does not violate the safe distance from the bed edge, the minimum turning radius is pre-derived by the overall configuration initialization module. The geometric feasibility of the transition actions between grids is checked to ensure that the obtained path satisfies constraints such as obstacles and safety edges, and can be realized by the actual kinematics of the differential drive wheel.

[0051] After obtaining the discrete grid path from s to g, the main control board converts the grid path into a continuous reference trajectory in the bed coordinate system through interpolation and spline smoothing, and represents the reference trajectory as a two-dimensional function under the path parameter u. , Furthermore, the desired heading angle for each path point is calculated in the body coordinate system of the bed cleaning robot. .

[0052] The main control board in each control cycle The internal position is estimated based on the current position of the bed cleaning robot. , , ) and nearest reference path point ( , , The deviation is used to determine the current path segment to be tracked, and this information is passed to the local obstacle avoidance and velocity sampling control process so that the robot can move along the reference path direction as much as possible while taking obstacles into account. After the bed cleaning robot completes path tracking of the current target coverage area, the main control board updates the endpoint grid g to the next uncovered partition grid, and repeats the path planning and trajectory generation in the same way until the entire bed area is covered.

[0053] To maintain the safety and smoothness of the path even in the presence of dynamic obstacles or rapid changes in the local environment, the main control board checks the validity of the current reference path in each control cycle based on E(k) generated by the multi-infrared drop sensing modeling module. When an occupancy probability corresponding to a certain segment of the reference path at the current moment is detected... Exceeding the safety threshold If the path segment crosses a new high-risk area along the edge of the bed, the main control board immediately triggers a local path replanning process near that segment. The affected path segment is treated as a local problem between the start and end points of the new path planning, and a local A-search is rerun to generate a new local reference trajectory segment. This segment is then seamlessly stitched onto the original path, allowing the bed cleaning robot to bypass newly appearing obstacles or risk areas and continue completing the cleaning task. Through this combination of global path planning and local replanning, the dynamic path planning process can flexibly respond to environmental changes without completely overturning the existing coverage strategy.

[0054] After obtaining the reference trajectory and local update results, the main control board initiates a local obstacle avoidance and velocity sampling control process based on the dynamic window concept, transforming the reference trajectory tracking problem into real-time evaluation and selection of linear velocity and angular velocity control variables v and ω. In the k-th control cycle, the main control board first initializes the velocity constraints in the initialization module according to the overall machine configuration. , With acceleration constraints , Construct a dynamic window in the current velocity space Where v(k−1) and ω(k−1) are the actual linear velocity and angular velocity output in the previous control cycle. The acceleration and deceleration limits are defined as linear velocity limits, and the acceleration constraints ensure that the bed cleaning robot will not experience excessively large speed changes within a single control cycle; To set the angular velocity acceleration / deceleration limits, ensuring smooth steering maneuvers. When constructing the dynamic window, the main control board operates within the range [...]. , ]and[ , Within the range, v and ω are discretely sampled according to a predetermined resolution to obtain several candidate velocity pairs. , ).

[0055] For each pair of candidate velocities in the dynamic window ( , The main control board is in the current pose of the bed cleaning robot. , , Based on this, forward trajectory prediction is performed, with a prediction time of [missing information]. ,in This represents the prediction step number. In the l-th prediction step, the main control board updates the model using the following pose: ; ; ; in, , , As the initial state for prediction , , Let represent the pose estimation of the bed cleaning robot at prediction step l. Within each prediction step, the main control board maps the position corresponding to the predicted pose to the current integrated grid map M(k), and reads the occupancy probability of several grid cells near that position. With edge risk marking To determine the candidate speed The Future · Whether a collision with an obstacle or bed edge will occur within a given timeframe. If, during a prediction step, the probability of grid occupancy on the predicted path exceeds a collision threshold... Or exist The high-risk grid at the edge of the bed will be marked as infeasible by the main control board; It's important to note that when a bed-cleaning robot is deployed on a raised bed or narrow bed with one side against a wall and the other side suspended, the bed surface is typically covered with a thick mattress and multiple layers of bedding. Near the head of the bed are bedside tables, power outlets, and tangled charging cables. Because the mattress is soft, there are areas of sagging and bulging. The edges of the multiple layers of bedding naturally droop, forming flexible overhangs. Some power cables also dangle from the edges of the bedside tables to near the bed edge, creating a combination of rigid obstacles, flexible obstacles, and areas of significant drop around the bed. When performing full-coverage cleaning, the bed-cleaning robot must ensure efficient coverage of the central area while maintaining sufficient safety margins near the bed edge, overhanging corners of bedding, and dangling cables to prevent falls, entanglement, or pulling on the power cables. Simultaneously, it cannot move slowly along the bed edge at extremely low speeds for extended periods, thus completing the cleaning task within the limited time and battery power.

[0056] In such environments, mattress deformation, and the swaying of bedding and cables can cause changes in the distance and location of obstacles observed by multiple infrared obstacle avoidance sensors within different time windows. The risks associated with bed edges and flexible overhangs exhibit dynamic changes with posture and time. If only linear velocity and angular velocity are selected as single factors, it is easy to encounter problems such as gradually approaching the high-risk area of ​​the bed edge under the cumulative effect of multiple cycles, while frequently and significantly slowing down to avoid it in order to maintain distance, creating uncleaned blind spots at the bed corners.

[0057] Therefore, in this embodiment, in the dynamic path planning control module, the main control board introduces a multi-dimensional risk budget and cleaning benefit coupling strategy for the aforementioned bed surface scenario, based on the traditional dynamic window speed sampling. Specifically, based on the local environment representation E(k) and comprehensive grid map M(k) output by the multi-infrared fall perception modeling module, the main control board additionally maintains an edge exposure intensity index for each passable grid cell (m,n). Flexible disturbance intensity index and supporting stability indicators .

[0058] Among them, the edge exposure intensity index By statistically analyzing the correlation between this grid and high-risk bedside markers within a preset time window... The cumulative duration of the disturbance is obtained from the time spent in the adjacent or bed edge buffer zone; flexible disturbance intensity index. This is obtained by calculating the variance or frequency of infrared ranging values ​​near the grid within a time window, and is used to characterize the swaying activity of the suspension angle and the drooping cable; support stability index. Then, based on the equivalent height sequence measured by the drop sensor when the bed cleaning robot passes through the area. The calculation of short-term changes shows that when the local mattress deformation is gentle and the support is stable... When the altitude approaches 1, it fluctuates drastically within a short period of time. The reduction is used to characterize the reliability of the soft mattress in supporting the posture of the bed cleaning robot.

[0059] During the initialization phase, the main control board normalizes the above three indicators to the [0,1] interval, and updates them in each control cycle based on the movement of the bed cleaning robot and sensor data. Internal recursive refresh, thereby reducing the probability of traditional obstacle occupancy. High-risk markings at the edge of the bed In addition, a multi-dimensional risk field is formed that fits the scenarios of raised beds, flexible bedding, and cables.

[0060] In each control cycle, the main control board constructs a dynamic window based on the speed and acceleration constraints set by the overall configuration initialization module. And generate several candidate velocity pairs ( , After that, instead of relying solely on the nearest obstacle distance or target direction deviation for selection, it considers each pair of candidate velocities ( , Perform forward trajectory prediction, setting the prediction step size to... The predicted pose is obtained in the l-th prediction step. , , ) and mapped to raster cells ( , The main control board reads the edge exposure intensity of the corresponding grid from M(k) in each prediction step. Flexible disturbance strength and supporting stability indicators And construct the overall risk accumulation of this candidate velocity pair over the entire prediction time window: ; in , , These represent the weighting coefficients for edge risk, flexible entanglement risk, and support instability risk, respectively, with Δt being the prediction step size. Simultaneously, the main control board updates the cleaning coverage matrix maintained by the cleaning execution closed-loop update module. Estimate the candidate velocity in relation to the number of new sweeping coverage units that may be added within the prediction time window. Based on this, a cleaning revenue function is defined: ; in To cover the return weight, Weighted by speed utilization rate This represents the total number of accessible grid cells on the bed surface.

[0061] The main control board further maintains a dynamically changing risk budget B(k) at the system level. This risk budget is updated based on the time the bed cleaning robot spends operating within the safe area and the time it spends in the bed edge buffer zone and flexible high-risk areas. For example, it is updated in each control cycle according to... Recursion in the form of, in , , These represent the duration of the current cycle within the safe zone, the bed edge buffer zone, and the flexible high-risk zone, respectively. , , This is the corresponding adjustment coefficient;

[0062] The clip(·) function is used to limit the risk budget value within the parentheses to a preset range. , Within ], when the calculation result is less than Time to take When the calculation result is greater than Time to take This is to avoid numerical divergence in risk budgets.

[0063] Meanwhile, when filtering candidate speeds, the main control board first eliminates all those that meet the requirements. From the candidate velocity pairs, only the set of velocity pairs whose overall risk is acceptable under the current risk budget constraint is retained; then, the sweep payoff function is selected from this set. The largest candidate velocity pair ( , (This serves as a reference value for the linear velocity and angular velocity in this cycle.) , This technology enables the robot to adapt to and balance cleaning speed and coverage efficiency in complex bed environments where raised beds and flexible hanging objects coexist. This is achieved while ensuring safety from falling off the bed edges and avoiding cable entanglement. It significantly improves the robot's path planning creativity and environmental adaptability in complex bed scenarios.

[0064] In obtaining and Then, based on the differential drive mapping established in the overall configuration initialization module, the main control board calculates the target angular velocity that the left and right drive wheels should achieve in the current control cycle. and Specifically, according to: ; ; The main control board solves the problem through algebraic transformations. and In conjunction with the maximum angular velocity and acceleration limits of the drive motor, for and Amplitude clipping and smooth acceleration / deceleration are performed to generate the final control command for the left and right drive wheels in the kth control cycle.

[0065] It should be noted that the above-mentioned dynamic path planning and local obstacle avoidance control quantity generation process is repeated in each control cycle, enabling the bed cleaning robot to continuously generate motion control quantities that meet both the reference trajectory tracking requirements and the safety constraints of obstacles and bed edges, supported by a multi-infrared fusion environment model.

[0066] In the cleaning execution closed-loop update module, the dynamic path planning control module outputs the linear velocity and angular velocity reference values ​​for the current control cycle. , The target angular velocities of the left and right drive wheels are calculated through differential mapping. , After that, the bed cleaning robot performs each control cycle. The system performs cleaning operations along a dynamic path while simultaneously updating the environmental representation and cleaning coverage status in a closed loop to ensure that obstacle avoidance and fall prevention constraints are always met while completing full coverage cleaning of the bed surface.

[0067] Specifically, at the start of the k-th control cycle, the main control board calculates the path using the dynamic path planning control module. , As a speed command input, the motor drive circuit is invoked to perform pulse width modulation control on the drive motors of the left and right drive wheels. By adjusting the duty cycle of the drive voltage at both ends of the windings of the left and right drive motors, the actual angular velocity is... , Try to get as close as possible to the desired angular velocity , To suppress speed errors caused by load disturbances and changes in the bed surface friction coefficient, the main control board can employ a proportional-integral control algorithm within the motor drive layer, adjusting the angular velocity error based on the speed within each sub-sampling cycle. , The control increment is calculated to reduce steady-state error while ensuring response speed. During the rotation of the left and right drive wheels at the commanded angular velocities, the actual linear velocity and angular velocity of the bed cleaning robot in the bed coordinate system are determined in real time by the differential relationship, as follows: ; ; Where r is the radius of the drive wheel, and L is the center distance between the left and right drive wheels. At the end of each control cycle, the main control board updates the pose of the bed cleaning robot according to the above v(k) and ω(k), and updates the estimated pose at the end of the k-th control cycle ( , , The recursive calculation is performed as follows: ; ; ; Thus, without relying on external positioning equipment, continuous pose estimation of the bed cleaning robot relative to the bed surface can be achieved solely by the kinematics of the drive wheels; Synchronized with the drive unit's actions, the main control board, within the cleaning execution closed-loop update module, controls the roller brush motor, vacuum motor, and ultraviolet disinfection lamp to work collaboratively according to preset cleaning condition parameters. Before entering the cleaning trajectory tracking stage, the main control board sets the target rotation speed of the roller brush based on the bed surface material and the user-selected cleaning mode. Set the target power level of the vacuum cleaner motor to After entering the cleaning execution closed-loop update module, in each control cycle, the main control board performs closed-loop adjustment of the roller brush motor speed according to the current linear velocity v(k). When it detects that v(k) is close to... When the speed of the roller brush approaches or nears the preset high-speed threshold, the speed of the roller brush is controlled at... The upper limit range is maintained to ensure sufficient number of beats on the bed surface during high-speed travel; when v(k) is detected to drop to a low speed or close to zero speed, the roller brush speed is slightly reduced to reduce fabric wear caused by localized repeated beating. The vacuum cleaner motor is controlled by adjusting the motor drive duty cycle to maintain a certain level. The corresponding negative pressure level ensures that dust and mites stirred up by the roller brush are fully sucked into the dust cup through the suction port and dust cup air duct. The main control board simultaneously adjusts the cleaning progress based on the bed surface and remaining battery power. In a timely manner Fine-tuning was performed when the remaining energy approached the threshold. However, before cleaning is complete, slightly reduce the vacuuming power to extend the battery life while maintaining basic mite removal effect.

[0068] The control of the ultraviolet disinfection lamps follows the principle of synchronous irradiation of the coverage area. The main control board estimates the position of the lamps based on the position of the lamps. , , When the bed cleaning robot enters a new un-illuminated section based on the reference trajectory progress, the ultraviolet disinfection lamp is turned on and the corresponding path length or time length is determined. or The system keeps the UV disinfection lamps active, ensuring the bed surface is irradiated with UV light after being brushed and vacuumed. When the bed cleaning robot repeatedly passes over an already irradiated area, the main control board can choose to keep the UV disinfection lamps off to reduce unnecessary UV accumulation. To achieve this control, the main control board adds an irradiation status marker to each bed surface grid cell on the integrated grid map M(k), and determines whether to turn on the UV disinfection lamps by comparing the status of the markers at the current location of the grid cell. Throughout the process, the voice speaker broadcasts status information based on instructions from the main control board at key stages such as entering cleaning, low battery, and cleaning completion, reminding the user of the current work stage.

[0069] While the bed cleaning robot performs cleaning along a dynamic path, the cleaning execution closed-loop update module continuously calls the multi-infrared obstacle avoidance sensor and fall sensor acquisition process in the multi-infrared fall perception modeling module to perform closed-loop updates on the local environment representation E(k) and the comprehensive grid map M(k). Specifically, in the k-th control cycle, as the pose of the bed cleaning robot ( , , The change in distance, the smooth distance measured by multiple infrared obstacle avoidance sensors Corresponding obstacle coordinates ( , The grid is remapped to the origin of the new body coordinate system of the bed cleaning robot, and the main control board updates the grid occupancy probability according to the geometric transformation and log-likelihood recursion method given in the multi-infrared fall perception modeling module. Bed edge risk marking This reflects the distribution of obstacles and bed edges within the new field of view after the bed cleaning robot moves. When the bed cleaning robot approaches the bed edge or a newly appearing obstacle, the corresponding grid... or Significant changes will occur, and these changes will be immediately utilized in the dynamic path planning and obstacle avoidance control variable generation process of the dynamic path planning control module in the next control cycle, so that the calculated... , It automatically veers away from high-risk areas, achieving real-time closed-loop coupling between the path and the environment.

[0070] To characterize the cleaning coverage status, the main control board maintains a cleaning coverage marker for each accessible grid cell (m,n) on the integrated grid map M(k). During initialization, the sweep coverage markers for all bed surface grids are set to... =0 indicates an uncleaned state. During the operation of the cleaning execution closed-loop update module, the main control board updates the status of the bed cleaning robot at the end of each control cycle based on the current pose of the robot. , The effective cleaning width of the roller brush is used to mark the grid cells that intersect with the projected area of ​​the roller brush as cleaned, i.e., the corresponding... Set to 1. To improve the stability of coverage information, the main control board can employ a multiple coverage confirmation strategy within a single brush width. Only when the same grid cell is determined to be within the brush coverage area for several consecutive control cycles will the coverage be confirmed. The coverage matrix is ​​reset from 0 to 1 to filter out instantaneous coverage jitter caused by small errors in pose estimation. As the control cycle progresses, the coverage matrix is ​​cleaned. The system will gradually evolve from a state of all zeros to a state close to all one. At the end of each path planning round, the main control board calculates the current bed surface cleaning coverage η(k), for example, through... The calculation is performed in the following manner, where This indicates the total number of sweep coverage markers marked as 1 in all accessible grid cells on all beds. This represents the total number of accessible grid cells on the bed surface. When η(k) is greater than or equal to the preset target coverage... At that time, the main control board considers the cleaning task of the current bed surface area to be completed.

[0071] When the current coverage rate is detected to be reached Or the remaining energy of the lithium battery Reduced to a safe threshold The following is the main control board initiating the cleaning task's finalization process within the cleaning execution closed-loop update module. Specifically, the main control board first selects a retreat path on the integrated grid map M(k) that returns the cleaning robot from its current position to the center of the safe area on the bed or a location easily accessible to the user. A simplified A* algorithm is used to generate this retreat path without increasing the cleaning area. Subsequently, within several control cycles, the system outputs the generated retreat path. , The system controls the bed cleaning robot to gradually move away from the edge of the bed and stop at a preset position. During the retreat, the main control board gradually reduces the control inputs of the roller brush motor and the vacuum motor, slowly reducing the roller brush speed and vacuum power from their operating values ​​to zero, and preemptively turning off the ultraviolet disinfection lamp to avoid unnecessary irradiation in non-cleaned areas. When the bed cleaning robot reaches the preset stopping position and the angular speed of the left and right drive wheels... , When the system is controlled to zero, the main control board shuts down all actuators, maintaining only power management and necessary standby circuits. It then announces the completion of the cleaning process via a voice speaker and displays the final coverage matrix. Integrated raster map M(k) Several key parameters are recorded in the internal memory for subsequent maintenance or algorithm parameter optimization.

[0072] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0073] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0074] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and inventive constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0075] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0076] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0077] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A dynamic path planning system for a bed cleaning robot using multi-infrared fusion, characterized in that, include: The system includes a whole-machine configuration initialization module, a multi-infrared drop sensing modeling module, a dynamic path planning and control module, a cleaning execution closed-loop update module, and signal connections between the modules. The overall configuration initialization module is used to complete the structural installation and electrical connection of the overall components of the bed cleaning robot after it is powered on. The main control board is configured with parameters for multiple infrared obstacle avoidance sensors and multiple drop sensors. The multi-infrared fall perception modeling module is used to synchronously collect data from multiple infrared obstacle avoidance sensors and multiple fall sensors through the main control board, and generate a local environment representation with obstacle occupancy probability and bed edge risk marking; The dynamic path planning control module is used to construct a multi-dimensional risk field by maintaining edge exposure intensity index, flexible disturbance intensity index and support stability index for grid cells through the main control board. It performs integrated analysis on the risks, sweeping coverage and sweeping benefits of candidate linear velocity and angular velocity, and eliminates candidate velocities with excessive risks by combining the dynamically updated risk budget and determines the reference values ​​of linear velocity and angular velocity. The cleaning execution closed-loop update module is used by the main control board to perform differential drive and closed-loop update based on the reference values ​​of linear velocity and angular velocity, and to link the roller brush, vacuuming and ultraviolet disinfection lamp to complete the cleaning.

2. The dynamic path planning system for a bed cleaning robot based on multi-infrared fusion according to claim 1, characterized in that: After being powered on, the bed cleaning robot completes the structural installation and electrical connection of the shell, front and side multi-channel infrared obstacle avoidance sensors, bottom drop sensors, roller brush and dust collection unit, ultraviolet disinfection lamp, left and right drive wheels, lithium battery and voice speaker components. The main control board records the installation position, installation posture and bed edge height judgment parameters for each multi-channel infrared obstacle avoidance sensor and each pair of drop sensors, and loads kinematic constraints such as wheel radius, wheel track, maximum running speed and minimum turning radius.

3. The dynamic path planning system for a bed cleaning robot based on multi-infrared fusion according to claim 2, characterized in that: After the power self-test and link handshake are normal, the sensor sampling period and the start-up constraints of the actuator are set to generate parameterized power-on initialization results.

4. The dynamic path planning system for a bed cleaning robot based on multi-infrared fusion according to claim 1, characterized in that; The main control board synchronously collects ranging information from multiple infrared obstacle avoidance sensors and reflection height information from multiple drop sensors according to the control cycle. It performs linear calibration and smoothing filtering on the ranging results of multiple infrared obstacle avoidance sensors and converts them into obstacle observation points in the body coordinate system of the bed cleaning robot in combination with the installation posture. The obstacle observation points are written into the discrete grid environment model around the bed cleaning robot and the obstacle occupancy probability of each grid is updated recursively according to obstacle observation and idle observation. At the same time, based on the bed edge detection results of multiple drop sensors in multiple consecutive control cycles, the corresponding grid is marked as a bed edge area that cannot be passed. The main control board outputs a local environment representation containing obstacle occupancy probability and bed edge risk marking under a unified time reference.

5. The dynamic path planning system for a bed cleaning robot based on multi-infrared fusion according to claim 1, characterized in that: The main control board establishes a working grid map within the entire bed surface area based on the local environment representation output by the multi-infrared fall perception modeling module. It divides the area into passable and inaccessible grids according to the mattress outline and the area outside the bed, and overlays obstacle occupancy probability and bed edge risk markers to construct a comprehensive grid map. On the integrated grid map, the main control board uses a heuristic search path planning algorithm with obstacle penalty and bed edge penalty to generate a reference trajectory from the current position of the bed cleaning robot to the target cleaning grid that satisfies the passage constraint and minimum turning constraint, and replans the reference trajectory in a local area when it is affected by new obstacles or new bed edge risks. Meanwhile, based on dynamic window velocity sampling and short-time trajectory prediction, the main control board selects reference values ​​of linear velocity and angular velocity that meet obstacle avoidance and kinematic constraints from candidate velocities that satisfy velocity and acceleration constraints, and drives the bed cleaning robot to complete the cleaning along the reference trajectory.

6. The dynamic path planning system for a bed cleaning robot based on multi-infrared fusion according to claim 5, characterized in that: Based on the comprehensive grid map output by the multi-infrared drop sensing modeling module, a multi-dimensional risk field is constructed to fit the scenarios of raised beds, flexible bedding, and drooping cables by additionally maintaining edge exposure intensity index, flexible disturbance intensity index, and support stability index for each passable grid cell. Within each control cycle, the bed cleaning robot performs forward trajectory prediction for each set of candidate linear and angular velocities based on dynamic window velocity sampling. It accumulates the three types of risk indicators corresponding to the grid cells traversed by the predicted trajectory over time to obtain the comprehensive cumulative risk of the candidate speed within the current prediction time window. The cleaning execution closed-loop update estimates the number of cleaning coverage units and speed utilization rate that the candidate speed can add within the prediction time window, and calculates the corresponding cleaning benefits.

7. The dynamic path planning system for a bed cleaning robot based on multi-infrared fusion according to claim 6, characterized in that: The main control board dynamically updates the risk budget based on the dwell time of the bed cleaning robot in the safe area, the bed edge buffer zone, and the flexible high-risk area. All candidate speeds with a total risk exceeding the current risk budget are eliminated. Only the optimal set of linear and angular velocities from the remaining candidate speeds is selected as the speed reference values ​​for the current control cycle, based on the cleaning benefits from high to low. Under the premise of ensuring bed edge safety and avoiding cable entanglement, the cleaning speed and coverage efficiency are adaptively balanced.

8. The dynamic path planning system for a bed cleaning robot based on multi-infrared fusion according to claim 1, characterized in that: Based on the linear and angular velocity reference values ​​provided by the dynamic path planning control module, the main control board calculates the target rotational speeds of the left and right drive wheels through the differential drive relationship and performs closed-loop control under maximum speed and acceleration / deceleration constraints to continuously estimate the pose of the bed cleaning robot.

9. The dynamic path planning system for a bed cleaning robot based on multi-infrared fusion according to claim 8, characterized in that: Based on the current linear speed, remaining battery energy, and cleaning progress, the robot adjusts the roller brush speed, suction power, and UV disinfection lamp status in a coordinated manner. It continuously updates the environmental grid and cleaning coverage markers using multiple infrared obstacle avoidance sensors and drop sensors, and calculates the cleaning coverage rate. When the coverage rate reaches the set target or the battery level drops to a safe threshold, it generates an exit path, controls the bed cleaning robot to exit the bed in an orderly manner, and completes the cleaning task.

Images