A home-based elderly care service robot

Abstract

The application relates to a home-based family service robot, characterized by comprising a cleaning base, a main body and a head with a touch screen from bottom to top. The main body comprises a plurality of storable dishes, and a separate compartment with a mainboard is arranged in the main body; when the dishes are stored, heating and / or heat preservation can be carried out in the main body; and the robot can find a path to the service object according to the current position of the robot in the path and walk to the service object. Path calculation and long and short memory model training are used, so that the robot can intelligently operate household items and the cared object at a fixed time, deliver the items and communicate through voice, and efficient assistance is realized for the life and living of the old people.

Description

Technical Field

[0001] This invention relates to home service robots, and in particular to a home-based elderly care service robot, belonging to the field of intelligent service robots. Background Technology

[0002] Existing service robots rely on a dual infrastructure of data storage and remote operation, or are equipped with the ability to deliver items (such as groceries and water), essentially providing services on demand. However, in reality, elderly people may not be able to operate robots, or may even have lost their ability to express themselves and communicate, requiring unattended operation. This is especially true before the development of robots that can assist with household items (such as cooking and cleaning) and proactively control the elderly person's physical movements (such as helping, lifting, moving, temporarily maintaining a posture, feeding, toileting, and bathing). How can robots intelligently understand the different lifestyles of elderly people, pay attention to their physical and mental state, and dynamically adjust their assistance strategies while still maintaining a human-centric approach? Current technologies have not addressed this issue. Summary of the Invention

[0003] In view of the above-mentioned problems in the prior art, the present invention takes the following aspects into consideration: first, it introduces basic cleaning and delivery functions into the robot; second, it adopts intelligent algorithms to learn the elderly’s living habits and thus adopt corresponding service strategies; and third, it communicates remotely with personnel to collect corresponding human services, thereby saving human effort.

[0004] Based on the above considerations, the present invention provides a home-based elderly care service robot, comprising, from bottom to top, a cleaning base, a main body, and a head with a touch screen, wherein...

[0005] The cleaning base includes a shell, a cleaning structure, a walking mechanism, a dust collection chamber, a water tank, a sludge suction box, multiple camera devices on the side of the cleaning base, and a first intelligent chip for performing 3D modeling and path calculation based on images captured in real time by the camera devices. The main body includes multiple retractable plates and has a compartment with a separate main board inside. When the plates are retracted, the main body can be heated and / or kept warm inside. The robot will then find a path to reach the service recipient based on its current position and walk to the service recipient.

[0006] The motherboard is equipped with a second intelligent chip, which can perform cleaning work at set times, remotely control the household appliances needed by the service recipients at set times, and communicate remotely with personnel to remind them of the service time and the current physiological state of the service recipients through a pre-trained model. Personnel can modify the cleaning work and the time of remote control operation of household appliances through a touch screen. The motherboard periodically retrains the model based on the modified historical data, and if the timing result obtained based on the newly trained model is inconsistent with the previous modification record, it remotely informs the personnel to manually determine whether to select the timing result obtained by the newly trained model or the previous modification record, so as to remotely execute the remote control operation of household appliances and / or cleaning work.

[0007] Optionally, the motherboard has a voice dialogue module that can recognize the voice of the service recipient and the voice of the personnel, so that when the service recipient and / or personnel have a voice dialogue, the voice is given higher priority than the pre-trained model and the remote execution, so as to perform the cleaning work and / or remote control operation.

[0008] Optionally, voice conversations between personnel can include both in-person and remote conversations.

[0009] Optionally, the cleaning base includes omnidirectional wheels controlled by the motherboard, which first perform dry sweeping and then wet sweeping. The cleaning structure includes a roller brush that can perform both dry and wet sweeping. The water tank is connected to the water inlet pipe via an electronic valve controlled by the motherboard to introduce water into the surface of the roller brush for wetting. The dust collection bin is located above the roller brush, and a shovel extends from one end of the dust collection bin to shovel off debris from the roller brush and continuously accumulate it in the dust collection bin. A water collection trough is formed between the suction box and the roller brush. Wastewater in the water collection trough is sucked out by a water pump controlled by the motherboard and located on one side of the water tank via a water suction pipe. The water suction pipe starts from the water collection trough, passes through the suction box, and connects to the suction pipe. The section of the water suction pipe that passes through the inside of the suction box has an opening to allow the sucked water to overflow and fall into the suction box. The water pump outlet returns water to the suction box.

[0010] Preferably, a filter is installed at the joint connecting the water pump to the water pump and / or in the return water section that extends into the sludge suction box.

[0011] Optionally, the camera device is equipped with a distance sensor to define an initial position, so that the omnidirectional wheel controls the cleaning base to perform a first rotation, collects images captured by the camera device before and after the first rotation, as well as distance data collected during the first rotation, and continues to complete a second rotation, completing one full rotation, and similarly continues to collect distance data collected during the second rotation. The first intelligent chip performs three-dimensional modeling based on the images and all distance data.

[0012] Preferably, the plurality of camera devices are evenly distributed in a circle, and the first rotation angle is half of the acute central angle between the positions of two adjacent camera devices and the center of the circle.

[0013] Optionally, the step of performing 3D modeling based on the image and all distance data includes:

[0014] S1 establishes a three-dimensional rectangular coordinate system in the modeling software, locates all distance data based on the center position of the base, and connects each location point in sequence.

[0015] S2 is in a three-dimensional Cartesian coordinate system, in which it identifies whether the object belongs to furniture based on the images before and after the first rotation, locates the furniture according to the center position of the captured circle, and builds a three-dimensional furniture body according to the preset height estimate of the furniture.

[0016] S3 determines the projection outline of the three-dimensional furniture on the ground, and establishes an initial path in the non-furniture projection outline area, and walks along the initial path. Every preset time period, stops continuing to proceed with steps S1-S2, and continues to establish more paths until the scanning is completed, obtaining all non-furniture projection outline areas.

[0017] S4 finds the shortest path to the vicinity of the current location of the service object among all established paths, and defines this path as the fastest path to reach the current location.

[0018] The initial path and other paths are established by the user within the non-furniture projection outline area displayed on the touch screen or smartphone, and after determination, the main board controls the omnidirectional wheels to make the robot walk according to the path.

[0019] It is easy to understand that, with the establishment of a three-dimensional rectangular coordinate system and the determination of the current center position of the base, the walking parameters can be easily determined based on the equation of the path.

[0020] Preferably, after a path is established, if the motherboard calculates and finds that a collision will occur when walking along the established path based on the current furniture and non-furniture projection outline areas, it will prompt the user to redetermine a new path until the calculation result indicates that no collision will occur.

[0021] Understandably, the distance between symmetrical points on both sides of the vertical path is calculated based on the maximum distance along the robot's shoulder width. This serves as a benchmark to determine whether a collision is possible. It's generally best to leave an additional 2-5cm of clearance, just in case.

[0022] Optionally, users can set a starting point for the robot, which will return to the starting point to wait after completing all tasks.

[0023] Optionally, the method for pre-training the model includes the following steps:

[0024] Q1. Obtain the timestamps of on-site user settings, voice history, and remote modifications on the touchscreen, and divide the same type of control commands for each day into training and validation sets.

[0025] Q2 establishes a long short-term memory model, inputs the control command training set into different model network units in chronological order, calculates the loss function, continuously optimizes the network parameters, uses a validation set for verification, and stops training when the loss function value is minimized, thus obtaining the pre-trained model.

[0026] Optionally, the method of finding a path to the service object based on the current position of the robot and walking to the service object is to draw a straight line from the current position to the starting point of the shortest path in step S4. If the motherboard calculates that a collision will occur in the straight line connection, the user can adjust the calculation through the touch screen or remote smartphone until a collision is calculated. Beneficial effects

[0027] By utilizing path calculation and long short-term memory model training, the robot can intelligently and regularly operate household items and deliver items to those being cared for, as well as communicate via voice, thus providing efficient assistance for the elderly's daily life. Attached Figure Description

[0028] Figure 1 This invention provides a schematic diagram of a home-based elderly care service robot, including a bottom view of the base, a side view, and a side view of the main body and part of the base shell.

[0029] Figure 2 A schematic diagram showing the positional changes of the camera device before and after rotation and the range of its front and rear fields of view.

[0030] Figure 3 A schematic diagram of the 3D modeling of the floor and the establishment of the shortest path.

[0031] Figure 4 A simplified diagram of the long and short memory model training. Detailed Implementation

[0032] Figure 1 This invention discloses a home-based elderly care service robot, comprising, from bottom to top, a cleaning base, a main body, and a head with a touch screen, wherein...

[0033] The cleaning base includes a T-shaped shell, a cleaning structure inside the T-shaped shell, omnidirectional wheels, a dust collection bin, a water tank, a sludge suction bin, multiple camera devices on the side of the cleaning base, and a first intelligent chip for performing 3D modeling and path calculation based on images captured in real time by the camera devices. The main body includes four retractable plates and has a compartment with a separate mainboard inside. When the plates are retracted, the main body can be heated and kept warm inside. The robot will then find a path to reach the service recipient based on its current position and walk to the service recipient.

[0034] The motherboard is equipped with a second smart chip (not shown in the figure), which can perform cleaning work at set times, remotely operate the household appliances needed by the service recipients at set times, and communicate remotely with personnel to remind them of the service time and the current physiological state of the service recipients. Personnel can modify the cleaning work and the time of remote operation of household appliances through the touch screen. The motherboard retrains the model periodically based on the modified historical data, and if the timing result obtained based on the newly trained model is inconsistent with the previous modification record, it remotely informs the personnel to manually determine whether to select the timing result obtained by the newly trained model or the previous modification record, so as to remotely execute the remote operation of household appliances and cleaning work.

[0035] Specifically, the motherboard also has a voice dialogue module, which can recognize the voice of the service recipient and the voice of the personnel collected, so that when the service recipient and the personnel have a voice dialogue in on-site or remote mode, the voice dialogue takes priority over the model and the remote execution to carry out the cleaning work and remote control operation.

[0036] The cleaning structure includes Figure 1 The omnidirectional wheels, controlled by the mainboard, perform dry sweeping followed by wet sweeping. The cleaning structure includes a roller brush that performs both dry and wet sweeping. A water tank is connected to a water inlet pipe via an electronic valve controlled by the mainboard, introducing water to wet the roller brush surface. A dust collection bin is located above the roller brush, with a scraper extending from one end to remove debris from the roller brush and continuously accumulate it in the bin. A water collection trough is formed between the suction box and the roller brush. Wastewater in the collection trough is pumped out by a water pump, also controlled by the mainboard and located on one side of the water tank, through a water suction pipe. The water suction pipe starts from the water collection trough, passes through the suction box, and connects to the suction pipe. The section of the water suction pipe passing through the inside of the suction box has an opening to allow the pumped water to overflow into the suction box. The water pump outlet returns water to the suction box. A filter (not shown in the figure) is installed at the joint connecting the water suction pipe and / or the return section inside the suction box.

[0037] like Figure 1Viewed from the front and back, the control lines for the water pump and electronic valve are designed to pass through the outer casing and connect to the main board inside the main body. The serving tray is either rotating or a drawer-style retractable or pull-out mechanism. A camera is also installed in the head to recognize the frontal facial image of the person being cared for, allowing AI to transform it into a dynamic image. This can reassure some patients with severe dementia, who may mistake the robot for a family member, thus increasing the quality of service. A rotating disk is located between the main body and the T-shaped shell to facilitate the camera's aiming and capture of the user's face.

[0038] The heating and keeping of food on the plate can be set via the touch screen.

[0039] like Figure 2 As shown, each of the four equally divided circular camera devices is equipped with a distance sensor (not shown in the figure). An initial position A is defined, causing the omnidirectional wheel to control the cleaning base to perform a first rotation, i.e., a 45° clockwise rotation. The four cameras collect images before and after the first rotation, as well as distance data collected during the first rotation. The system then continues with a second rotation, completing one full revolution, and similarly continues to collect distance data during the second rotation. The first intelligent chip performs a 3D model based on the images and all distance data. Thus, the field of view of adjacent cameras and the focal point of the cameras after the first rotation are P and P', to capture images of the entire perimeter of circle O as comprehensively as possible.

[0040] Among them, such as Figure 3 As shown, the steps for performing 3D modeling based on the image and all distance data include:

[0041] S1 establishes a three-dimensional rectangular coordinate system in the modeling software (its XY plane is established on the floor), locates all distance data based on the center position of the base, and connects each location point in sequence; in the figure, starting from position a, the distance sensor scans the walls of the living room and kitchen, as well as the legs of the sofa and TV cabinet, and the areas not scanned are still represented by thin lines.

[0042] S2 is in a three-dimensional Cartesian coordinate system, in which the image before and after the first rotation will be used ( Figure 2 (As shown at the top), it identifies whether it belongs to furniture, locates the furniture based on the center position of the circle captured, and builds a three-dimensional furniture body according to the furniture's preset height (generally not exceeding 90cm, even if the furniture height exceeds 90cm, it will not affect the recognition).

[0043] S3 projects three objects onto the ground to obtain the furniture projection outline. In the non-furniture projection outline area, an initial path is established and the object walks along the initial path. Every preset time interval (usually 2-4 seconds), the object stops and arrives at position b in the figure. At this time, steps S1-S2 are continued, and more paths are established until the scan is completed, obtaining all non-furniture projection outline areas.

[0044] S4 finds the shortest path among all established paths leading to the vicinity of the current location of the service object, and defines this path as the fastest path to reach the object based on its current location. The path indicated in the diagram is this shortest path, leading to the bedside.

[0045] Once a path is established, the motherboard calculates and determines that following the established path will result in a collision based on the current furniture and non-furniture projected outline areas. If this happens, the user is prompted to determine a new path, and this process continues until the calculated path avoids collisions. Furthermore, the initial path and other paths are determined by the user within the non-furniture projected outline areas displayed on the touchscreen or smartphone. After determination, the motherboard controls the omnidirectional wheels to move the robot according to the path.

[0046] like Figure 4 As shown, the method for pre-training a model includes the following steps:

[0047] Q1. Obtain the timestamps of on-site user settings, voice history, and remote modifications on the touchscreen, and divide the same type of control commands for each day into training and validation sets.

[0048] Q2 establishes a Long Short-Term Memory (LSTM) model. The control command training set is input into different network units of the model in chronological order, and the loss function is calculated. Network parameters are continuously optimized, and validation is performed using a validation set. Training stops when the loss function value is minimized, resulting in a pre-trained model. Validation is performed on... Figure 4 In this case, you can simply replace the training set with the validation set.

[0049] The method for finding a path to the service recipient based on the robot's current position is to draw a straight line ca from the current position to the kitchen entrance c, and then connect the shortest path starting point ca in step S4. If the mainboard calculates that a collision will occur in the straight line ca, the user can adjust the line via the touch screen or remote smartphone until a collision is calculated.

Claims

1. A home care service robot, characterized by, It comprises a cleaning base, a main body and a head with a touch screen from bottom to top in turn, wherein, The cleaning base comprises a shell, a cleaning structure arranged inside the shell, a walking mechanism, a dust storage bin, a water tank, a sewage suction tank, a plurality of cameras arranged on the side of the cleaning base, and a first intelligent chip for three-dimensional modeling and path calculation according to images captured by the cameras in real time. The main body comprises a plurality of storable dish plates, and a separate compartment with a mainboard arranged in the main body. When the dish plates are stored, heating and / or heat preservation can be performed inside the main body, and a path to the service object is found according to the current position of the robot in the path to walk to the side of the service object; The second intelligent chip is arranged on the mainboard, which can perform cleaning work at regular intervals, remotely control household appliances required by the service object at regular intervals, and remotely communicate with the personnel, remind the artificial service time and the current physiological state of the service object. The personnel can modify the cleaning work and the household appliance remote control operation time through the touch screen. The mainboard re-trains the model according to the modified historical data at regular intervals, and when the timing result based on the newly trained model is inconsistent with the last modification record, the personnel is remotely informed to manually determine whether to select the timing result obtained by the newly trained model or the last modification record to remotely execute the household appliance remote control operation and / or cleaning work. The cleaning base comprises a universal wheel controlled by the mainboard; the cameras are provided with distance sensors to define the initial position, so that the universal wheel controls the cleaning base to rotate first, collects the images captured by the cameras before and after the first rotation, and the distance data collected during the first rotation, and continues to complete the second rotation, completes a circle of rotation, and also continues to collect the distance data collected during the second rotation. The first intelligent chip performs three-dimensional modeling according to the images and all distance data. The plurality of cameras are equally distributed around the circumference. The angle of the first rotation is half of the acute central angle between the positions of the adjacent two cameras and the line connecting the center. The mainboard has a voice dialogue module, which can recognize the voice of the served object and the voice of the personnel according to the collected voice, so as to prioritize the pre-trained model and the remote execution when the served object and / or the personnel voice dialogue, to perform the cleaning work and / or remote control operation. The personnel voice dialogue includes on-site dialogue and remote dialogue.

2. The robot of claim 1, wherein, The step of performing three-dimensional modeling according to the images and all distance data comprises: S1: establishing a three-dimensional rectangular coordinate system in the modeling software, positioning all distance data based on the base center position, and sequentially connecting each positioning point; S2: in the three-dimensional rectangular coordinate system, based on the images before and after the first rotation, identifying whether it belongs to furniture, positioning the furniture according to the photographed center position, and establishing a three-dimensional body of the furniture according to the preset height estimate value of the furniture. S3 determines the three-dimensional projection contour of the furniture on the ground, and establishes an initial path in the non-furniture projection contour area, and walks along the initial path. Every preset time interval, stop and continue steps S1-S2, and continue to establish more paths until the scanning is completed, and all non-furniture projection contour areas are obtained; S4 finds the nearest path to the vicinity of the current location of the service object among all established paths, and defines the path as the fastest path to reach based on the current location.

3. The robot of claim 2, wherein, Wherein the establishment of the initial path and other paths is determined by the user in the non-furniture projection contour area displayed on the touch screen or smart phone, and after determination, it is handed over to the mainboard to control the universal walking wheel to make the robot walk; After the path is established, the mainboard finds that based on the current furniture and non-furniture projection contour area, walking according to the established path will cause a collision, so the user will determine a new path, until the calculation result is that there will be no collision; The user can set the robot to set the starting point, and after completing all tasks, it returns to the starting point to standby.

4. The robot of claim 3, wherein, The method for pre-training the model comprises the following steps: Q1 obtains the time stamp of the touch screen field user setting, voice history record, remote modification, and divides the same kind of control instruction every day into training set and verification set; Q2 establishes a long short memory model, inputs the control instruction training set into different model network units according to time sequence, calculates the loss function, continuously optimizes the network parameters, verifies the verification set, stops training when the loss function value is minimum, and obtains the pre-training model.

5. The robot of claim 4, wherein, The method for finding a path to reach the service object according to the current position of the robot and walking to the side of the service object is that a straight line is drawn from the current position to the starting point of the nearest path in step S4, and when the mainboard calculates that a collision will occur in the straight line, the user adjusts it through the touch screen or remote smart phone until no collision is calculated.

6. The robot of claim 4, wherein, The head is also provided with a camera for identifying the front face image of the cared object, so as to change it into a dynamic picture by AI, wherein a rotating disc is arranged between the main body and the T-shaped shell to enable the camera to conveniently find and align the collected face; the heating and heat preservation are set by the touch screen.

Images