Patient falling protection robot system based on flexible buffering

Abstract

The invention provides a patient falling protection robot system based on flexible buffering. According to the method, an environment sensing module, a central processing module, a high-speed moving chassis and a flexible buffering module are included; the environment sensing module is used for sensing user posture information and surrounding environment information; the central processing module is used for processing the posture information of the user and the surrounding environment information and predicting a falling track and a falling position of the user; the high-speed mobile chassis is connected with the flexible buffer module, and the flexible buffer module is moved to a corresponding position based on the predicted tumble track and tumble position of the user; the flexible buffer module is used for performing fall protection on the user; a flexible buffering mechanism is adopted, secondary injuries such as hurting injuries and collision injuries possibly caused by rigid supporting are avoided, and the safety requirement of a human body is better met.

Description

Technical Field

[0001] This invention belongs to the field of service robot technology, specifically relating to a patient fall prevention robot system based on flexible cushioning. Background Technology

[0002] With the increasing aging of society, patient falls have become a serious public health problem. Falls can lead to serious consequences such as fractures and traumatic brain injury, and even endanger life. Currently, fall prevention products on the market are mainly divided into two categories: one is wearable devices, such as fall detection wristbands / belts, which can sound an alarm after a fall is detected, but cannot actively intervene to prevent injury; the other is environmental modifications, such as installing handrails and laying soft flooring, but these measures are static and passive and cannot provide protection in all areas where patients move around.

[0003] In the field of robotics, some research has attempted to use robots for assisted walking or fall detection, but most of these studies remain at the "detection-alarm" stage. A few studies have proposed the idea of ​​providing physical support during falls, but these typically suffer from problems such as slow response speed, high rigidity of the support structure potentially causing secondary injuries, or overly simplistic action strategies that cannot cope with complex real-world environments.

[0004] Therefore, existing technologies lack an innovative solution that can proactively, quickly, and accurately intervene physically during a fall and effectively absorb impact energy in a flexible manner, thereby directly mitigating injury. Summary of the Invention

[0005] To address the problems existing in the current technology, this invention proposes a patient fall protection robot system based on flexible cushioning. The system includes: an environmental perception module, a central processing module, a high-speed moving chassis, and a flexible cushioning module. The environmental perception module is used to perceive user posture information and surrounding environment information. The central processing module processes the user posture information and surrounding environment information to predict the user's fall trajectory and fall location. The high-speed moving chassis is connected to the flexible cushioning module and moves the flexible cushioning module to the corresponding position based on the predicted user fall trajectory and fall location. The flexible cushioning module is used to protect the user from falls.

[0006] The beneficial effects of this invention are:

[0007] The system of this invention can transform from "post-incident alarm" to "in-incident intervention," taking action before injury occurs. This invention combines high-speed prediction and high-speed movement, completing deployment before the patient's body hits the ground. This invention employs a flexible cushioning mechanism, avoiding secondary injuries such as chafing and impact injuries that may be caused by rigid support, thus better meeting human safety requirements. Attached Figure Description

[0008] Figure 1 This is an overall structural block diagram of the robot system described in this invention;

[0009] Figure 2 This is a schematic diagram of the workflow of the method described in this invention;

[0010] Figure 3 This is a system state diagram of the method described in this invention;

[0011] Figure 4 This is a system interaction diagram of the method described in this invention;

[0012] Figure 5 This is a data flow diagram of the method described in this invention;

[0013] Figure 6 This is a typical scenario diagram of a robot predicting a fall and moving to a predicted position to cushion the impact.

[0014] Figure 7 This is a cross-sectional view of one specific implementation of the flexible buffer module. Detailed Implementation

[0015] 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.

[0016] A patient fall prevention robot system based on flexible cushioning is disclosed. The system includes: an environmental perception module, a central processing module, a high-speed moving chassis, and a flexible cushioning module. The environmental perception module senses user posture information and surrounding environment information. The central processing module processes the user posture information and surrounding environment information to predict the user's fall trajectory and fall location. The high-speed moving chassis is connected to the flexible cushioning module and moves the flexible cushioning module to a corresponding position based on the predicted user fall trajectory and fall location. The flexible cushioning module is used to protect the user from falls.

[0017] In this embodiment, the central processing module processes the input data by: preprocessing the user's posture information and surrounding environment information; fusing the preprocessed data and inputting the fused data into the fall prediction model to obtain the user's fall prediction result; determining the user's fall location information based on the surrounding environment information and planning the movement path of the protective robot based on the location information; generating movement commands through the decision and control unit based on the movement path and transmitting the movement commands to the high-speed mobile chassis.

[0018] Preprocessing of user posture information and surrounding environment information includes:

[0019] Step 1: Preprocess user pose information; normalize keypoint coordinates by transforming the 3D coordinates of human keypoints from the image coordinate system to the world coordinate system to eliminate the influence of distance and angle. The formula is as follows:

[0020]

[0021] in, These are the coordinates of a point in the pixel coordinate system; This is the intrinsic parameter matrix of the robot system's camera. This is the inverse of the camera intrinsic parameter matrix; This is the transformation matrix from the camera coordinate system to the world coordinate system; These are the coordinates of a three-dimensional point in the world coordinate system.

[0022] Data smoothing filtering uses a Kalman filter to smooth the coordinates of key points and eliminate measurement noise. The formula is as follows:

[0023]

[0024] in, These are the original measurement values ​​(key point coordinates) at the current time t. This is the filtered estimate from the previous time step (t-1); This is the filtered estimate (smoothed output) for the current time. The smoothing factor (filter coefficient) has a value range of [0, 1] and controls the weight of the new and old data.

[0025] Attitude sequence standardization involves time alignment and length standardization of the attitude sequences of consecutive frames to ensure the temporal consistency of the input sequence.

[0026] Step 2: Environmental Information Preprocessing. Point cloud downsampling and filtering: Voxel mesh filtering is used to downsample the original point cloud, preserving key structural features while reducing data volume. Ground plane extraction: The ground plane is extracted from the point cloud to provide a baseline for supporting polygon calculations. The formula is as follows:

[0027]

[0028] Where a, b, c and d are plane parameters, and (a, b, c) constitute the normal vector of the plane; x, y and z are the coordinates of a point in three-dimensional space, and d is the signed distance parameter from the plane to the origin.

[0029] Obstacle clustering and identification: The Euclidean clustering algorithm is used to segment the non-ground point cloud into independent obstacles and calculate their bounding boxes and centroid positions.

[0030] Step 3: Multi-sensor data fusion preprocessing. Timestamp synchronization: Time-align the data streams from different sensors to ensure temporal consistency between attitude and environmental data. Coordinate system one: Transform the vision, IMU, and other data into the robot's base coordinate system. The formula is...

[0031]

[0032] in, The transformation matrix from the sensor to the world coordinate system is obtained through sensor calibration; This is the transformation matrix from the world to the robot's base coordinate system, obtained through robot localization; This is the final transformation matrix from the sensor to the robot's base coordinate system.

[0033] Data quality assessment involves real-time monitoring of data quality from various sensors, and imputation or labeling of outliers or missing data.

[0034] Step 4: Feature Engineering Preprocessing. Dynamic feature extraction: Extracting dynamic features such as centroid trajectory, joint angle change rate, and support base area change from the posture sequence. Environmental feature encoding: Encoding environmental information such as obstacle distribution, ground flatness, and spatial accessibility into feature vectors. Temporal feature construction: Constructing statistical features (mean, variance, extreme values) within a sliding window as model input.

[0035] Through the above preprocessing steps, the raw multimodal sensor data is transformed into standardized, high-quality feature representations, providing a reliable input basis for subsequent fall risk prediction, path planning, and buffering decisions, effectively improving the system's real-time performance and robustness.

[0036] The fusion of preprocessed data includes:

[0037] Step 1. Spatiotemporal registration of multi-source heterogeneous data. Time series alignment: Dynamic time warping algorithm is used to synchronize the attitude sequence and environmental perception data. Spatial coordinate unification: A unified global coordinate system is established, mapping the data from each sensor to the same spatial reference frame.

[0038] Step 2. Information fusion based on a Bayesian framework. Multimodal state estimation: IMU and visual data are fused using an extended Kalman filter to estimate the user's complete motion state. Uncertainty quantification: The measurement uncertainties of each sensor are modeled, and optimal weighted fusion is achieved based on the covariance matrix.

[0039] Step 3. Feature-level deep fusion. Multi-scale feature extraction: Spatiotemporal features are extracted from pose sequences, environmental point clouds, and IMU signals, respectively. Attention mechanism fusion: A cross-modal attention network is used to adaptively adjust the contribution weights of each modality feature. Graph neural network fusion: A user-environment interaction graph is constructed, and structured information fusion is achieved through graph convolutional networks.

[0040] Step 4. Decision-level Fusion Strategy. Multi-expert system integration, combining dynamics-based prediction models, deep learning-based recognition models, and rule-based logic models. Confidence-weighted voting, dynamically adjusting the weight of each model in the final decision based on its historical accuracy. DS evidence theory fusion, handling uncertainties and conflicts in sensor data, improving system robustness.

[0041] Step 5. Real-time Adaptive Fusion Mechanism. Online learning and adjustment: Fusion parameters are dynamically updated based on feedback signals to adapt to different user and environmental scenarios. Anomaly detection and fault tolerance: The reliability of each data source is monitored in real time, and abnormal sensor data is downweighted or excluded. Resource-aware fusion: The complexity of the fusion algorithm is dynamically adjusted according to computing resource conditions to ensure real-time requirements.

[0042] Through the aforementioned multi-level and multi-dimensional data fusion technology, the system can extract more accurate and robust user status and environmental perception from redundant and complementary multi-source information, providing a solid data foundation for accurate prediction of fall risk and protective decisions.

[0043] The processing of fused data using a fall prediction model includes:

[0044] Step 1. Multimodal temporal feature analysis. Attitude dynamics feature extraction: Based on the fused attitude data, calculate features such as the center of mass trajectory, angular momentum, and joint kinematic chains. The formula is as follows: ,in Joint torque represents the rotational force acting on a human joint. The moment of inertia matrix describes the inertia of an object rotating about an axis; ω is angular acceleration, which represents the rate of change of angular velocity. This is the angular velocity vector, representing the speed and direction of rotation; This refers to the vector cross product operation, which calculates the cross product of two vectors. Environmental interaction feature modeling analyzes environmental interaction features such as the relative position and distance change rate between the user and obstacles. The formula is...

[0045]

[0046] in, Minimum danger distance: the shortest distance between the user and the nearest obstacle; This is the user's location coordinate vector, typically representing the user's centroid or the location of key body parts; The obstacle's position coordinate vector represents the location of the obstacle in the environment. Given the Euclidean norm, calculate the straight-line distance between two points; This is a minimum function that finds the closest obstacle among all obstacles.

[0047] Step 2. Spatiotemporal graph convolutional prediction based on attention mechanism. Spatiotemporal graph construction: The human skeleton is constructed as a graph structure, with nodes representing key points and edges representing skeletal connections and temporal relationships. Hierarchical attention mechanism: Spatial attention and temporal attention are used to capture key key points and key time frames respectively, as shown in the formula:

[0048]

[0049] in, Let l be the node feature matrix of the l-th layer; The output is the node feature matrix of the (l+1)th layer. Let be the k-th adjacency matrix, representing the spatial connectivity of the graph; Let K be the k-th trainable weight matrix of the l-th layer; K is the number of adjacency matrices, which usually correspond to different relation types. Nonlinearity is introduced for activation functions, such as ReLU; For the summation operation, the convolution results of different adjacency matrices are aggregated.

[0050] Step 3. Multi-task joint learning framework. Fall probability prediction, outputting fall risk probability values ​​over continuous time series. Fall direction classification, and simultaneous prediction of fall direction category. {Front, Back, Left, Right, Unknown}. Collision time estimation, predicting the estimated time of a collision. ,in To predict the collision time, the estimated moment when the body makes contact with the ground / obstacle is determined. The reference time for prediction is the current time. The prediction time interval is the time difference from the present to the expected collision.

[0051] Step 4. Trajectory prediction based on physical constraints. Human kinematic constraints are embedded in the prediction, including physical constraints such as the range of motion of human joints and kinematic chains. ,in (and Let and represent the minimum (and maximum) joint angles, respectively, the minimum (and maximum) angles that the joint can achieve. Environmental physical constraints, considering environmental physical factors such as ground reaction force and friction, are given by the formula:

[0052]

[0053] in, For the ground's supporting force on the human body, For human body mass; This is the acceleration due to gravity, approximately 9.8 m / s². 2 ; The acceleration in the vertical direction is the acceleration of the human body in the vertical direction.

[0054] Step 5. Uncertainty Quantification and Confidence Assessment. Probability output calibration: The probability values ​​output by the model are calibrated using histogram binning. The calibrated probabilities... ,in Here, z is the sigmoid function, z is the original output of the model, and T is the temperature parameter used to adjust the smoothness of the probability distribution. The prediction uncertainty is estimated by quantifying the prediction uncertainty through ensemble learning, as shown in the formula:

[0055]

[0056] Where is the prediction variance, representing the degree of uncertainty in the model's prediction; M is the number of model samples; f m (x) is the model's predicted output for input x at the m-th sampling; This is the average predicted value.

[0057] Step 6. Real-time adaptive prediction mechanism. Online model updates dynamically adjust prediction model parameters based on the latest observation data. ,in, Let be the model parameters at time t. These are the updated model parameters; The learning rate controls the size of the parameter update step. The gradient of the loss function represents the direction of parameter updates; Let be the input data and the ground truth label at time t. Personalized adaptation adjusts the prediction threshold and sensitivity based on the user's historical behavior patterns, using the formula:

[0058]

[0059] in, A personalized threshold, a fall detection threshold customized for this user; This is a global threshold, based on a baseline threshold set for the general population. This is a historical threshold, adjusted based on the user's historical behavior data; To integrate weights, the contributions of global thresholds and historical thresholds are balanced.

[0060] Step 7. Multi-scale prediction fusion. Short-term accurate prediction: predicting the precise state 0.1-0.5 seconds into the future based on high-frequency sensor data. Long-term trend prediction: predicting the fall risk trend 1-3 seconds into the future based on historical behavior patterns. Adaptive weighted fusion: dynamically adjusting the fusion weights of short- and long-term predictions according to the prediction time scale and confidence level, using the following formula:

[0061]

[0062] in, The final fusion probability is the combined result of multi-scale predictions; For short-term probability prediction, accurate prediction is based on high-frequency data; For long-term probability prediction, trend prediction based on historical patterns; The weighting of short-term predictions and their importance in the fusion process; Long-term prediction weights are used to illustrate the importance of long-term predictions in fusion.

[0063] Through the multi-level, multi-scale fall prediction model processing described above, the system can accurately identify fall risks from fused data, predict fall trajectories and collision times, and provide accurate and reliable input for subsequent robot movement planning and buffering decisions, ensuring that protective measures are taken at the optimal time.

[0064] In this embodiment, the flexible buffer module includes:

[0065] Module 1: Intelligent Compression Buffer Structure. Utilizing a multi-layered composite compression core material, it achieves a composite structure of shape memory polymer and highly elastic foam, resulting in a high compression ratio. Recovery time ,in, The initial thickness is the original thickness of the material when it is not under pressure. Thickness after compression, the actual thickness of the material after being compressed; Recovery time is the time required for the body to recover from a compressed state to its initial state. Simultaneously, a variable stiffness support frame is employed, with graded stiffness adjustment based on a trapezoidal wave spring array, and the stiffness coefficient... Where K is the instantaneous stiffness coefficient; This is the initial stiffness coefficient; This represents the amplitude of stiffness variation. This is the stiffness attenuation coefficient. This module also employs a honeycomb energy absorption layer, constructed using a negative Poisson's ratio material, to meet... ,in Poisson's ratio is the ratio of transverse strain to axial strain. For lateral strain; For axial strain.

[0066] Module 2. Mechanical rapid deployment mechanism. The deployment system uses a torsion spring drive, with a high energy density. , unfolding time The shape memory alloy is used as a trigger, and the NiTi alloy wires in it have phase transition temperatures of... strain recovery rate It also uses a linkage amplification mechanism with a transmission ratio of... This enables rapid and large-scale deployment.

[0067] Module 3. Contact Interface Optimization Design. A biomimetic surface bonding layer is used, based on B-spline surface modeling:

[0068]

[0069] in, The surface parametric equations; These are the coordinates of the control points; represents the B-spline basis functions; m and n are the control point grid dimensions; p and q are the B-spline degrees; The coefficient of friction; This refers to the operating temperature.

[0070] Meanwhile, it employs a temperature-adaptive coating and thermochromic materials, with a working temperature range of [missing information]. .

[0071] Module 4. Gas-free Energy Management System. Employs a flywheel energy storage system with a rotational inertia... angular velocity It employs piezoelectric energy recovery materials, achieving high energy conversion efficiency. Output power A supercapacitor buffer is used, and its capacitance value is... Charge and discharge efficiency .

[0072] Module 5. Intelligent Sensing and Control Unit. Employs a fiber Bragg grating sensor network with wavelength resolution... Strain measurement accuracy Employs an embedded AI processor, with high computing power. Power consumption Response time .

[0073] Module 6. Structural Reliability Design. A carbon fiber composite shell is used, with a density of... compressive strength It uses modular quick-change design with magnetic connection interfaces, reducing disassembly time. Environmentally adaptable design, operating humidity range .

[0074] Module 7. Performance parameters, including energy absorption efficiency Peak force attenuation (correspond (Impact); Cyclic service life Second; System quality Noise level .

[0075] Through the above design, the flexible buffer module realizes a buffering mechanism that combines pure mechanics with smart materials. When a fall risk is detected, it can be quickly deployed through mechanical energy storage and achieve efficient energy absorption by utilizing the intrinsic properties of smart materials, thus maintaining excellent buffering performance and reliability.

[0076] In this embodiment, the environmental perception module includes an RGB-D camera, an inertial strategy unit (IMU), a lidar, and a millimeter-wave radar.

[0077] Figure 1 This is a block diagram of the overall structure of the robot system described in this invention; the flowchart clearly illustrates the four core modules of the patient fall prevention robot system and their collaborative working mechanism. The environmental perception module collects patient posture and environmental data in real time through multi-sensor fusion. This data is fed into the central processing module, where AI algorithms predict fall risk and plan a path. The decision-making unit then issues commands in parallel: the high-speed mobile chassis rapidly moves to the estimated landing point, while the flexible buffer module enters a standby state. Finally, just before the patient hits the ground, the flexible body precisely deploys to complete the buffering intervention, forming a complete protective closed loop from perception and decision-making to execution.

[0078] Figure 2 This is a schematic diagram illustrating the workflow of the method described in this invention; the flowchart clearly shows the five-stage workflow of the patient fall prevention robot. The system continuously monitors the patient's status through an environmental perception module (Stage 1). Once the fall prediction algorithm identifies a high risk, it immediately generates movement and cushioning commands (Stage 2). Subsequently, the high-speed moving chassis and the flexible cushioning module are activated in parallel, performing maneuvering and pre-deployment tasks respectively (Stage 3). At the final moment, the system fully activates the cushioning or cancels the action based on contact judgment (Stage 4). Finally, post-incident processing is completed through audible and visual alarms and remote notifications (Stage 5), forming a complete closed-loop protection system.

[0079] The fall prediction model, the centroid-support polygon model, includes a human centroid calculation model and a fall risk index model, and its expression is:

[0080] The human body's center of mass calculation model is as follows:

[0081]

[0082] The fall risk indicator model is as follows:

[0083]

[0084] in, Let m be the three-dimensional coordinates of the center of mass of the human body at time t; i Let be the mass of the i-th body segment (e.g., thigh, torso); Let t represent the three-dimensional position coordinates of the i-th body segment at time t; M represents the total mass of the human body. The velocity vector is the projection of the centroid, calculated from the coordinate difference of consecutive frames; It is the shortest distance from the centroid projected onto the boundary of the supporting polygon (the convex hull formed by the two legs); The magnitude of the center-of-mass velocity indicates the severity of the instability. and These are weighting coefficients used to balance the contributions of distance and speed factors to risk. and These are the normalization parameter and the threshold, respectively, used to unify physical quantities of different dimensions into risk scores.

[0085] A prediction model based on temporal deep learning (AI method) for predicting fall probability and direction:

[0086]

[0087] in, This represents the fall probability output by the model. Classify the fall direction output by the model (e.g., front, back, left, right). X represents the model parameters, which are obtained through training with a large amount of fall and daily behavior data; X is the input sequence, which contains historical posture data from the past T frames.

[0088] Figure 3 This is a system state diagram of the method described in this invention; the state diagram fully presents the intelligent decision-making logic of the protective robot. The system starts in a standby patrol state, continuously monitoring the patient's posture through data analysis. Once an anomaly is detected, fall prediction is immediately initiated. After confirming a high risk, it enters an emergency response sub-state including path planning, high-speed maneuvering, and buffer preparation. Upon reaching the buffer-ready state, the system makes a final judgment based on the actual contact situation: if a collision is detected, the buffer device is fully activated and post-event processing begins; if it is determined to be a false alarm, it directly returns to the standby state, demonstrating the system's adaptability to complex scenarios.

[0089] Figure 4This is a system interaction diagram of the method described in this invention; this sequence diagram dynamically presents the complete collaborative process of robot fall protection. When a patient experiences instability, the sensing module immediately sends a data stream to the central processing unit. The processor uses AI algorithms to cyclically analyze the data, and after confirming a high risk, issues movement and cushioning preparation commands in parallel. Simultaneously, the mobile chassis precisely maneuvers to the contact point, and the cushioning module completes pre-inflation. At the critical moment of body contact, the proximity sensor triggers the final command, and the flexible cushioning interface fully deploys to absorb the impact. Alarms are sent to the cloud platform synchronously throughout the process, achieving seamless collaboration from detection and response to alarm.

[0090] The landing site prediction model includes the following:

[0091] System state vector:

[0092]

[0093] Kalman filter prediction step:

[0094]

[0095]

[0096] Landing point prediction using a uniform acceleration model:

[0097]

[0098] Robot's final goal:

[0099]

[0100] The state vector at time t includes position p, velocity v, and acceleration a. The state prior estimate at the predicted time t is also included. For prediction at time t The state prior estimate at time step 1. F is the state transition matrix, describing how the system evolves from the current state to the next state. P is the error covariance matrix, representing the uncertainty of the state estimate. Q is the process noise covariance matrix, representing the uncertainty of the model. The coordinates of the predicted body landing point. For the prediction time window, such as 0.5 seconds in the future. These are the coordinates of the target point that the robot needs to move to. To adjust the offset vector, calculations are performed based on the fall direction and local environmental obstacles to ensure effective robot intervention.

[0101] Figure 5This is a data flow diagram of the method described in this invention; the flowchart clearly illustrates the data flow and closed-loop control process of the protective robot. The system takes multimodal sensor data as input, which is fused and processed by the perception module to form structured environmental information. The central processing module, acting as the intelligent hub, synchronously outputs path coordinates and buffer commands, respectively driving the mobile chassis to achieve precise maneuvering and controlling the flexible body to complete buffer preparation. The entire system achieves closed-loop control through position feedback and, simultaneously with intervention, sends alarms to the cloud via the communication module, forming a complete protective closed loop integrating perception, decision-making, execution, and alarm functions.

[0102] Impact energy of falling:

[0103]

[0104] Energy absorbed by the buffer:

[0105]

[0106] Airbag insulation process model:

[0107]

[0108] Approximate calculation of buffer force:

[0109]

[0110] in, This refers to the kinetic energy generated when a patient falls. For the quality of patients. The impact velocity when the patient comes into contact with the buffer. F(d) represents the energy absorbed by the flexible buffer through deformation. F(d) is the function of the buffer force and the deformation depth d, which depends on the material properties. P and V are the instantaneous pressure and volume inside the buffer module. and The initial pressure and initial volume of the airbag. is the adiabatic index, approximately 1.4 for air. F is the average cushioning force provided by the buffer. A is the effective contact area between the buffer and the patient's body.

[0111] Figure 6 This is a schematic diagram of a typical scenario where a robot anticipates a fall and moves to a predicted position to cushion the impact. This sequence of diagrams clearly illustrates the four typical stages of a robot's fall protection:

[0112] Phase 1: Monitoring and Prediction. Process: The patient exhibits instability while walking (e.g., slipping or tripping). Technical Action: The robot's environmental perception module (e.g., the top camera) continuously captures key data about the human body. The core algorithm of the central processing module analyzes this data in real time, predicting an impending fall by calculating the centroid shift velocity and the relationship between the supporting polygons, and calculating the body's estimated trajectory and landing point.

[0113] Phase Two: Decision Making and Maneuvering. Process: The robot rapidly starts from its standby position. Technical Actions: The system generates two parallel commands: Movement Command: The calculated estimated landing point coordinates are sent to the high-speed moving chassis. The chassis immediately plans an optimal path to avoid obstacles (such as chairs) and maneuvers at high speed. Buffer Preparation Command: The flexible buffer module enters a ready-to-trigger state.

[0114] Phase Three: Buffer Intervention. Process: The patient tilts and comes into contact with the positioned robot. Technical Action: The robot precisely arrives at the intervention point just before the patient's body hits the ground. At this moment, the flexible buffer module has fully deployed, using its soft surface to support the patient's body and absorb impact energy through material deformation, forming an effective first line of defense.

[0115] Phase Four: Post-Incident Handling. Process: After the fall is successfully cushioned. Technical Actions: The robot initiates the post-incident handling process, including issuing a local audible and visual alarm to attract the attention of those nearby, and simultaneously sending an alert message containing the location and situation of the incident to pre-set emergency contacts (such as children, community centers) via wireless network.

[0116] Figure 7 This is a cross-sectional view of a specific implementation of the flexible buffer module. The view clearly illustrates the internal hierarchical structure and working principle of the flexible buffer module:

[0117] Outer Layer: Contact and Sensing Layer. High-Elasticity / Antibacterial Skin: The outermost layer, in direct contact with the human body. Made of soft, skin-friendly, and easy-to-clean materials to ensure comfort and hygiene during cushioning. Contact Pressure Sensor Array: Distributed beneath the skin, used to sense the occurrence of contact and the distribution of contact force, providing data for subsequent fine-tuning of the cushioning force. Proximity Sensor: Used to detect whether the human body is approaching at extremely high speed in the final stage, providing the final signal to trigger cushioning.

[0118] Buffering Function Layer: Multi-level Buffering Module (Core) This is a hierarchical, collaborative system designed to achieve optimal buffering performance.

[0119] Level 1: Rapid Response Main Buffer Module. Characteristics: Low pressure, large area. Utilizes rapidly deployable compressible material, aiming to deploy within an extremely short time (milliseconds) to make initial contact with the human body, avoiding rigid collisions and providing initial deceleration. Level 2: Energy Absorption Core Module. Characteristics: Higher internal pressure, or filled with non-Newtonian fluid. When the first-level airbags make contact, the impact force is transferred to this layer. Under immense pressure, this layer absorbs and disperses most of the impact energy through gas compression or fluid hardening, which is crucial for protection. Level 3: Bottom Support Bag. Characteristics: Controllable stiffness. It provides stable base support for the entire buffer structure and participates in the final energy dissipation through its own controllable deformation, preventing the robot from shifting backward or tipping over due to impact.

[0120] Bottom Layer: Control and Structural Layer. High-Strength Flexible Substrate: Connects to the robot's main body, supports all internal structures, and requires a certain level of strength and flexibility. Miniature Air Pump and Air Tank: Provides the air source for inflating the airbags. In active systems, a compressed gas generator may also be included for extremely rapid inflation. High-Speed ​​Solenoid Valve: A key component controlling the direction and flow of gas, with extremely fast response time, directly driven by the control unit.

[0121] In this embodiment, the control unit receives signals from the robot's central processor and its own sensors, precisely controls the opening and closing of the air pump and solenoid valve, and realizes the preparation, graded inflation and de-inflation of the airbags. The workflow is as follows: (1) The central processor issues a "buffer preparation" command, the control unit starts the air pump, and the first-stage main airbag begins to pre-inflate. (2) The proximity sensor confirms that the human body is about to collide and issues a final trigger signal. (3) The control unit opens the high-speed solenoid valve, and all airbags are fully inflated in an instant, forming a multi-layer buffer structure. (4) The human body impacts the soft skin, and the impact force is absorbed and dispersed step by step. (5) Afterwards, the control unit opens the de-inflation valve to release the gas, the buffer module is reset, and it is ready for the next task.

[0122] refer to Figure 1 The system consists of four main hardware components working together. The environmental perception module uses an RGB-D camera and a built-in IMU. The central processing module is an embedded industrial computer. The mobile chassis is a four-wheel differential drive chassis. The flexible cushioning module is a layer of slow-rebound memory foam about 10cm thick that wraps around the robot's shell, and integrates a main airbag and two side airbags on the front.

[0123] Workflow (Reference) Figure 2 ):

[0124] Step 1: The RGB-D camera tracks 17 key points on the patient's body at a frequency of 30Hz.

[0125] Step 2: The central processing unit calculates the relationship between the human body's center of mass projection and the supporting surface in real time. When a rapid and significant deviation of the center of mass from the supporting surface is detected (e.g., severe body tilt and angular velocity exceeding a threshold), the algorithm is immediately triggered.

[0126] Step 3: The algorithm predicts the trajectory of the body within the next 0.5 seconds based on the current body posture and angular velocity, and calculates the most likely body landing area (e.g., the left hip and left arm area).

[0127] Step 4: The system uses the center coordinates of the landing area as the robot's target point and plans the shortest path to avoid the tables and chairs.

[0128] Step 5: The chassis starts with maximum acceleration and moves 3 meters in 0.8 seconds to reach the target point.

[0129] Step 6: During movement, the main airbag on the front begins to pre-inflate. Upon reaching the target point, a proximity ToF sensor confirms the patient's approach, and the main and side airbags fully inflate within 100 milliseconds, forming a large, soft contact surface.

[0130] Step 7: The patient's body falls onto the inflatable airbag, effectively absorbing the impact. Simultaneously, the robot sends an alert message via its 4G module to the patient's children's mobile phones stating, "A fall has occurred, and cushioning has been applied."

[0131] In another embodiment, the flexible cushioning module employs non-Newtonian fluid technology. Under normal conditions, the pouch filled with a shear-thickening fluid is soft. Upon high-speed impact, the fluid instantly hardens, thus efficiently dispersing the impact force. The robot can selectively activate different regions of the non-Newtonian fluid pouch based on the predicted fall intensity, achieving an adaptive cushioning effect.

[0132] The above-described embodiments further illustrate the purpose, technical solution, and advantages of the present invention. It should be understood that the above-described embodiments are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made to the present invention within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A patient fall prevention robot system based on flexible cushioning, characterized in that, include: The system comprises an environmental perception module, a central processing module, a high-speed mobile chassis, and a flexible buffer module. The environmental perception module senses user posture information and surrounding environment information. The central processing module processes the user posture information and surrounding environment information to predict the user's fall trajectory and fall location. The high-speed mobile chassis is connected to the flexible buffer module and moves the flexible buffer module to the corresponding position based on the predicted user fall trajectory and fall location. The flexible buffer module is used for fall protection for the user.

2. The patient fall prevention robot system based on flexible cushioning according to claim 1, characterized in that, The environmental perception module includes an RGB-D camera, an inertial strategy unit (IMU), a lidar, and a millimeter-wave radar.

3. The patient fall prevention robot system based on flexible cushioning according to claim 1, characterized in that, The central processing module processes the input data including: preprocessing user posture information and surrounding environment information; fusing the preprocessed data and inputting the fused data into the fall prediction model to obtain the user fall prediction result; determining the user's fall location information based on the surrounding environment information and planning the movement path of the protective robot based on the location information; generating movement commands through the decision and control unit based on the movement path and transmitting the movement commands to the high-speed mobile chassis.

4. The patient fall prevention robot system based on flexible cushioning according to claim 3, characterized in that, Preprocessing of user posture information and surrounding environment information includes: Step 1: Preprocess user pose information; specifically including: normalizing key point coordinates, transforming the 3D coordinates of human key points from the image coordinate system to the world coordinate system; smoothing key point coordinates using a Kalman filter; and performing time alignment and length normalization on the pose sequence of consecutive frames. Step 2: Preprocess the environmental information; specifically including: downsampling and filtering the point cloud, i.e., using voxel mesh filtering to downsample the original point cloud; separating the ground plane from the point cloud; using the Euclidean clustering algorithm to segment the non-ground point cloud into independent obstacles, and calculating their bounding boxes and centroid positions; Step 3: Time-align the data streams from different sensors, monitor the data quality of each sensor in real time, and interpolate or mark outliers or missing data. Step 4: Dynamic feature extraction, extracting dynamic features such as centroid trajectory, joint angle change rate, and support base area change from the posture sequence; environmental feature encoding, encoding obstacle distribution, ground flatness, and spatial accessibility environmental information into feature vectors; temporal feature construction, constructing statistical features within a sliding window as model input.

5. The patient fall prevention robot system based on flexible cushioning according to claim 3, characterized in that, The fusion of preprocessed data includes: Step 1. Spatiotemporal registration of multi-source heterogeneous data; specifically including: using a dynamic time warping algorithm to synchronize the attitude sequence and environmental perception data in time; establishing a unified global coordinate system and mapping the data from each sensor to the same spatial reference system; Step 2. Information fusion based on the Bayesian framework; specifically including: fusing IMU and visual data using an extended Kalman filter to estimate the user's complete motion state; modeling the measurement uncertainties of each sensor, and achieving optimal weighted fusion based on the covariance matrix; Step 3. Feature-level deep fusion; specifically including: multi-scale feature extraction, extracting spatiotemporal features from pose sequences, environmental point clouds and IMU signals respectively; attention mechanism fusion, using a cross-modal attention network to adaptively adjust the contribution weights of each modality feature; graph neural network fusion, constructing a user-environment interaction graph, and achieving structured information fusion through graph convolutional networks; Step 4. Decision-level fusion strategy; specifically including: multi-expert system integration, integrating dynamics-based prediction models, deep learning-based recognition models, and rule-based logic models; confidence-weighted voting, dynamically adjusting the weight of each model in the final decision based on its historical accuracy; and DS evidence theory fusion, fusing uncertainties and conflicts in sensor data. Step 5. Real-time adaptive fusion mechanism; specifically including: online learning and adjustment, dynamically updating fusion parameters based on feedback signals; anomaly detection and fault tolerance, real-time monitoring of the reliability of each data source, and downweighting or excluding abnormal sensor data; resource-aware fusion, dynamically adjusting the complexity of the fusion algorithm according to the computing resource status.

6. The patient fall prevention robot system based on flexible cushioning according to claim 3, characterized in that, The processing of fused data using a fall prediction model includes: Step 1. Multimodal temporal feature analysis; specifically including: posture dynamics feature extraction, calculating the center of mass trajectory, angular momentum, and joint kinematic chain features based on the fused posture data; environmental interaction feature modeling, analyzing the environmental interaction features of the relative position and distance change rate between the user and obstacles; Step 2. Spatiotemporal graph convolutional prediction based on attention mechanism; specifically including: spatiotemporal graph construction, constructing the human skeleton as a graph structure, nodes as key points, and edges representing bone connections and temporal correlations; hierarchical attention mechanism, using spatial attention and temporal attention to capture key key points and key time frames respectively; Step 3. Multi-task joint learning framework; specifically including: fall probability prediction, outputting fall risk probability values ​​over continuous time series. Fall direction classification, and simultaneous prediction of fall direction category. {Front, Back, Left, Right, Unknown}; Collision time estimation, predicting the expected time of collision; Step 4. Construct trajectory prediction based on physical constraints; that is, human kinematic constraints, embedding physical constraints such as the range of motion of human joints and kinematic chains into the prediction. ,in and These are the minimum and maximum joint angles, respectively; environmental physical constraints include ground reaction force, friction, and other environmental physical factors. Step 5. Uncertainty Quantification and Confidence Assessment; Probability Output Calibration: The probability values ​​output by the model are calibrated using histogram binning. The calibrated probabilities are... ,in Here, z is the Sigmoid function, z is the original output of the model, T is the temperature parameter, and the prediction uncertainty is estimated by ensemble learning to quantify the prediction uncertainty, as shown in the formula: ； in, The prediction variance represents the degree of uncertainty in the model's predictions; M is the number of model samples; f m (x) is the model's predicted output for input x at the m-th sampling; This is the average predicted value; Step 6. Real-time adaptive prediction mechanism; online model update, dynamically adjusting prediction model parameters based on the latest observation data; personalized adaptation, adjusting prediction thresholds and sensitivity according to users' historical behavior patterns; Step 7. Multi-scale prediction fusion; short-term accurate prediction, predicting the precise state in the next 0.1-0.5 seconds based on high-frequency sensor data; long-term trend prediction, predicting the fall risk trend in the next 1-3 seconds based on historical behavior patterns; adaptive weight fusion, dynamically adjusting the fusion weights of short-term and long-term predictions according to the prediction time scale and confidence level.

7. The patient fall prevention robot system based on flexible cushioning according to claim 1, characterized in that, The high-speed mobile chassis includes a drive motor, a wheel structure, and a power management system; the drive motor is connected to the wheel structure and controls the movement of the wheel structure through the drive motor; the power management system is used to provide voltage to the drive motor.

8. The patient fall prevention robot system based on flexible cushioning according to claim 1, characterized in that, The flexible buffer module includes: an intelligent compression buffer structure, a mechanical rapid deployment mechanism, an optimized contact interface, an airless energy management system, and an intelligent sensing and control unit. The intelligent compression buffer structure uses a multi-layer composite compression core material to achieve a composite structure of shape memory polymer and highly elastic foam. The mechanical rapid deployment mechanism uses a torsion spring driven deployment system, which uses a shape memory alloy trigger and linkage amplification mechanism to achieve rapid large-area deployment. The optimized contact interface uses a biomimetic curved surface bonding layer. The airless energy management system uses a flywheel energy storage system. The intelligent sensing and control unit uses a fiber Bragg grating sensor network.

9. A patient fall prevention robot system based on flexible cushioning according to claim 8, characterized in that, The flexible buffer module also includes structural reliability design, using a carbon fiber composite shell with a high density. compressive strength Modular quick-change design with magnetic connection interface, reducing disassembly time. Environmentally adaptable design, operating humidity range .

10. A patient fall prevention robot system based on flexible cushioning according to claim 8, characterized in that, The flexible buffer module also includes performance parameter optimization, namely energy absorption efficiency. Peak force attenuation Cycle life Second; System quality Noise level .

Images