A barrier avoidance early warning system suitable for electric wheelchairs

Abstract

The application relates to the technical field of intelligent auxiliary travel equipment, and particularly provides an obstacle avoidance early warning system suitable for an electric wheelchair. The system comprises a multi-sensor perception module for collecting multi-modal data of a surrounding environment; a data processing and risk assessment module for constructing a local environment dynamic map and performing multi-dimensional risk quantification; an intelligent early warning and decision module for executing hierarchical early warning and flexible active intervention such as vision, hearing and touch according to a risk level; and a personalized learning module for optimizing risk assessment parameters and early warning strategies through reinforcement learning based on user historical behaviors. The system can realize all-around obstacle identification and dynamic avoidance, has high adaptability and learning ability, significantly improves the driving safety and user experience of the electric wheelchair in a complex environment, and has good practicability and popularization prospect.

Description

Technical Field

[0001] This invention relates to the field of intelligent assisted mobility equipment technology, and in particular to an obstacle avoidance and early warning system suitable for electric wheelchairs. Background Technology

[0002] Currently, electric wheelchairs, as assistive mobility devices, are widely used in the daily lives of the elderly, people with physical disabilities, and those with limited mobility. To improve safety during use, some high-end electric wheelchairs have attempted to integrate basic obstacle detection or simple collision warning functions. These typically use ultrasonic or infrared sensors to determine the distance to obstacles and issue simple warnings such as beeps when approaching an obstacle. However, such systems have several technical limitations: First, the perception dimension is limited, lacking the ability to understand environmental semantics. Existing systems typically operate solely based on point-to-point distance sensors, making it difficult to accurately identify the type, size, movement trend, and potential threat level of obstacles. This is especially true in dynamic environments (such as moving pedestrians, animals, and bicycles), where the systems lack the ability to make judgments and cannot achieve true "risk assessment."

[0003] Secondly, the warning methods are simplistic and crude, resulting in a poor user experience. Most existing systems only provide alerts through a single beep or flashing light, lacking differentiation of warning levels and failing to incorporate multimodal output based on user perception channels. For example, people with hearing impairments cannot perceive audible warnings, and those with poor vision have difficulty capturing visual cues, leading to unreliable warnings and even causing interference or misleading information in some scenarios.

[0004] Third, the system lacks an intelligent response mechanism and cannot dynamically adjust according to changes in user status and scenario. Most systems use a fixed threshold triggering mechanism, which cannot optimize parameters according to individual differences or environmental changes. For example, frequent triggering of warnings in a confined environment may lead to "warning fatigue," while in a dangerous environment, the intervention opportunity may be missed because the threshold is not reached for a long time.

[0005] Furthermore, in certain high-risk scenarios, such as when a user continuously operates the control lever towards an obstacle, existing systems often rely solely on the user's autonomous judgment to avoid the obstacle, lacking an active intervention mechanism. If the user reacts slowly or makes a mistake, accidents such as collisions or falls are very likely to occur, posing significant safety hazards.

[0006] In summary, existing obstacle avoidance and early warning systems for electric wheelchairs suffer from problems such as weak perception capabilities, limited early warning methods, rigid response mechanisms, and a lack of personalized adaptation and proactive intervention capabilities. There is an urgent need for a new system with multi-dimensional environmental perception, intelligent risk assessment, hierarchical human-computer interaction, and flexible intervention functions to meet the safety assistance needs in complex environments. Summary of the Invention

[0007] In view of this, the purpose of the present invention is to provide an obstacle avoidance and early warning system suitable for electric wheelchairs, so as to solve the problems of single perception dimension, inaccurate risk judgment, single and crude early warning method, and lack of active intervention and personalized adaptation in the prior art, thereby realizing an active safety assistance system that combines multimodal environmental perception, intelligent risk assessment and humanized interaction.

[0008] To achieve the above objectives, the present invention provides the following technical solution: In one embodiment of the present invention, an obstacle avoidance and early warning system suitable for electric wheelchairs is provided, comprising: a multi-sensor perception module, installed on the electric wheelchair, for real-time acquisition of multimodal data of the surrounding environment; a data processing and risk assessment module, for time synchronization and spatial fusion of the multimodal data, constructing a 360-degree dynamic environmental map centered on the wheelchair, and performing quantitative risk assessment based on at least three dimensions, outputting a risk level signal; an intelligent early warning and decision-making module, for executing a four-level hierarchical early warning strategy based on the risk level signal, including visual, auditory, and tactile prompting units and an active intervention unit; and a personalized learning module, for dynamically optimizing the weights and thresholds of the risk assessment model based on the user's historical operation and early warning result data, and supporting switching between multiple early warning modes.

[0009] Furthermore, the multi-sensor sensing module includes at least two of the following sensors: lidar, ultrasonic sensor, camera, and inertial measurement unit, for the purpose of acquiring and fusing multi-source information.

[0010] Furthermore, the multiple dimensions used in the risk assessment include, but are not limited to: obstacle distance, relative speed, direction of movement, ambient light, road surface conditions, and user behavior characteristics, of which at least three dimensions are selected to participate in the risk quantification calculation.

[0011] Preferably, the visual warning unit includes a graphic display screen and an RGB LED light strip disposed around the wheelchair base, used to synchronously display an environmental semantic map and indicate the direction of risk through color and directional lighting effects.

[0012] Preferably, the auditory warning unit includes a dual-channel speaker and a speech synthesis chip, wherein the speaker supports 3D sound field rendering, and the speech synthesis chip generates graded voice prompts according to the risk level.

[0013] Preferably, the tactile warning unit includes linear resonant actuators embedded in the left and right armrests, which generate tactile feedback of different parts and intensities through pulses to indicate the location and level of risk.

[0014] Furthermore, the active intervention unit is activated when the highest level of risk is detected and the user does not respond in a timely manner. The unit sends control commands to the wheelchair drive controller via the CAN bus to perform progressive deceleration or assisted steering operations.

[0015] Preferably, the progressive deceleration controls the wheelchair's deceleration in a smooth upward manner, with a maximum deceleration not exceeding 2 m / s². 2 Assisted steering achieves steering tendency by adjusting the torque difference between the left and right wheel motors, and this tendency can be actively controlled by the user.

[0016] Furthermore, the personalized learning module employs a Q-learning-based reinforcement learning algorithm with the optimization goal of reducing the false positive rate and improving the success rate of risk avoidance, and automatically adjusts the weight coefficients and judgment thresholds of each risk factor.

[0017] Optionally, the system supports switching between novice mode, standard mode, and expert mode. Each mode has different risk sensitivity and intervention proactivity to adapt to the needs of users with different ability levels.

[0018] Based on the above technical solutions, the obstacle avoidance and early warning system for electric wheelchairs of the present invention achieves accurate identification and response to dynamic obstacle environments through a multi-sensor perception module, a risk assessment model based on multi-dimensional fusion calculation, and an intelligent hierarchical early warning and active intervention mechanism. It effectively solves the problems of single perception dimension, high false alarm and false alarm rate, and slow user response in the prior art, and significantly improves the safety and intelligence level of electric wheelchairs in complex environments.

[0019] This invention can construct a dynamic local environment map centered on the wheelchair in real time, supporting the comprehensive identification and tracking of core elements such as obstacle location, movement trend, and hazard level. By introducing multiple evaluation dimensions such as distance, relative speed, movement direction, ambient lighting, road surface conditions, and user behavior habits, the system can not only adapt to different physical scenarios (such as indoor / outdoor, day / night, flat road / slope), but also match different users' reaction abilities and operating habits to achieve highly personalized risk assessment and intervention strategies.

[0020] In terms of early warning interaction, the multimodal and humanized prompting system proposed in this invention combines three sensory channels: visual (map highlighting and directional light strips), auditory (3D sound source localization and voice broadcasting), and tactile (handrail vibration encoding). This effectively improves the transmission efficiency and accessibility of early warning information, ensuring that users with different cognitive abilities and sensory functions can accurately perceive risks and reduce the probability of being startled, misoperating, or missing risks.

[0021] For extremely high-risk scenarios, this invention features a progressive and flexible intervention mechanism. By precisely controlling the deceleration and slight steering trends of the wheelchair, it provides safety assistance while preserving the user's control, thus avoiding the risk of secondary injuries such as falls and being thrown out due to abrupt braking in traditional systems.

[0022] Furthermore, this invention continuously records user behavior data through a personalized learning module and dynamically optimizes risk weights and warning thresholds using reinforcement learning algorithms, thereby continuously improving the system's adaptability and intelligence over long-term use. The system offers three preset modes: beginner, standard, and expert, which users or caregivers can switch between with a single click according to their needs, making it widely applicable to various groups such as the elderly, people with mobility impairments, and patients undergoing rehabilitation training.

[0023] The system of this invention is built on a mature embedded platform and commercial sensor solutions, and has good integration, maintainability and cost control capabilities. It supports modular upgrades of software and hardware, is easy to promote and apply in existing electric wheelchair products, and has significant market application value and industrialization prospects. Attached Figure Description

[0024] To more clearly illustrate the technical solution of the present invention, embodiments of the present invention will be further described below with reference to the accompanying drawings. The drawings do not constitute a limitation on the scope of protection of the present invention.

[0025] Figure 1 This is a schematic diagram of the obstacle avoidance and early warning system of the present invention applicable to electric wheelchairs; Figure 2 This is a schematic diagram of the functional modules and workflow of the obstacle avoidance and early warning system of the present invention. Detailed Implementation

[0026] To make the technical solution of this invention clearer and more complete, the obstacle avoidance and warning system for electric wheelchairs described in this invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that these embodiments are for illustrative purposes only and do not constitute a limitation on the scope of protection of this invention. Without departing from the principles of this invention, those skilled in the art can make various forms of substitutions, modifications, or equivalent transformations, all of which should be covered within the scope of protection of this invention.

[0027] I. System Overall Structure Description In one embodiment of the present invention, such as Figure 1As shown, an obstacle avoidance and early warning system for electric wheelchairs is integrated into the structural frame of the electric wheelchair body to form a highly integrated safety assistance unit. It mainly includes: a multi-sensor perception module (100), a data processing and risk assessment module (200), an intelligent early warning and decision-making module (300), and a personalized learning module (400). The above modules together construct a closed-loop active safety system.

[0028] The multi-sensor sensing module (100) is fixedly installed at the front, sides, bottom, and top of the backrest of the electric wheelchair. It employs a combination of various sensors, including lidar, ultrasonic sensors, a wide-angle camera, and an inertial measurement unit (IMU), to collect multimodal information about the environment surrounding the electric wheelchair from all directions. Preferably, lidar is used for horizontal two-dimensional mapping, ultrasonic sensors are used for short-range obstacle detection, the camera is used to identify obstacle types and movement trends, and the IMU is used to assist in determining the wheelchair's posture and ramp status.

[0029] The data processing and risk assessment module (200) is connected to each sensor via a high-speed data bus and consists of an embedded processor, including a central computing unit, an edge computing module, and a temporary cache module. This module performs time synchronization, coordinate unification, and sensor fusion calculations to construct a 360-degree local dynamic environment map with the wheelchair as the reference coordinate system. The map updates the position, size, relative speed, and type information of obstacles in real time, and outputs the corresponding risk level signal for each obstacle through a risk calculation model.

[0030] Preferably, the risk assessment model supports the dynamic selection of at least three dimensions from six dimensions, namely "obstacle distance", "relative speed", "direction of movement", "ambient light", "road surface condition" and "user behavior habits", to obtain a comprehensive risk score, which is then mapped to a multi-level risk level for subsequent modules to call.

[0031] The intelligent early warning and decision-making module (300) includes visual, auditory, and tactile early warning units and an active intervention control unit, which are installed near the wheelchair base, armrests, and main control circuit. It is used to trigger corresponding early warning behaviors or active intervention control based on risk level information, forming a hierarchical response mechanism of "prompt-warning-alarm-intervention".

[0032] The personalized learning module (400) continuously records the user's response behavior to the warning (such as the direction of the control lever, braking operation, etc.) through the recording subsystem connected to the main control chip. Combined with the event results (such as whether a collision or emergency avoidance is triggered), the module uses a Q-learning-based reinforcement learning method to dynamically optimize the weight factors and warning thresholds in the risk assessment model, so as to achieve individual adaptability and long-term evolution capability.

[0033] The overall system adopts a modular design and realizes information interaction between multiple modules through CAN bus or other vehicle communication interfaces. It features fast response speed, high control precision, and strong scalability, and can achieve seamless integration without changing the existing electric wheelchair drive structure.

[0034] II. Detailed Explanation of the Intelligent Early Warning and Decision-Making Module In one embodiment of the present invention, such as Figure 1 and Figure 2 As shown, the intelligent early warning and decision-making module (300) is the core of the response execution of the entire system. Its function is to provide multimodal prompts for the risk level signals output by the data processing and risk assessment module (200) and to perform active intervention operations in extreme risk situations to ensure the driving safety of users in complex environments.

[0035] 2.1 Visual Early Warning Unit The visual warning unit includes a 7-inch capacitive touchscreen (301) located in front of the armrest or in the central control area and an RGB LED light strip (302) surrounding the edge of the electric wheelchair base.

[0036] The touchscreen is used to display a semantic map of the wheelchair's surroundings in real time, in which obstacles of different directions and risk levels are highlighted with colors, such as red for high risk, orange for medium risk, and blue for low risk. At the same time, auxiliary information such as the wheelchair's own position and direction of travel is marked on the map, so that users can understand the source of risk and the path of travel at a glance.

[0037] The RGB LED light strip (302) is used to provide immersive orientation warnings. Preferably, the system illuminates the corresponding light strip area according to the location of the obstacle and uses color coding to indicate the risk level. For example, when there is a high-risk obstacle to the left front, the left front light strip will be solid red, enhancing the user's intuitive perception of spatial risks.

[0038] 2.2 Auditory Early Warning Unit The auditory warning unit includes dual-channel speakers (303) symmetrically arranged on both sides of the seat or inside the backrest, and a configured speech synthesis chip (304).

[0039] The dual-channel speakers support 3D sound field rendering. When an obstacle is detected in a certain direction, the system will enhance the volume or directionality of the channel in that direction, so that the sound "comes" from the direction of the risk, helping users to establish spatial awareness.

[0040] The speech synthesis chip (304) generates corresponding prompts or voice broadcasts based on the risk level, preferably including graded prompts such as "attention," "careful," and "danger," while enhancing risk level perception by adjusting the speech rate and tone. For example, the broadcast is gentle and slow when the risk is low, and rapid and rising in tone when the risk is high.

[0041] 2.3 Tactile Early Warning Unit The tactile warning unit includes linear resonant actuators (LRA, 305) installed inside the left and right armrests, which provide non-invasive tactile feedback through vibration coding. It is suitable for users with hearing or visual impairments and enhances the diversity and coverage of warning information transmission.

[0042] The preferred vibration coding rules are as follows: A single pulse on the left indicates a risk on the left side; A single pulse on the right side indicates a risk on the right side; Bilateral alternating pulses: Indicates a risk in front or behind; Bilateral continuous strong earthquake: indicates that the current situation is under emergency braking or active intervention.

[0043] Haptic feedback allows for adjustment of vibration intensity and duration via software parameters to suit the different perceptual abilities of users.

[0044] 2.4 Active Intervention Control Unit In a preferred embodiment of the present invention, when the system detects that the risk level has reached the highest level (e.g., level 4) and the user does not change the control command within the specified response time (e.g., continues to move forward), it is determined to be a potential error or operational failure, and the system will trigger the active intervention unit.

[0045] The unit sends intervention commands to the drive controller (307) of the electric wheelchair via the CAN bus (306) to perform the following actions: Gradual deceleration: Based on the current speed curve and the speed at which the obstacle approaches, the speed of acceleration is smoothly reduced, with a maximum of no more than 2 m / s², to avoid the inertial impact and risk of falling caused by sudden braking; Assisted steering: By adjusting the output torque of the left and right motors, a slight steering tendency is created to deviate from the risky direction, guiding the wheelchair to naturally avoid obstacles. This tendency can be overridden by the user, ensuring "human-machine co-driving" and not depriving the user of control.

[0046] The aforementioned intervention mechanism respects user intent while ensuring safety, and its practicality and robustness in complex scenarios have been verified through multiple experiments. It is particularly suitable for user groups with long response times or rapidly changing environments.

[0047] III. Implementation Methods of Personalized Learning Modules In a preferred embodiment of the present invention, such as Figure 2 As shown, the personalized learning module (400) is used to implement dynamic learning and adaptive optimization functions based on user behavior. By continuously recording the user's operation patterns and risk response history, this module can automatically adjust the risk assessment weights and early warning strategies according to the usage characteristics of different users, thereby achieving highly personalized behavior adaptation and long-term system evolution.

[0048] 3.1 User behavior data recording mechanism The personalized learning module includes a data recording subsystem for real-time collection and storage of user responses to different warning events. Specific recorded content includes, but is not limited to: Event type: risk level, obstacle direction and category; Operation response: changes in joystick commands, whether the braking signal is triggered, and whether the direction of movement is changed; Results feedback: whether obstacle avoidance was successful, closest distance, and whether a collision was triggered; Environmental conditions: such as light intensity, road surface type and other auxiliary parameters.

[0049] The aforementioned data is stored locally and periodically written to non-volatile storage media, and used for behavioral modeling in subsequent learning processes. Preferably, the data sampling period can be set to 100ms~500ms to ensure a balance between recording accuracy and system response performance.

[0050] 3.2 Reinforcement Learning Optimization Strategies In an embodiment of the present invention, the learning module optimizes the parameters in the risk assessment engine based on the Q-learning algorithm. Q-learning is a model-free reinforcement learning method suitable for behavioral decision-making scenarios with high-dimensional state spaces and sparse feedback.

[0051] The learning objective function is defined as simultaneously minimizing the number of false alarms and unnecessary interventions, while maximizing the success rate of avoiding genuine dangers. A reward mechanism is constructed using the weights (w_1, w_2, ..., w_n) of each dimension in risk assessment and the risk thresholds (T_1, T_2, ..., T_n) of each level as state parameters, combined with user action results. Successfully avoiding obstacles in advance → Positive reward; Failure to react or misjudgment leading to a collision → negative reward; Frequent false alarms with no risk → Penalize false alarms; Timely response but system intervention → penalty for redundant intervention.

[0052] The system updates the Q-value table after each warning cycle and periodically adjusts the weights and threshold parameters. As usage time accumulates, the system strategy gradually converges to the optimal behavioral strategy, more accurately reflecting user behavioral preferences and risk tolerance.

[0053] 3.3 Mode Switching Mechanism To accommodate the behavioral differences among different user groups (such as beginners, the elderly, or professional rehabilitation personnel), the system of this invention provides three preset operating modes: "Beginner Mode," "Standard Mode," and "Expert Mode," each corresponding to different sensitivity parameters and intervention strategies.

[0054] Beginner Mode: The system has the highest alert sensitivity and tends to intervene proactively, making it suitable for those who are in the early stages of learning or have a slower cognitive response. Standard mode: Moderate sensitivity, balanced human-machine control; Expert mode: The system only intervenes in extremely high-risk situations, relying more on the user's subjective judgment, and is suitable for skilled operators.

[0055] Users or caregivers can manually switch modes through a graphical interface, or the system can automatically recommend and prompt modes based on behavioral data. For example, if the system records for several consecutive days that a user frequently ignores prompts but has a high success rate in obstacle avoidance, it will suggest switching to "Expert Mode".

[0056] The introduction of this personalized learning module enables the system of this invention to have the ability to continuously evolve, autonomously optimize, and adapt to individual users, effectively improving user experience, reducing false alarm rate, and enhancing the reliability and flexibility of security protection in long-term use.

[0057] In summary, this invention provides an obstacle avoidance and early warning system suitable for electric wheelchairs. It possesses comprehensive environmental perception, multi-dimensional risk assessment, multi-modal hierarchical early warning, and flexible proactive intervention capabilities. Combined with a personalized learning module, it can dynamically optimize system parameters based on user behavior, significantly improving the safety, intelligence, and user adaptability of electric wheelchairs in complex environments. The system has a compact structure, clearly defined modules, and good scalability and industrial application prospects.

[0058] It should be noted that the specific embodiments described above are merely preferred embodiments of the present invention, intended to aid in understanding the technical solution of the present invention, and are not intended to limit the scope of protection of the present invention. All equivalent modifications or substitutions made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An obstacle avoidance and early warning system suitable for electric wheelchairs, characterized in that, include: A multi-sensor sensing module, installed on the electric wheelchair, is used to collect multimodal data of the surrounding environment in real time; The data processing and risk assessment module is used to perform time synchronization and spatial fusion of the multimodal data, construct a 360-degree dynamic environment map centered on the wheelchair, and perform quantitative risk assessment based on at least three dimensions, outputting risk level signals. The intelligent early warning and decision-making module executes a four-level hierarchical early warning strategy based on the risk level signal, including visual, auditory and tactile prompting units as well as an active intervention unit; The personalized learning module is used to dynamically optimize the weights and thresholds of the risk assessment model based on the user's historical operation and warning result data, and supports switching between multiple warning modes.

2. The system according to claim 1, characterized in that, The multi-sensor sensing module includes at least two types of sensors selected from lidar, ultrasonic sensors, cameras, and inertial measurement units, for the purpose of acquiring and fusing multi-source information.

3. The system according to claim 1, characterized in that, The risk assessment uses multiple dimensions, including but not limited to: obstacle distance, relative speed, direction of movement, ambient light, road surface conditions, and user behavior characteristics, of which at least three dimensions are selected for risk quantification calculation.

4. The system according to claim 1, characterized in that, The visual warning unit includes a graphic display screen and RGB LED light strips set around the wheelchair base, which are used to synchronously display an environmental semantic map and indicate the direction of risk through color and directional lighting effects.

5. The system according to claim 1, characterized in that, The auditory warning unit includes a dual-channel speaker and a speech synthesis chip. The speaker supports 3D sound field rendering, and the speech synthesis chip generates graded voice prompts based on the risk level.

6. The system according to claim 1, characterized in that, The tactile warning unit includes linear resonant actuators embedded in the left and right armrests, which generate tactile feedback of different parts and intensities through pulses to indicate the location and level of risk.

7. The system according to claim 1, characterized in that, The active intervention unit is activated when the highest level of risk is detected and the user does not respond in time. The unit sends control commands to the wheelchair drive controller via the CAN bus to perform progressive deceleration or assisted steering operations.

8. The system according to claim 7, characterized in that, The progressive deceleration controls the wheelchair's deceleration in a smooth, upward manner, with a maximum deceleration not exceeding 2 m / s². 2 Assisted steering achieves steering tendency by adjusting the torque difference between the left and right wheel motors, and this tendency can be actively controlled by the user.

9. The system according to claim 1, characterized in that, The personalized learning module employs a Q-learning-based reinforcement learning algorithm with the optimization goal of reducing false positive rate and improving risk avoidance success rate, and automatically adjusts the weight coefficients and judgment thresholds of each risk factor.

10. The system according to claim 1, characterized in that, The system supports switching between novice mode, standard mode, and expert mode. Each mode has different risk sensitivity and intervention proactivity to adapt to the needs of users with different ability levels.

Images