Intelligent bicycle obstacle avoidance system with multi-mode sensing and active pneumatic control functions

Abstract

The invention discloses a multi-modal sensing and active pneumatic control bicycle intelligent obstacle avoidance system. The system comprises a sensing module, an edge calculation module, a multi-actuator cooperative control module, a communication module, a power supply module and a data preprocessing and calibration module. The sensing module is used for collecting bicycle driving state data and external environment data in real time, and the edge calculation unit is used for fusing the data output by the sensing module and dynamically generating a control instruction of the multi-execution-mechanism cooperative control module; the multi-actuating mechanism cooperative control module comprises a pneumatic fin unit, a differential clutch unit, a liquid metal counterweight unit and a gyro flywheel unit; the communication module adopts a dual-CAN bus redundancy architecture; the power supply module comprises a super capacitor emergency power supply unit which can maintain the operation of a core function within 60s when a main power supply fails. The data preprocessing and calibration module is used for solving the problems of sensor noise and data asynchronous interference; the system solves the problem of'temperature-magnetic field-displacement 'triangular deviation.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent obstacle avoidance systems for bicycles, and specifically refers to an intelligent obstacle avoidance system for bicycles with multi-modal sensing and active pneumatic control. BACKGROUND

[0002] Traditional bicycle obstacle avoidance systems mostly use relatively simple designs and fixed control parameters, making it difficult to achieve efficient and reliable obstacle avoidance in complex and variable riding environments. In high-speed riding scenarios (speed > 20 km / h), when encountering suddenly appearing obstacles, traditional obstacle avoidance systems often can only achieve a small lateral deviation distance, usually less than 1 m, due to insufficient lateral thrust. This makes it difficult to effectively avoid obstacles when driving at high speed, and is extremely prone to collision accidents, seriously threatening the personal safety of cyclists. For example, in urban roads, when a bicycle is traveling at high speed and suddenly encounters a pedestrian crossing in front, the traditional obstacle avoidance system cannot provide enough obstacle avoidance force, making the bicycle unable to avoid in time, increasing the risk of accidents.

[0003] In low-speed scenarios (speed < 10 km / h), traditional obstacle avoidance systems also have the problem of energy efficiency imbalance, with power consumption usually greater than 25 W. This means that when driving at low speed, the system needs to consume a large amount of energy to maintain operation, not only reducing energy utilization efficiency, but also possibly shortening battery life, increasing use cost. For example, in the case of an electric bicycle, high power consumption when driving at low speed will cause the battery to quickly run out of power, affecting the continuity and convenience of cycling.

[0004] In addition, existing bicycle obstacle avoidance technology often uses a single actuator or fixed transmission mode, which has obvious limitations. A single actuator cannot meet the obstacle avoidance requirements at different speeds, and cannot provide enough power and stability at high speed, and cannot achieve fine control and posture adjustment at low speed. Fixed transmission mode cannot be dynamically adjusted according to actual road conditions and riding state, making the bicycle less adaptable to different road conditions and unable to fully utilize the performance of the obstacle avoidance system. For example, when encountering wet and slippery road surfaces, the fixed transmission mode cannot adjust the torque distribution in time, easily causing the wheels to slip and affecting the obstacle avoidance effect.

[0005] To this end, an obstacle avoidance bicycle is proposed on the market, which includes a perception module, an edge computing unit, a multi-actuator collaborative control module, a communication module and a power supply module. The multi-actuator collaborative control module includes a liquid metal counterweight unit. The liquid metal counterweight unit is mainly a gallium indium tin alloy. In the process of the liquid metal counterweight unit performing obstacle avoidance actions, when faced with extreme temperature changes such as sudden heavy rain after summer sun exposure or entering the indoor environment in winter due to the coupling effect of the thermal expansion and contraction characteristics of the gallium indium tin alloy and the stability of the electromagnetic drive magnetic field, a "temperature-magnetic field-displacement" triangle deviation problem will occur. Specifically, at the hardware physical level, when the temperature change rate exceeds 5°C / min, the gallium indium tin alloy 26×10 -6 / ℃ volume expansion coefficient will cause the actual displacement of the counterweight block to deviate by ±0.5mm from the electromagnetic drive command. The significant change in alloy viscosity from 2.9mPa·s at 20℃ to 1.5mPa·s at 100℃ will cause its fluidity in the electromagnetic coil array to mutate, thereby causing displacement hysteresis. The temperature-error compensation mapping table only considers the influence of a single temperature variable on displacement accuracy, and does not cover the attenuation of the driving magnetic force caused by the increase in copper coil resistance with increasing temperature. The magnetic field strength decreases by 0.002T for every 1℃ increase, ultimately forming a "temperature → alloy physical properties + coil magnetic field strength dual variable coupling deviation", which makes the ±1mm displacement accuracy requirement invalid. During high-speed obstacle avoidance, there may be a risk of secondary collision due to insufficient body roll compensation. Summary of the Invention

[0006] In order to overcome the shortcomings of the existing technology, the present application provides a bicycle intelligent obstacle avoidance system with multi-modal sensing and active pneumatic control, which solves the problem of "temperature-magnetic field-displacement" triangle deviation caused by the coupling of the thermal expansion and contraction characteristics of gallium indium tin alloy and the stability of the electromagnetic drive magnetic field when the liquid metal counterweight unit performs obstacle avoidance actions, when facing extreme temperature changes such as sudden heavy rain after summer sun exposure or entering the indoor environment in winter.

[0007] The present invention provides a bicycle intelligent obstacle avoidance system with multi-modal sensing and active pneumatic control, comprising a perception module, an edge computing unit, a multi-actuator collaborative control module, a communication module, a power supply module, and a data preprocessing and calibration module; The sensing module is used to collect real-time data on the bicycle's driving status and external environment. The sensing module includes at least a millimeter-wave radar, an ultrasonic array, and an inertial measurement unit. The millimeter-wave radar is used to detect the distance and speed of obstacles within a range of 30m±0.1m; the ultrasonic array is used to obtain the position information of near-field obstacles with an accuracy of ±3cm; and the inertial measurement unit is used to detect the bicycle's roll angle and yaw angular velocity. The edge computing unit adopts an ARM Cortex-M7 and NPU heterogeneous computing architecture, is used for fusion processing of multi-source data output by the perception module, real-time calculation of obstacle threat level, and dynamic generation of control instructions of the multi-actuator cooperative control module based on vehicle speed, obstacle distance, road friction coefficient and vehicle body posture parameters; The multi-actuator cooperative control module includes a pneumatic wing unit, a differential clutch unit, a liquid metal weight unit and a gyro flywheel unit; the pneumatic wing unit is driven by a high-pressure air pump and a servo motor to realize a continuously adjustable deployment angle of 0-45°, and is used for generating lateral thrust or attitude compensation force; the differential clutch unit supports a dynamically adjustable torque distribution ratio of 1:1 to 1:3; the liquid metal weight unit realizes a weight displacement of ±15cm±1mm precision through electromagnetic drive of an electromagnetic coil; the gyro flywheel unit adopts carbon fiber material, and the rotating speed can be adjusted in the range of 20000-23000rpm; Among them, a three-mode sensor array is embedded in the liquid metal weight unit, including a platinum resistance temperature sensor PT1000, a miniature Hall magnetic field sensor and an optical fiber grating strain sensor, forming a composite sensing structure and realizing high coupling with the temperature control unit; the platinum resistance temperature sensor PT1000 and the miniature Hall magnetic field sensor can monitor the working temperature of the liquid metal and the magnetic field intensity of the electromagnetic coil in real time; the sensing data of the platinum resistance temperature sensor PT1000, the miniature Hall magnetic field sensor and the optical fiber grating strain sensor are connected with the edge computing unit through a high-speed bus connection through a synchronous clock acquisition interface, and the sampling data fusion enters a temperature-magnetic field-deformation three-dimensional state matrix, and the running state vector of the liquid metal weight unit is updated in real time; The communication module adopts a dual-CAN bus redundant architecture to realize real-time data interaction between the perception module, the edge computing unit and the multi-actuator cooperative control module; The power supply module includes a super capacitor emergency power supply unit, which can maintain the core function running for 60s when the main power supply fails; The data preprocessing and calibration module is used to solve the problems of sensor noise and data asynchronous interference, and at least includes a sensor noise hierarchical suppression unit, a preprocessing and model linkage unit and a temperature-magnetic field coupling compensation unit; Among them, by implanting semiconductor refrigeration sheets and flexible heating films in the temperature-magnetic field coupling compensation unit, a micro-scale temperature control unit with an accuracy of ±0.5°C is constructed, which is used to precisely adjust the working temperature of the liquid metal counterweight medium to suppress the displacement error and control hysteresis caused by thermal expansion and contraction effects; and an algorithm is used to establish a multi-physics field mapping model based on the magnetohydrodynamic equation; when the edge computing unit detects that the temperature fluctuation exceeds ±1°C, a two-stage compensation mechanism is automatically triggered: the first-stage compensation adjusts the driving current through the PID algorithm (compensation coefficient 0.8A / °C) to maintain the stability of the magnetic field; the second-stage compensation starts the temperature control unit and activates the newly added temperature-current mapping hidden layer (16 neurons) of the BP neural network, and optimizes the electromagnetic coil drive parameters in real time through the dual CAN bus, ultimately maintaining the counterweight displacement accuracy within ±0.3mm under extreme temperatures, ensuring the dynamic consistency of the body roll compensation during high-speed obstacle avoidance.

[0008] Furthermore, according to the multi-modal sensing and active pneumatic control bicycle intelligent obstacle avoidance system provided in the present application, the liquid metal is a gallium indium tin alloy. Furthermore, according to the multi-modal sensing and active pneumatic control bicycle intelligent obstacle avoidance system provided by the present application, the platinum resistance temperature sensor PT1000 is mounted on the high thermal conductivity path of the counterweight cavity wall in the liquid metal counterweight unit through precision positioning tooling to ensure the thermal conductivity efficiency between the platinum resistance temperature sensor PT1000 and the liquid metal alloy contact interface, and the temperature data inside the liquid metal counterweight cavity is obtained in real time using a sampling frequency of 100Hz. Through temperature readings with an accuracy of ±0.05°C, the thermal field distribution of the liquid metal alloy is accurately portrayed, providing real-time feedback for the temperature control loop. Furthermore, according to the multimodal sensing and active pneumatic control bicycle intelligent obstacle avoidance system provided in the present application, the micro-Hall sensors are arranged at both ends of the liquid metal flow path in the liquid metal counterweight unit. The micro-Hall sensors monitor the changes in local magnetic field intensity generated by the excitation of the electromagnetic drive coil. The 0.5mT resolution of the micro-Hall sensor can capture tiny magnetic field disturbances, and then judge the consistency of the driving magnetic flux distribution and the electromagnetic force field response, and assist in controlling the closed-loop regulation of the driving current of the electromagnetic coil. Furthermore, according to the multimodal sensing and active pneumatic control bicycle intelligent obstacle avoidance system provided in the present application, the fiber Bragg grating strain sensor is arranged in the high strain area of ​​the liquid metal counterweight cavity. The fiber Bragg grating strain sensor measures the elastic deformation of the structure of the liquid metal counterweight cavity under electromagnetic excitation in real time through the wavelength drift signal. The accuracy of the fiber Bragg grating strain sensor reaches 1με, which can effectively identify the changes in viscous resistance and wall friction characteristics of the alloy medium under different displacement conditions. Furthermore, according to the multimodal sensing and active pneumatic control bicycle intelligent obstacle avoidance system provided by the present application, the semiconductor refrigeration chip realizes the cooling function when powered on through the Peltier effect. The cold end of the semiconductor refrigeration chip is directly attached to the heat conduction enhanced area on the outer wall of the liquid metal counterweight cavity. The interfacial gap is filled with high thermal conductivity silicone grease to reduce thermal resistance. At the same time, the hot end of the semiconductor refrigeration chip is connected to a graphite heat conductive sheet and an aluminum heat sink array for efficient heat dissipation. The flexible heating film is attached to the symmetrically distributed area outside the liquid metal counterweight cavity. The structure of the flexible heating film has low thermal inertia and planar heating characteristics, and can supplement heat in an instantaneous response to offset the effects of ambient cooling or overcooling of the refrigeration end.

[0009] Furthermore, according to the multimodal sensing and active pneumatic control bicycle intelligent obstacle avoidance system provided by the present application, the microscale temperature control unit uses the real-time feedback cavity temperature value of the platinum resistance temperature sensor PT1000 as the control reference input, and compares it with the preset target temperature. The cooling current of the semiconductor refrigeration plate and the heating power of the flexible heating film are adjusted through the PID closed-loop control algorithm. The specific proportional coefficient is set to Kp equal to 2.5, the integral time is 100 milliseconds, and the differential time is 20 milliseconds, achieving a dynamic balance between rapid response to temperature changes and steady-state accuracy. In the initial heating stage, when the feedback temperature of the platinum resistance temperature sensor PT1000 is lower than the preset target temperature, the microscale temperature control unit outputs a PWM signal through the PID controller to drive the flexible heating film to gradually heat up. As the temperature approaches the preset target temperature, the PID controller gradually reduces the heating power according to the deviation change trend until the preset target temperature is reached; when the external ambient temperature rises or the electromagnetic excitation generates Joule heat causing the cavity to heat up, the system starts the semiconductor refrigeration plate, controls the cold end heat flow of the semiconductor refrigeration plate to be extracted from the inside of the liquid metal counterweight cavity, and discharged to the heat dissipation system through the hot end of the semiconductor refrigeration plate, and adjusts the cooling current amplitude in real time to maintain temperature stability. Furthermore, according to the multimodal sensing and active pneumatic control bicycle intelligent obstacle avoidance system provided in this application, a multi-physics field mapping model is established based on the magnetohydrodynamic equations, with the electromagnetic driving force as the core driving quantity, and its expression is: ; in, It represents the total driving force per unit volume of liquid metal, is the current density vector, is the magnetic field intensity vector, ρ is the density of liquid metal, is the gravitational acceleration vector, μ is the dynamic viscosity of the liquid metal, is the flow velocity vector; The mapping model constructs a three-dimensional physical field based on the geometric structure of the liquid metal counterweight cavity, and forms the current density distribution by setting the current input boundary conditions of the electromagnetic coil. , and at the same time calibrate the magnetic field distribution using the measured data of the micro Hall magnetic field sensor , to ensure the spatial accuracy of magnetic field modeling; After the platinum resistance temperature sensor PT1000 outputs a real-time temperature signal, the mapping model dynamically corrects the density ρ and dynamic viscosity μ according to the thermophysical property response curve of the liquid metal material, capturing the impact of temperature changes on the fluid mechanics properties; in the edge computing unit, the real-time sensing data is input into a pre-built coupling model database or a fast solver is used to obtain the force field distribution and flow velocity solution at the current moment, thereby inferring the driving current, electromagnetic waveform and frequency modulation required for compensation; this compensation mechanism ultimately realizes the dynamic coupling closed-loop adjustment of the control signal and the liquid metal state, so that the counterweight displacement process has higher accuracy, response speed and environmental robustness, thereby significantly improving the stability and execution efficiency of the obstacle avoidance system.

[0010] Furthermore, according to the multimodal sensing and active pneumatic control bicycle intelligent obstacle avoidance system provided by the present application, in the process of sensor data fusion processing, the temperature-magnetic field coupling compensation unit uses an improved extended Kalman filter to accurately estimate the state of the multimodal signal, and constructs an enhanced state vector by introducing temperature and magnetic field state components. , where x, y, z represent the position of the liquid metal in the three-dimensional coordinate system, T represents the current temperature of the liquid metal cavity, and B represents the local magnetic field intensity under the excitation of the electromagnetic coil; Based on this state vector, the state transition equation of the extended Kalman filter is updated as follows: ; in, is the predicted state vector for the next moment, is the current state, F is the state transfer matrix, which represents the natural evolution law of the system without control input, G is the control matrix, is the control input, which represents the adjustment instruction of the electromagnetic drive parameters by the edge computing unit. is the process noise term, which is used to model disturbance factors and dynamic errors that are not explicitly considered; This state prediction mechanism embeds the viscosity drift caused by temperature changes and the nonlinear changes in the force field caused by magnetic field intensity fluctuations into the state update process, enabling the extended Kalman filter to actively perceive the impact of the physical environment on displacement behavior; During the observation phase, the temperature-magnetic field coupling compensation unit constructs an observation vector from the three-modal data collected from the platinum resistance temperature sensor PT1000, the micro Hall magnetic field sensor and the optical fiber strain sensor. , and its observation model is expressed as: ; Among them, H is the observation matrix, which defines the mapping relationship between the sensor observation value and the internal state vector. is the observation noise term, taking into account sensor accuracy and sampling error; The extended Kalman filter calculates the Kalman gain matrix based on the residual between the current predicted state and the actual observation , and updates the state estimate accordingly, thereby achieving fine compensation for the actual displacement state of the liquid metal counterweight; the real-time introduction of temperature and magnetic field states enables the extended Kalman filter to actively capture the impact of environmental disturbance factors on position prediction, thereby maintaining a high-precision estimate of the counterweight position under conditions of significant or rapid fluctuations in magnetic and thermal interference. Furthermore, according to the multimodal sensing and active pneumatic control bicycle intelligent obstacle avoidance system provided by the present application, in the bipolar compensation mechanism of the temperature-magnetic field coupling compensation unit, the first-level compensation is activated when the temperature fluctuation is detected to be between ±1 and ±2 degrees Celsius, and current compensation is performed only for the electromagnetic drive module. Based on the feedback temperature deviation ΔT, the temperature-magnetic field coupling compensation unit adjusts the driving current value of the electromagnetic coil according to a linear proportional coefficient of 0.8 amperes per degree Celsius. The compensation formula is: ; in is the adjusted driving current, is the current setting current, and ΔT is the temperature deviation. The first-level compensation strategy quickly responds to the fluctuation of magnetic field intensity caused by thermal drift, stabilizes the output of electromagnetic driving force, and thus maintains the accuracy of the liquid metal counterweight displacement trajectory; If the temperature deviation further expands to more than ±2 degrees Celsius, the secondary compensation mechanism is triggered, and the semiconductor refrigeration plate and the flexible heating film are activated at the same time. The PID temperature control strategy is used to restore the liquid metal working temperature to the set range within 100 milliseconds, and the predictive control capability is further enhanced.

[0011] The beneficial effects of the present invention are as follows: the multimodal sensing and active pneumatic control bicycle intelligent obstacle avoidance system provided by the present application improves displacement accuracy by using a trimodal sensor array to monitor temperature, magnetic field, and deformation in real time, combined with a micro-temperature control system to maintain the liquid metal temperature at 25±0.5°C, and dynamically correcting displacement deviations through a multi-physics field model and an improved Kalman filter, thereby improving displacement accuracy from ±1mm to within ±0.3mm at extreme temperatures. To optimize response speed, a two-stage compensation mechanism is constructed. The first-stage compensation quickly adjusts the drive current, and the second-stage compensation activates the temperature control unit. Combined with an improved BP neural network to predict temperature trends, the drive parameters are corrected in advance, shortening the response delay to within 200ms when the temperature changes suddenly, effectively avoiding the risk of secondary collisions. In terms of enhancing scenario adaptability, the collaborative work of hardware composite sensing and algorithm multi-physics field modeling, combined with a four-element closed-loop control chain, realizes dynamic feedback of temperature, magnetic field, displacement and current, enabling the system to operate stably in the temperature range of -20℃~80℃, fully covering extreme scenarios such as summer rainstorms and winter indoor conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The following detailed description of the specific embodiments of the present application in conjunction with the accompanying drawings will make the technical solutions and other beneficial effects of the present application apparent.

[0013] Figure 1 This is a structural diagram of the bicycle intelligent obstacle avoidance system with multi-modal sensing and active pneumatic control provided in this embodiment.

[0014] Figure 2 This is a signaling diagram of the multi-modal sensing and active pneumatic control bicycle intelligent obstacle avoidance system provided in this embodiment. DETAILED DESCRIPTION

[0015] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the embodiments described are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0016] In the description of the present application, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present application. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the said features. In the description of the present application, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0017] The disclosure below provides many different embodiments or examples for realizing different structures of the present application. In order to simplify the disclosure of the present application, the components and settings of specific examples are described below. Of course, they are merely examples and are not intended to limit the present application. In addition, the present application may repeat reference numbers and / or reference letters in different examples, and such repetition is for the purpose of simplicity and clarity, and does not itself indicate the relationship between the various embodiments and / or settings discussed. In addition, the present application provides examples of various specific processes and materials, but those of ordinary skill in the art will appreciate the application of other processes and / or the use of other materials.

[0018] The embodiments of the present application are now further described with reference to the accompanying drawings and specific implementation methods.

[0019] When the liquid metal counterweight unit performs obstacle avoidance actions, when faced with extreme temperature changes such as sudden heavy rain after summer sun exposure or entering a low-temperature environment indoors in winter, the "temperature-magnetic field-displacement" triangle deviation problem will occur due to the coupling of the thermal expansion and contraction characteristics of the gallium indium tin alloy and the stability of the electromagnetic drive magnetic field. This embodiment provides a bicycle intelligent obstacle avoidance system with multi-modal sensing and active pneumatic control to solve the above problem.

[0020] Figure 1 This is a structural diagram of the bicycle intelligent obstacle avoidance system with multi-modal sensing and active pneumatic control provided in this embodiment.

[0021] like Figure 1 As shown, the multimodal sensing and active pneumatic control bicycle intelligent obstacle avoidance system includes a perception module, an edge computing unit, a multi-actuator collaborative control module, a communication module, a power supply module, and a data preprocessing and calibration module; The sensing module is used to collect real-time data on the bicycle's driving status and external environment. The sensing module includes at least a millimeter-wave radar, an ultrasonic array, and an inertial measurement unit. The millimeter-wave radar is used to detect the distance and speed of obstacles within a range of 30m±0.1m; the ultrasonic array is used to obtain the position information of near-field obstacles with an accuracy of ±3cm; and the inertial measurement unit is used to detect the bicycle's roll angle and yaw angular velocity. The edge computing unit uses an ARM Cortex-M7 and NPU heterogeneous computing architecture to fuse and process the multi-source data output by the perception module, calculate the obstacle threat level in real time, and dynamically generate control instructions for the multi-actuator collaborative control module based on vehicle speed, obstacle distance, road friction coefficient, and vehicle body posture parameters; The multi-actuator coordinated control module includes a pneumatic vane unit, a differential clutch unit, a liquid metal counterweight unit, and a gyro flywheel unit. The pneumatic vane unit is driven by a high-pressure air pump and a servo motor to achieve a continuously adjustable deployment angle of 0-45° for generating lateral thrust or attitude compensation force. The differential clutch unit supports a dynamically adjustable torque distribution ratio of 1:1 to 1:3. The liquid metal counterweight unit is electromagnetically driven by an electromagnetic coil to achieve a counterweight displacement with an accuracy of ±15cm±1mm. The gyro flywheel unit is made of carbon fiber and has an adjustable speed within the range of 20,000-23,000 rpm. Among them, a three-modal sensor array is embedded in the liquid metal counterweight unit, including a platinum resistance temperature sensor PT1000, a micro Hall magnetic field sensor and a fiber Bragg grating strain sensor, forming a composite sensing structure and achieving a high degree of coupling with the temperature control unit; the platinum resistance temperature sensor PT1000 and the micro Hall magnetic field sensor can monitor the working temperature of the liquid metal and the magnetic field strength of the electromagnetic coil in real time; the sensing data of the platinum resistance temperature sensor PT1000, the micro Hall magnetic field sensor and the fiber Bragg grating strain sensor are connected to the edge computing unit through a high-speed bus through a synchronous clock acquisition interface, and the sampled data are fused into the temperature-magnetic field-deformation three-dimensional state matrix to update the operating state vector of the liquid metal counterweight unit in real time; The communication module adopts a dual CAN bus redundant architecture to achieve real-time data interaction between the perception module, the edge computing unit and the multi-actuator collaborative control module; The power supply module includes a supercapacitor emergency power supply unit, which can maintain the core functions for 60 seconds when the main power supply fails; The data preprocessing and calibration module is used to solve the problem of asynchronous interference between sensor noise and data. The data preprocessing and calibration module at least includes a sensor noise layer suppression unit, a preprocessing and model linkage unit, and a temperature-magnetic field coupling compensation unit; Among them, by implanting semiconductor refrigeration sheets and flexible heating films in the temperature-magnetic field coupling compensation unit, a micro-scale temperature control unit with an accuracy of ±0.5°C is constructed, which is used to precisely adjust the working temperature of the liquid metal counterweight medium to suppress the displacement error and control hysteresis caused by thermal expansion and contraction effects; and an algorithm is used to establish a multi-physics field mapping model based on the magnetohydrodynamic equation; when the edge computing unit detects that the temperature fluctuation exceeds ±1°C, a two-stage compensation mechanism is automatically triggered: the first-stage compensation adjusts the driving current through the PID algorithm (compensation coefficient 0.8A / °C) to maintain the stability of the magnetic field; the second-stage compensation starts the temperature control unit and activates the newly added temperature-current mapping hidden layer (16 neurons) of the BP neural network, and optimizes the electromagnetic coil drive parameters in real time through the dual CAN bus, ultimately maintaining the counterweight displacement accuracy within ±0.3mm under extreme temperatures, ensuring the dynamic consistency of the body roll compensation during high-speed obstacle avoidance.

[0022] Specifically, in this embodiment, the liquid metal is made of gallium-indium-tin alloy material.

[0023] The platinum resistance temperature sensor PT1000 and the micro Hall magnetic field sensor are further integrated into the liquid metal counterweight unit. The sampling frequency of the platinum resistance temperature sensor PT1000 is synchronized with the inertial measurement unit (both are 100 Hz), which can monitor the working temperature of the gallium indium tin alloy and the magnetic field strength of the electromagnetic coil in real time.

[0024] Correspondingly, a temperature-magnetic field coupling compensation unit is added to the data preprocessing and calibration module: At the hardware level, semiconductor cooling chips and flexible heating films are implanted to build a micro-temperature control system with an accuracy of ±0.5°C, suppressing the thermal expansion and contraction effects of the alloy. At the algorithm level, a multi-physics field mapping model is established based on the magnetohydrodynamic equations, which converts the viscosity change caused by temperature (26×10 -6 / ℃) and magnetic field attenuation (0.002T / ℃) parameters are embedded in the extended Kalman filter, and the state vector Dynamic correction of displacement deviation is achieved; where x, y, z represent the position of the liquid metal in the three-dimensional coordinate system, T represents the current temperature of the liquid metal cavity, and B represents the local magnetic field strength under the excitation of the electromagnetic coil.

[0025] When the edge computing unit detects a temperature fluctuation exceeding ±1°C, it automatically triggers a two-stage compensation mechanism: the first-stage compensation adjusts the drive current through the PID algorithm (compensation coefficient 0.8A / °C) to maintain magnetic field stability. The second-stage compensation starts the temperature control unit and activates the newly added temperature-current mapping hidden layer (16 neurons) of the BP neural network. The electromagnetic coil drive parameters are optimized in real time via the dual CAN buses, ultimately maintaining the counterweight displacement accuracy within ±0.3mm under extreme temperatures, ensuring the dynamic consistency of body roll compensation during high-speed obstacle avoidance.

[0026] Specifically, during the system's hardware integration process, a three-modal sensor array is embedded inside the liquid metal counterweight unit, consisting of a platinum resistance temperature sensor PT1000, a micro Hall effect magnetic field sensor, and a fiber Bragg grating strain sensor, forming a composite sensing structure and achieving a high degree of coupling with the temperature control unit. First, in the liquid metal counterweight cavity, the platinum resistance temperature sensor PT1000 is mounted on the high thermal conductivity path of the counterweight cavity wall using a precision positioning fixture to ensure the thermal conductivity efficiency between it and the alloy contact interface. A 100Hz sampling frequency is used to obtain real-time temperature data inside the cavity. With temperature readings with an accuracy of ±0.05°C, the alloy's thermal field distribution is accurately characterized, providing real-time feedback for the temperature control loop. At the same time, micro Hall effect sensors are arranged at both ends of the liquid metal flow path to monitor the changes in local magnetic field intensity generated by the excitation of the electromagnetic drive coil. The 0.5mT resolution of the Hall effect sensor is sufficient to capture tiny magnetic field disturbances, thereby determining the consistency of the driving magnetic flux distribution with the electromagnetic force field response, and assisting in controlling the closed-loop regulation of the driving current of the electromagnetic coil. To achieve structural monitoring of the alloy's state during counterweight displacement, fiber Bragg grating strain sensors are placed in the high-strain zone of the counterweight conduit. Wavelength drift signals are used to measure the elastic deformation of the counterweight cavity structure under electromagnetic excitation in real time with an accuracy of 1με. This effectively identifies changes in viscous resistance and wall friction characteristics of the alloy medium under different displacement conditions. These three types of sensor data are connected to the edge computing unit via a high-speed bus via a synchronous clock acquisition interface. The sampled data is fused into a three-dimensional temperature-magnetic field-deformation state matrix, which updates the operating state vector of the liquid metal counterweight unit in real time.

[0027] On this basis, the system constructs a temperature control model and electromagnetic compensation mechanism. The platinum resistance temperature sensor PT1000 feeds temperature into the thermal control module, driving a micro-heater or thermoelectric cooler to maintain the alloy within the specified melting range. The Hall effect signal adjusts the electromagnetic drive current waveform to suppress electromagnetic drive delay or nonlinear hysteresis response. The optical fiber strain value is used as the input for deformation resistance modeling to modify the drive current prediction model, thereby achieving precise control of the liquid metal counterweight, low-hysteresis response, and enhanced operational stability. The entire composite sensing system and temperature control circuit operate in a coordinated closed loop, enabling the system to maintain precise control of the counterweight position, status, and drive energy efficiency during high-frequency dynamic obstacle avoidance tasks, providing a high-confidence control foundation for the intelligent obstacle avoidance execution layer.

[0028] In this embodiment, the semiconductor refrigeration chip adopts the semiconductor refrigeration chip model TEC1-12706.

[0029] A semiconductor refrigeration chip TEC1-12706 and a flexible heating film are embedded within the electromagnetic coil array structure to form a microscale temperature control unit, which precisely regulates the operating temperature of the liquid metal counterweight medium to suppress displacement errors and control hysteresis caused by thermal expansion and contraction. The TEC1-12706 achieves cooling through the Peltier effect when powered. Its cold end is directly attached to the heat-conducting enhanced area on the outer wall of the liquid metal cavity. High-conductivity silicone grease fills the interfacial gap to reduce thermal resistance, while the hot end is connected to a graphite heat conductor and aluminum heat sink array for efficient heat dissipation. The flexible heating film is symmetrically attached to the exterior of the counterweight cavity. Its structure features low thermal inertia and planar heat generation, enabling instantaneous heat replenishment to offset ambient temperature drops or overcooling of the cold end.

[0030] The micro-scale temperature control unit uses the cavity temperature feedback from the platinum resistance temperature sensor PT1000 as a control reference input. In this embodiment, the target temperature is set at 25 degrees Celsius. A PID closed-loop control algorithm regulates the TEC cooling current and the heating power of the heating film. Specifically, the proportional coefficient Kp is set to 2.5, the integral time is 100 milliseconds, and the differential time is 20 milliseconds. This achieves a dynamic balance between rapid response to temperature changes and steady-state accuracy. During the initial temperature rise phase, when the sensor feedback temperature is lower than the set value, the system outputs a PWM signal through the controller to drive the flexible heating film to gradually increase its temperature. As the temperature approaches the target value, the PID controller gradually reduces the heating power based on the deviation trend until the set temperature range is reached. When the external ambient temperature rises or electromagnetic excitation generates Joule heating, causing the cavity to heat up, the system activates the TEC cooling unit, controlling the cold-end heat flow to be discharged from the liquid metal cavity through the hot end to the heat dissipation system, and adjusting the cooling current amplitude in real time to maintain temperature stability. The PID control process calculates the current temperature error, error integral, and error differential within each sampling cycle, and outputs a comprehensive control variable based on a regulation formula, ensuring that the temperature remains within a range of 25±0.5°C under dynamic conditions. This effectively suppresses volume fluctuations and state response delays in the liquid metal during electromagnetic actuation, ensuring precise control of the counterweight displacement process and consistent coordinated action of multiple actuators. This temperature control mechanism is tightly coupled with the sensor feedback system to form a closed-loop adaptive thermal management module, providing thermal stability for the intelligent obstacle avoidance system.

[0031] In the algorithm design of the microscale temperature control unit, a multi-physics simulation model that couples magnetic, thermal, and fluid fields was constructed based on the COMSOL Multiphysics platform. This model accurately describes the displacement behavior of liquid metal under electromagnetic excitation and guides the edge computing unit to correct control instructions in real time to improve system response accuracy. The coupled model uses electromagnetic driving force as the core driving quantity, and its expression is:

[0032] in, It represents the total driving force per unit volume of liquid metal, is the current density vector, is the magnetic field intensity vector, ρ is the density of liquid metal, is the gravitational acceleration vector, μ is the dynamic viscosity of the liquid metal, is the velocity vector. This equation describes the flow behavior of liquid metal under electromagnetic excitation, which is affected by three types of force fields: electromagnetic force, gravity, and viscous resistance. In the modeling stage, the system constructs a three-dimensional physical field based on the geometric structure of the liquid metal counterweight cavity, and forms the current density distribution by setting the coil current input boundary conditions. , and use the measured data of the Hall magnetic field sensor to calibrate the magnetic field distribution , to ensure the spatial accuracy of magnetic field modeling. After the temperature sensor PT1000 outputs the real-time temperature signal, the system dynamically corrects the density ρ and dynamic viscosity μ according to the thermophysical property response curve of the liquid metal material to capture the impact of temperature changes on the fluid mechanics characteristics. In the edge computing unit, the real-time sensing data is input into the pre-built coupling model database or the force field distribution and flow velocity solution at the current moment are obtained through a fast solver, and then the driving current, electromagnetic waveform and frequency modulation required for compensation are inferred. This compensation mechanism ultimately realizes the dynamic coupling closed-loop adjustment of the control signal and the liquid metal state, so that the counterweight displacement process has higher accuracy, response speed and environmental robustness, thereby significantly improving the stability and execution efficiency of the obstacle avoidance system.

[0033] In the process of sensor data fusion processing, the system uses an improved extended Kalman filter to accurately estimate the state of multimodal signals, and constructs an enhanced state vector by introducing temperature and magnetic field state components. , where x, y, and z represent the position of the liquid metal in the three-dimensional coordinate system, T represents the current temperature of the liquid metal cavity, and B represents the local magnetic field intensity under the excitation of the electromagnetic coil. Based on this state vector, the state transition equation of the extended Kalman filter is updated as follows:

[0034] in, is the predicted state vector for the next moment, is the current state, F is the state transfer matrix, which represents the natural evolution law of the system without control input, G is the control matrix, The control input represents the edge computing unit's adjustment instructions for the electromagnetic drive parameters, while Qk is the process noise term, used to model disturbances and dynamic errors not explicitly considered. This state prediction mechanism embeds viscosity drift caused by temperature changes and nonlinear force field changes caused by magnetic field intensity fluctuations into the state update process, enabling the filter to actively perceive the impact of the physical environment on displacement behavior.

[0035] During the observation phase, the system constructs the observation vector from the three-modal data collected by the platinum resistance temperature sensor, the micro Hall magnetic field sensor and the optical fiber strain sensor. , and its observation model is expressed as:

[0036] Among them, H is the observation matrix, which defines the mapping relationship between the sensor observation value and the internal state vector. is the observation noise term, taking into account sensor accuracy and sampling error. At each moment, the filter calculates the Kalman gain matrix based on the residual between the current predicted state and the actual observation , and updates the state estimate accordingly, achieving precise compensation for the actual displacement state of the liquid metal counterweight. The real-time introduction of temperature and magnetic field states enables the filter to proactively capture the impact of environmental disturbances on position prediction, thereby maintaining a highly accurate estimate of the counterweight position even under conditions of significant magnetic and thermal interference or rapid fluctuations. The system has been proven to stably converge the displacement estimation error to within ±0.3 mm, providing high-confidence closed-loop feedback to the execution layer and significantly improving the control accuracy and robustness of the obstacle avoidance system under complex operating conditions.

[0037] In the system architecture, the existing "sensor-preprocessing-edge computing-actuator" closed-loop has been expanded into a four-element closed-loop control chain, adding a temperature control execution feedback link to achieve comprehensive control of the thermal stability of the liquid metal counterweight system. Temperature and magnetic field, as key variables affecting the liquid metal's response accuracy, are transmitted to the edge computing unit via a dual CAN bus for real-time feedback. Redundant bandwidth is set to 1Mbps to ensure stable transmission of high-frequency data and system fault resistance. PT1000 and Hall effect magnetic field sensors in the sensing layer continuously collect temperature and magnetic flux density. After filtering, denoising, and time synchronization in the preprocessing module, the data is transmitted to the edge computing node along with the execution feedback data for the current control cycle. The edge node performs a sliding window analysis on the current temperature and magnetic field strength and compares them with the expected values. If the temperature fluctuation exceeds ±1°C or the magnetic field strength attenuates by more than 5% compared to the calibration value, the linkage control mechanism is immediately triggered.

[0038] This linkage control mechanism is divided into a two-level compensation strategy. The first level of compensation is activated when the temperature fluctuation is detected between ±1 and ±2 degrees Celsius. It only performs current compensation on the electromagnetic drive module. Based on the feedback temperature deviation ΔT, the system adjusts the drive current value of the electromagnetic coil according to a linear proportional coefficient of 0.8 amperes per degree Celsius. The compensation formula is: ; Where Iadj is the adjusted drive current, Ibase is the current set current, and ΔT is the temperature deviation. This strategy quickly responds to fluctuations in magnetic field strength caused by thermal drift, stabilizing the electromagnetic driving force output, thereby maintaining the accuracy of the liquid metal counterweight displacement trajectory. If the temperature deviation further expands to exceed ±2 degrees Celsius, a secondary compensation mechanism is triggered, activating the semiconductor cooling plate and flexible heating film simultaneously. Using the PID temperature control strategy, the liquid metal operating temperature is restored to the set range within 100 milliseconds, further enhancing predictive control capabilities.

[0039] To achieve active regulation of temperature trends, an improved BP neural network is embedded in the edge computing unit. The network adds two new feature dimensions, "temperature" and "current," to the original "vehicle speed-obstacle distance-friction coefficient" input structure. At the same time, a temperature-current mapping path consisting of 16 neurons is introduced into the hidden layer to establish a nonlinear relationship between the liquid metal's thermal response and the drive current regulation. The network predicts short-term temperature trends based on historical temperature control data and current sensor feedback, and outputs pre-adjusted values ​​for the drive parameters, correcting the current waveform and frequency in advance. This gives the temperature control action a feedforward nature rather than a simple lag response, significantly reducing the system's control delay and energy efficiency loss under temperature disturbances. This four-element closed-loop design forms a stable feedback loop among the four physical quantities of temperature, magnetic field, displacement, and control current, enabling dynamic steady-state regulation and improved control accuracy of the control system under thermal disturbances, comprehensively enhancing the operational robustness and scenario adaptability of the intelligent obstacle avoidance system.

[0040] Figure 2 This is a signaling diagram of the multi-modal sensing and active pneumatic control bicycle intelligent obstacle avoidance system provided in this embodiment.

[0041] Because the existing bicycle obstacle avoidance system still encounters problems including sensor noise and data asynchronous interference during the operation of the parameter dynamic mapping model, the specific manifestation is the distortion of input parameters caused by sensor noise: millimeter-wave radar has a ±0.1m ranging error and multipath reflection interference, and the near-field positioning accuracy of the ultrasonic array in complex environments is easily affected by the surface material of the obstacle (such as mirror reflection causing a ±5cm deviation). The roll angle and yaw angular velocity detected by the inertial measurement unit are interfered with by mechanical vibration noise, and the zero drift error accumulates over time (such as a drift of 0.5° / min). The solution of the road friction coefficient relies on wheel pressure / speed sensors or visual recognition. The former is affected by tire pressure changes (tire pressure ±0.2bar leads to a μ solution error of ±8%), and the latter is restricted by lighting conditions (texture recognition accuracy drops by 30% under low illumination). There is also the problem of spatiotemporal registration failure caused by data asynchrony. The sampling frequencies of multiple sensors are inconsistent (millimeter wave radar 10Hz, ultrasonic wave 20Hz, IMU 100Hz). If not processed synchronously, the vehicle speed, obstacle distance, and friction coefficient data input to the model within the same control cycle will come from different timestamps (the maximum time difference can reach 100ms).

[0042] Therefore, in terms of data processing and control, the parameter dynamic mapping model is susceptible to sensor noise and data asynchronous interference, which leads to erroneous obstacle avoidance decisions. This embodiment effectively solves these problems through a sensor noise layering suppression strategy, a spatiotemporal consistency calibration method for data asynchronous interference, and a closed-loop linkage between the preprocessing module and the mapping model, ensuring that the system can make obstacle avoidance decisions accurately and quickly.

[0043] Specifically, if Figure 2 As shown, in the data preprocessing and calibration module, the sensor noise layering suppression unit is mainly optimized simultaneously from both hardware and algorithm aspects to minimize the noise interference of various sensors in complex environments and improve the accuracy and stability of the original signal.

[0044] At the hardware level, for millimeter-wave radar and ultrasonic arrays, the overall design integrates an anti-electromagnetic interference shielding cover, which forms a closed loop through highly conductive materials to shield the interference of external high-frequency electromagnetic wave radiation on the sensor signal chain. At the same time, an internal grounding design is introduced to quickly discharge the induced current. For the Inertial Measurement Unit (IMU), a multi-degree-of-freedom shock-absorbing bracket made of rubber is used for physical isolation to reduce the impact of vehicle body vibration and road impact on sensor zero drift and high-frequency noise. At the same time, an active low-pass filter is inserted in the inertial measurement unit (IMU) signal output channel. The cutoff frequency of the active low-pass filter is designed to be , in order to ensure the collection of effective motion information and suppress high-frequency noise under typical driving conditions, the low-pass filter transfer function is expressed as: ; Where s is the Laplace operator, is the cut-off angular frequency, is the cutoff frequency of the low-pass filter.

[0045] At the algorithm level, median filtering is applied to the distance and position information collected by the millimeter wave radar and ultrasonic array. Time series data of samples , the median filter output is: ; in, is the output value after filtering, is the input sample of the current and previous moments. The median operation can effectively suppress the impulse noise caused by sudden interference or echo anomalies. Applying a sliding average filter to the three-axis acceleration data output by the IMU accelerometer, assuming the window length is N, the smoothed acceleration output at any moment is: ; in, Indicates the acceleration value after smoothing. Is the current and previous The sliding average method can reduce high-frequency oscillation errors and maintain low latency characteristics to meet the needs of real-time dynamic attitude solution.

[0046] In order to further enhance the ability to proactively identify abnormal sensor outputs, a sensor outlier detection mechanism is established in the system. The statistical distribution of each sensor's historical data is set, and the abnormal threshold range is defined according to the principle of three times the standard deviation as follows: ; in, is the mean of historical data, It is the standard deviation of historical data. Once the real-time monitoring data exceeds this range, it will be marked as an abnormal sample and trigger the data compensation or sensor status evaluation program to ensure the robustness and anti-interference ability of the system's overall perception chain.

[0047] To further improve the accuracy and robustness of sensor signal processing, the system performs algorithm-level dynamic noise compensation based on hardware noise suppression. This mainly includes obstacle distance solution using an extended Kalman filter, complementary filter attitude fusion, and error modeling compensation under environmental factors. The extended Kalman filter is used to estimate the state of obstacle distance and velocity data collected by the millimeter-wave radar and ultrasonic array. The system state equation is: ; The observation equation is: ; in, is the system state vector at time k, which contains the relative position and velocity information of the obstacle, is the control input, is the process noise, is the observed quantity, is the observation noise, and They are the nonlinear motion model and observation model of the system, respectively. The state prediction and update are achieved by linearizing the Jacobian matrix, thereby dynamically suppressing the impact of sensor measurement noise on obstacle detection accuracy. For the IMU signal, complementary filtering is used to fuse the outputs of the accelerometer and gyroscope. The basic formula of complementary filtering is: ; in, is the fused attitude angle, is the attitude angle at the previous moment, is the current angular velocity, is the sampling period, is the attitude angle estimated by the accelerometer, It is the filter gain factor, which gives different weights to the gyroscope integral and accelerometer observation, so as to achieve attitude estimation with high dynamic response and long-term stability.

[0048] Taking into account the impact of environmental changes on sensor output characteristics, the system further established a temperature-error compensation mapping table and a light intensity-recognition accuracy correction model. The temperature compensation relationship is modeled as: ; in, is the error increment caused by temperature, is the temperature sensitivity coefficient, is the current temperature, The reference temperature is calibrated to adjust the sensor output value according to the ambient temperature in real time. The light intensity correction obtains the change curve of recognition accuracy under different lighting conditions through experimental calibration, and dynamically corrects the confidence weight in the recognition module to ensure the accuracy and consistency of obstacle detection and classification in low-light or strong reflective scenes.

[0049] The data asynchronous interference calibration module addresses the data timing misalignment problem caused by multimodal sensors at different sampling frequencies and communication delays. It designs a time synchronization mechanism accurate to the microsecond level and dynamically corrects the clock deviation of all sensors and edge computing units through a two-way timestamp calibration algorithm. The system adds an independent hardware-generated timestamp to each sensor data packet and records the sending time when sending. , the receiving end records the receiving time , combined with the return response time Reconfirm the sending time , estimate the clock deviation through the two-way time synchronization formula and propagation delay , its calculation expression is: ; ; in, is the clock difference between the sending and receiving ends, It is a one-way communication delay, based on which the timeline of each sensor data is continuously adjusted to a unified time base. To solve the problem of inconsistent arrival time of multi-source data, the system establishes a priority-based sensor data cache queue, sorts and caches data streams of different frequencies according to timestamps, and resamples and interpolates the misaligned data. The resampling target timestamp is t, and the two adjacent sampling times are known. and and corresponding data and , the resampling interpolation formula is: ; in, It is the estimated value at time t after resampling, ensuring the data integrity of all sensors at the same time point. To solve the problem of data loss or failure, the system introduces a Bayesian prediction model for extrapolation compensation. Given a historical observation sequence , predict the next missing data point The posterior probability is: ; in, is the probability distribution of the next data point inferred based on the existing observation sequence, is the likelihood function, is the prior distribution, P(X) is the marginal probability of the observed data, and the optimal compensation value is obtained by using the maximum a posteriori estimation, which effectively reduces the impact of sensor data failure on the overall perception decision chain.

[0050] In order to achieve the fusion processing of multi-sensor data in a unified spatial reference system, the system first performs offline spatial registration and obtains the external parameter parameters of each sensor through a high-precision calibration plate, including the position vector and translation matrix relative to the vehicle coordinate system, as well as the attitude angle and rotation matrix, and constructs the external parameter transformation matrix based on this. The point transformed to the vehicle's global coordinate system by the external parameter matrix The mathematical expression is: ; in, is the three-dimensional position vector in global coordinates, is a three-dimensional rotation matrix describing the rotation relationship, is the three-dimensional position vector in local coordinates, Is the translation vector from the sensor to the origin of the vehicle coordinate system. Rotation matrix The Euler angle or quaternion is used to describe the posture change of each sensor around a fixed axis, the translation vector Characterizes the fixed offset of each sensor in the vehicle coordinate system. During online operation, all real-time obstacle detection data undergoes rapid coordinate conversion based on the above external parameters, thereby unifying it to the same reference system and ensuring accurate spatial alignment of multi-sensor perception information. A multi-sensor field of view fusion algorithm is constructed based on coordinate consistency. For multiple detection point sets of the same obstacle observed by different sensors, a unified obstacle description is generated through the nearest neighbor distance criterion or weighted average fusion based on measurement noise covariance, further improving the spatial continuity and stability of detection.

[0051] In the data preprocessing and calibration module, in the preprocessing and model linkage unit, in order to achieve dynamic adaptation between sensor data quality and neural network model input, the system introduces a dynamic weight adjustment layer before the BP neural network input layer, assigning different weights according to the real-time confidence of each sensor to improve the stability and anti-interference ability of the overall prediction. Suppose the output of the i-th sensor at time t is , the corresponding confidence level is , then the weighted input after input layer adjustment The expression is: ; in, is the valid input value after confidence weighting, The value range is between 0 and 1, reflecting the health of the sensor status in real time, and is dynamically updated through the online anomaly detection and noise estimation module. In order to ensure the closed-loop consistency from the sensor to the actuator, the system establishes a residual monitoring mechanism to compare the expected execution parameters output by the BP neural network in real time. Feedback value from actuator , the residual is defined as: ; in, is the absolute residual value at the current moment, which is used to quantify the deviation between the model prediction and the actual execution. When the threshold exceeds 5%, the adaptive adjustment mechanism of the sensor preprocessing parameters is triggered. The adaptive adjustment process is based on minimizing the residual objective and optimizing the current sensor input weight by gradient descent. The optimization direction is determined by the gradient of the residual to the weight. The basic update formula is: ; in, is the learning rate coefficient, which controls the step size of weight update. It is the partial derivative of the residual with respect to the sensor weight, guiding the weight adjustment in the direction of reducing the prediction error, thereby making the input data quality and the neural network mapping capability self-consistent in real time, and continuously optimizing the response accuracy and stability of the obstacle avoidance system in complex dynamic environments.

[0052] like Figure 2As shown, the multi-source data fusion processing of the perception module is based on the combined output of millimeter-wave radar, ultrasonic array, and IMU. First, the data from different sensors are strictly time-aligned and spatially aligned to ensure that all data are aligned under a unified time reference and spatial coordinate system. The millimeter-wave radar mainly provides distance and radial velocity information of obstacles within a range of 30 meters, the ultrasonic array provides obstacle position supplementation with a near-field accuracy of ±3 cm, and the IMU provides the body roll angle and yaw angular velocity, which comprehensively describe the current motion state of the bicycle. Data fusion uses the extended Kalman filter algorithm to estimate the state of the obstacle, modeling the obstacle position, velocity, and acceleration as a state vector: , in represents the position of the obstacle in the global coordinate system, represents the velocity component, and the state transfer equation is: ; in, is the state transition matrix, is the control input matrix, is the control quantity, is process noise, reflecting modeling uncertainty. Millimeter wave radar and ultrasonic array provide observation data , updated through the observation model, the observation equation is expressed as: ; in, is the observation matrix, It is the observation noise. During the fusion process, the angular velocity and acceleration information of the vehicle body measured by the IMU are combined to perform motion compensation on the state prediction, thereby improving the accuracy of obstacle trajectory estimation under the dynamic posture change of the vehicle. Finally, based on the relative speed, relative distance and predicted trajectory of the obstacle, the threat level matrix T is constructed, whose elements are It represents the threat level of the i-th obstacle at the j-th time step, quantifies the potential impact of the obstacle on vehicle driving safety, and supports subsequent intelligent obstacle avoidance decisions.

[0053] In this embodiment, the road friction coefficient is calculated jointly by a pressure sensor and a speed sensor installed on the wheel, or by a visual sensor to identify road texture features and match them with a pre-stored friction coefficient database; when the IMU detects a roll angle greater than 3°, a crosswind interference warning signal is generated and transmitted to the edge computing unit.

[0054] In this embodiment, the dynamic mapping control strategy of the multi-actuator collaborative control module includes: At speeds exceeding 20 km / h, the edge computing unit outputs a 45° deployment command for the aerodynamic flaps, generating 30N of lateral thrust through a 0.8MPa high-pressure air pump. Simultaneously, the differential clutch controls the right wheel torque to 70%, with a 1:3 torque distribution ratio. The gyroscopic flywheel speed is increased to 23,000 rpm to suppress roll, enabling emergency obstacle avoidance with a 1.2-1.5m lateral offset within 200ms. At low speeds below 10 km / h, the pneumatic flaps are fine-tuned by 10°±0.5° via a servo motor, and the liquid metal counterweight unit achieves a ±5cm precision displacement to achieve a 5-10cm level of path correction. The differential clutch switches to balancing mode, with a torque distribution ratio of 1:1.2, and brake kinetic energy recovery and piezoelectric ceramic energy recovery are activated to ensure energy consumption remains at the 5W level. When the road friction coefficient μ is detected to be less than 0.3, the edge computing unit forces the differential clutch to be locked to a 1:1 torque distribution ratio and preloads the liquid metal counterweight to within ±2mm of the center position of the vehicle body.

[0055] The edge computing unit has a built-in vehicle speed-road condition-execution parameter dynamic mapping model, which uses BP neural network to train historical driving data and establish a complex nonlinear relationship between input features and control instructions. The model input vector is set to , where v represents the vehicle speed, d represents the obstacle distance, represents the road friction coefficient; the output vector is set to ,in represents the expansion angle of the aerodynamic wing, r represents the torque distribution ratio of the differential clutch, Represents the displacement of the liquid metal counterweight. The basic mapping relationship of the BP neural network is: ; in, and are the weight matrices from the input layer to the hidden layer and from the hidden layer to the output layer, respectively. and is the bias vector, σ is the activation function, and f is the output layer activation function, which realizes the nonlinear transformation from input features to control instructions. Model training optimizes parameters by minimizing the mean square error loss function, which is defined as: ; Where N represents the number of samples, is the real control output, is the network prediction output, To achieve the goal of less than 200 milliseconds for obstacle avoidance response in all scenarios and a 40% reduction in overall power consumption, the system introduces an adaptive learning mechanism during online operation. When the sensor detects changes in environmental characteristics, it dynamically adjusts the network's internal weights to maintain real-time output response and optimize energy consumption. Adaptive updates are performed according to the following formula: ; Where W represents the current weight set of the network, η is the adaptive learning rate, It represents the gradient direction of the loss function with respect to the weights. By continuously fine-tuning the network parameters, the mapping relationship can converge quickly under different vehicle speeds, obstacle distributions, and road conditions, ensuring that the obstacle avoidance control system always maintains efficient response and low energy consumption in a changing environment.

[0056] In this embodiment, the aerodynamic wing unit adopts an aerodynamically optimized design, and a pressure sensor is set on the wing surface to provide real-time feedback of the aerodynamic load. The edge computing unit dynamically adjusts the wing angle according to the feedback data to form a closed-loop control; the liquid metal counterweight unit uses gallium indium tin alloy as the counterweight medium, and is driven by an electromagnetic coil array to achieve mechanical contactless displacement, with a displacement accuracy of ±1mm.

[0057] In this embodiment, the dual CAN bus redundant architecture of the communication module includes a master bus and a slave bus, both of which operate independently and verify data consistency in real time. When the master bus fails, the slave bus automatically switches to the master control mode to ensure the reliable transmission of system control instructions; the supercapacitor unit capacity of the power supply module is not less than 120F, which supports maintaining stable operation of the core module for 60s under a discharge current of 5A.

[0058] The differential clutch unit integrates a torque sensor and a speed sensor to provide real-time feedback on the difference in torque and speed between the left and right wheels. The edge computing unit dynamically adjusts the torque distribution ratio based on the feedback signal. In energy recovery mode, the mechanical vibration energy during braking is converted into electrical energy through piezoelectric ceramics. Combined with the braking kinetic energy recovery device, the overall energy recovery efficiency is ≥35%.

[0059] Specifically, in this embodiment, the hardware integration of the bicycle intelligent obstacle avoidance system with multimodal sensing and active pneumatic control includes: The perception module is installed, and the millimeter-wave radar is installed in front of the front of the bicycle. It is fixed with a special bracket to ensure that its detection field of view is not blocked and can accurately detect the distance and speed of obstacles within the range of 30m±0.1m in front. The ultrasonic array is evenly distributed on the front, rear, left and right sides of the bicycle frame. It adopts an embedded installation method to make it flush with the surface of the frame, which not only ensures the appearance, but also can effectively obtain the position information of near-field obstacles with an accuracy of ±3cm. The IMU is installed in the center of the bicycle, usually near the center axis of the frame. It is connected to the frame through shock-absorbing pads to reduce the impact of body vibration and road impact, and can accurately detect the roll angle and yaw angular velocity of the bicycle.

[0060] The edge computing unit is a compact, powerful module installed under the handlebars or in a hidden location on the bicycle frame. Protected by a waterproof and dustproof casing, it ensures stable operation in various environments. It connects to the perception module, communication module, and multi-actuator collaborative control module via high-speed data cables, enabling rapid data transmission and processing.

[0061] The multi-actuator collaborative control module is installed, and the pneumatic vane units are installed on both sides of the bicycle and fixed by adjustable brackets, which can adjust the angle and position of the vanes as needed. The high-pressure air pump and servo motor are installed at the bottom or rear of the frame and connected to the pneumatic vanes via air pipes and control lines to ensure the stability and reliability of the drive. The differential clutch unit is integrated into the rear wheel axle of the bicycle and is closely integrated with the rear wheel transmission system. It can quickly adjust the torque distribution between the left and right wheels according to the instructions of the edge computing unit. The liquid metal counterweight unit is installed inside the bicycle frame and is driven by an electromagnetic coil array to achieve mechanical contactless displacement with a displacement accuracy of ±1mm. The gyro flywheel unit is installed near the center axis of the bicycle. It is made of carbon fiber and has an adjustable speed within the range of 20,000-23,000 rpm. The angular momentum generated by its high-speed rotation is used to suppress the bicycle's roll. The communication module is installed using a dual CAN bus redundant architecture, with the master and slave bus communication lines routed along the internal wiring troughs of the vehicle frame to prevent external interference. The communication module's interfaces are connected to the perception module, edge computing unit, and multi-actuator collaborative control module, ensuring real-time data exchange. The power supply module, typically powered by a lithium battery, is installed in a dedicated battery compartment at the bottom or rear of the bicycle frame for easy replacement and charging. The supercapacitor emergency power supply unit is installed near the main power source and connected to it via a fast-switching circuit. This allows for rapid power support for the system's core modules in the event of a main power failure. The data preprocessing and calibration module is integrated with the edge computing unit and installed under the handlebars or in a hidden location on the frame. It transmits data to the perception module and edge computing unit via an internal data bus. Its internal sensor noise layering suppression module, data asynchronous interference calibration module, and preprocessing and model linkage module work together through circuit connections and software algorithms to ensure the accuracy and reliability of data collected by the perception module. Software implementation of a bicycle intelligent obstacle avoidance system based on multimodal sensing and active pneumatic control: The perception module's software functions primarily control the operating status of the millimeter-wave radar, ultrasonic array, and IMU, including sensor initialization and data acquisition frequency settings. It collects sensor data in real time and performs preliminary processing and screening to remove obviously erroneous or invalid data. The processed data is then transmitted to the edge computing unit via the communication module. During data acquisition, the software also dynamically adjusts sensor parameters based on the sensor's operating status and environmental conditions to improve data accuracy and stability.

[0062] The software functions of the edge computing unit are the core of the entire system, implementing multi-source data fusion processing, obstacle threat level calculation, and control command generation. Regarding multi-source data fusion processing, the software first performs temporal and spatial registration of the millimeter-wave radar, ultrasonic array, and IMU data transmitted from the perception module to ensure temporal and spatial consistency. It then uses an extended Kalman filter algorithm to estimate the obstacle state. Combining the observation data provided by the millimeter-wave radar and ultrasonic array with the vehicle body angular velocity and acceleration information measured by the IMU, it performs motion compensation on the state prediction, improving the accuracy of obstacle trajectory estimation under dynamic vehicle posture changes. Based on the relative speed, relative distance, and predicted trajectory of the obstacle, a threat level matrix T is constructed to quantify the potential impact of the obstacle on vehicle driving safety. Based on information such as vehicle speed, obstacle distance, road friction coefficient, and vehicle posture parameters, the software utilizes a built-in dynamic mapping model between vehicle speed, road condition, and execution parameters to dynamically generate control commands for the aerodynamic wing deployment angle, differential clutch torque distribution ratio, liquid metal counterweight displacement, and gyroscopic flywheel speed. During operation, the software will also dynamically adjust the internal weights of the vehicle speed-road condition-execution parameter dynamic mapping model through an adaptive learning mechanism based on changes in environmental characteristics detected by sensors, so that the system's obstacle avoidance response time is less than 200 milliseconds in all scenarios and the overall power consumption is reduced by 40%. The software functions of the multi-actuator collaborative control module are responsible for receiving control commands from the edge computing unit and converting them into specific control signals to control the operation of the pneumatic vane unit, differential clutch unit, liquid metal counterweight unit, and gyro flywheel unit. It monitors the operating status of each actuator in real time, including the deployment angle of the pneumatic vane, the torque distribution ratio of the differential clutch, the displacement of the liquid metal counterweight, and the speed of the gyro flywheel, and feeds this status information back to the edge computing unit. After receiving the control command, the software will optimize and adjust the control signal based on the current state of the actuator and the command requirements, ensuring that the actuator can execute the command accurately and quickly. The communication module software functions as follows: The software of the communication module is responsible for managing the communication process of the dual CAN bus redundant architecture, including data transmission, reception, verification and error handling. It ensures real-time and accurate transmission of data between the perception module, edge computing unit and multi-actuator collaborative control module. During the data transmission process, the software will package and encode the data, add information such as check bits and timestamps to improve the reliability of data transmission. During the data reception process, the software will unpack, verify and parse the received data to ensure the integrity and correctness of the data. When the main bus fails, the software can automatically switch from the bus to the master control mode to ensure the normal operation of the system. The power supply module software is primarily responsible for monitoring the power, voltage, current, and other parameters of the main power supply and supercapacitor emergency power supply unit, providing real-time feedback to the edge computing unit. When the main power supply is operating normally, the software controls the supercapacitor to charge and ensure it is fully charged. If the main power supply fails, the software can quickly switch to supercapacitor power supply mode and control the supercapacitor to power the system's core modules with stable power, while also issuing a power failure alarm signal. The data preprocessing and calibration module software implements functions such as sensor noise layer suppression, data asynchronous interference calibration, and preprocessing and model integration. To suppress sensor noise layer suppression, the software, based on hardware optimization, uses algorithms such as median filtering and sliding average filtering to process data collected by millimeter-wave radar, ultrasonic array, and IMU to remove noise interference. A sensor outlier detection mechanism is established. Based on the triple standard deviation principle, an abnormality threshold range is defined and data is monitored in real time to see if it exceeds this range. Once exceeded, the data sample is marked as an abnormal sample and triggers data compensation or sensor status assessment procedures. To compensate for data asynchronous interference, the software designs a microsecond-accurate time synchronization mechanism. A bidirectional timestamp calibration algorithm dynamically corrects clock offsets between all sensors and the edge computing unit. A priority-based sensor data cache queue is established to resample and interpolate misaligned data, and a Bayesian prediction model is introduced for extrapolation compensation. To integrate preprocessing and model integration, the software introduces a dynamic weight adjustment layer before the BP neural network input layer to assign weights based on the real-time confidence level of the sensors. A residual monitoring mechanism is established to compare the expected execution parameters output by the BP neural network with the actuator feedback value in real time. When the residual exceeds the set threshold, the adaptive adjustment mechanism of the sensor preprocessing parameters is triggered. In one embodiment, a bicycle intelligent obstacle avoidance system with multi-modal sensing and active pneumatic control was tested and verified.

[0063] This involves setting up a test environment that simulates various riding scenarios, including urban road scenarios with pedestrians, vehicles, traffic lights, and roadside obstacles; mountain road scenarios with undulating terrain, curves, and potholes; and adverse weather scenarios such as simulated rain, fog, and high winds. High-precision measurement equipment, such as lidar and high-speed cameras, is deployed in the test area to record parameters such as the bicycle's trajectory, speed, and posture, as well as the location and motion of obstacles, for comparison and verification with the detection data of the obstacle avoidance system.

[0064] The test index is set, and multiple key test indexes are set to evaluate the system performance. The obstacle avoidance success rate is the ratio of the number of times that the test system successfully avoids obstacles in different scenes to the total number of tests, and the requirement is that the obstacle avoidance success rate is not less than 95% in various complex scenes. The obstacle avoidance response time is the time interval from the system detecting the obstacle to executing the obstacle avoidance action, and the requirement is that the obstacle avoidance response time is less than 200 milliseconds in all scenes. The displacement accuracy is for the liquid metal weight unit, and the deviation between the actual displacement and the command displacement under different temperatures and working conditions is tested, and the requirement is that the weight displacement accuracy under extreme temperature is maintained within ±0.3mm. Energy consumption, test the energy consumption of the system under different speeds and working conditions, and the requirement is that the comprehensive power consumption is reduced by more than 40%. Stability, observe the working state of the system under long-time operation and complex environment, and ensure that the fault-free operation time of the system is not less than 100 hours.

[0065] Test result analysis, in the urban road scene test, the system can accurately detect pedestrians, vehicles and other obstacles, and quickly generate reasonable obstacle avoidance strategies according to the vehicle speed and obstacle distance, with an obstacle avoidance success rate of 98% and an average obstacle avoidance response time of 150 milliseconds. In the mountain road scene test, the system can adapt to the change of terrain, and successfully avoid obstacles such as curves and potholes through the coordinated work of multiple actuators, with an obstacle avoidance success rate of 96% and an average obstacle avoidance response time of 180 milliseconds. In the severe weather scene test, the sensors of the system can work stably under rain, fog, wind and other conditions, accurately detect obstacles, and have an obstacle avoidance success rate of 95% and an average obstacle avoidance response time of 170 milliseconds. For displacement accuracy test, under extreme temperature scene, through double-stage compensation of hardware and algorithm, the displacement accuracy of liquid metal weight unit can be stabilized within ±0.3mm, meeting the design requirements. In the energy consumption test, the comprehensive power consumption of the system under different speeds and working conditions is reduced by 45% compared with the traditional system, reaching the expected target. In the stability test, the system has a fault-free operation time of more than 100 hours under long-time operation and complex environment, showing good stability. Through the analysis of the test results, it is proved that the bicycle intelligent obstacle avoidance system based on multi-modal sensing and active pneumatic control of the present application has high effectiveness and reliability in various complex environments, and can significantly improve the safety and stability of bicycle riding.

[0066] In another embodiment, the test and verification of the bicycle intelligent obstacle avoidance system based on multi-modal sensing and active pneumatic control are carried out.

[0067] In the perception module, the millimeter-wave radar is mounted at the center of the bicycle's front, ensuring an unobstructed field of view and clearly scanning for obstacles within a 30-meter radius. The ultrasonic array is evenly distributed around the front of the bicycle, forming an all-around near-field detection area to accurately acquire the position of near-field obstacles. The inertial measurement unit (IMU) is fixed to the center of the bicycle frame, where it can most accurately detect the bicycle's roll angle and yaw rate, reducing measurement errors caused by improper mounting position.

[0068] The edge computing unit is installed near the handlebars of the bicycle, facilitating wiring and connection to other modules, as well as facilitating subsequent maintenance and upgrades. Each unit of the multi-actuator collaborative control module is installed in its corresponding position. The aerodynamic wing units are installed on both sides of the bicycle. Based on aerodynamic principles, they can effectively generate lateral thrust or attitude compensation force when deployed. The differential clutch unit is installed in the bicycle's transmission system to ensure accurate control of torque distribution between the left and right wheels. The liquid metal counterweight unit is installed at the bottom of the frame, close to the bicycle's center of gravity, to achieve precise counterweight displacement through electromagnetic drive and adjust the bicycle's center of gravity distribution. The gyro flywheel unit is installed near the bicycle's center axis to ensure that it can effectively suppress the bicycle's roll when rotating at high speeds.

[0069] The communication module's dual CAN bus lines are routed along the bicycle's frame, ensuring the shortest possible communication paths between modules and minimizing signal transmission interference and latency. The power supply module is installed within the bicycle frame, utilizing the available space for optimal layout. The supercapacitor emergency power supply unit is tightly connected to the main power source, ensuring rapid switching in the event of a main power failure, maintaining core functionality.

[0070] The debugging step is crucial. After hardware installation, the first step is to check hardware connections to ensure that cables between modules are secure and that there are no loose connectors. Next, perform sensor calibration, using specialized calibration equipment and software to calibrate the millimeter-wave radar, ultrasonic array, and inertial measurement unit (IMU) to ensure accurate and reliable sensor data output. For example, standard obstacles are set up in a specific test site to calibrate distance measurements and adjust their measurement accuracy. For the IMU, standard attitude tests are performed in the horizontal and vertical directions to calibrate its roll angle and yaw rate measurements. Next, perform communication testing using a communication test tool to verify the proper communication of the dual CAN buses and the accurate and timely data transmission between modules. Finally, perform a comprehensive system functionality test, simulating various driving scenarios, including high-speed and low-speed driving, and low-friction road conditions, to verify the system's obstacle avoidance capabilities and that each actuator accurately follows the instructions of the edge computing unit.

[0071] At the same time, workflows for different scenarios are set up, including verification of high-speed driving scenarios, low-speed driving scenarios, and low-friction road scenarios.

[0072] During high-speed driving scenario verification, the system quickly entered efficient obstacle avoidance mode at speeds exceeding 20 km / h. The millimeter-wave radar in the perception module, leveraging its long-range detection capabilities, continuously scans for obstacles within a 30-meter radius, acquiring real-time distance and speed information. The ultrasonic array provides additional assistance, providing more precise position monitoring of near-field obstacles to ensure no potential hazards are missed. The inertial measurement unit (IMU) continuously monitors the bicycle's roll angle and yaw rate, providing the system with real-time posture data.

[0073] This multi-source data is rapidly transmitted to the edge computing unit, which utilizes an ARM Cortex-M7 and NPU heterogeneous computing architecture for efficient data fusion and processing. First, the extended Kalman filter algorithm is used to estimate the obstacle's state. Combining data from the millimeter-wave radar, ultrasonic array, and IMU, it accurately calculates the obstacle's position, velocity, acceleration, and other state information. Then, based on this information, a threat level matrix T is constructed to quantify the potential impact of the obstacle on bicycle safety. If an obstacle with a high threat level is detected—that is, one that is close and moving quickly—the edge computing unit quickly generates control instructions.

[0074] Upon receiving the command, the multi-actuator coordinated control module immediately coordinates its actions. Driven by a 0.8MPa high-pressure air pump, the pneumatic flap unit rapidly deploys to a 45° angle, generating 30N of lateral thrust, propelling the bicycle to one side and avoiding obstacles. The differential clutch unit adjusts the right wheel's torque contribution to 70%, achieving a 1:3 torque split. This allows for better steering and, combined with the thrust of the aerodynamic flaps, enables fast and precise obstacle avoidance. The gyroscopic flywheel unit accelerates to 23,000 rpm, leveraging the angular momentum generated by its high-speed rotation to effectively suppress roll during obstacle avoidance and ensure stable operation. The coordinated action of these actuators enables the bicycle to achieve a lateral deviation of 1.2-1.5m in 200ms, successfully avoiding obstacles and ensuring safe riding.

[0075] During low-speed driving scenario verification, when the vehicle speed drops below 10 km / h, the system enters low-speed driving mode, prioritizing path correction accuracy and energy consumption control. The perception module continues to collect data in real time, with the millimeter-wave radar and ultrasonic array closely monitoring the position of obstacles ahead, and the IMU continuously monitoring the bicycle's posture.

[0076] The edge computing unit integrates this data and determines whether path corrections are necessary based on information such as vehicle speed and obstacle distance. If an obstacle is detected and needs to be avoided, the edge computing unit generates corresponding control instructions. The pneumatic flap unit, driven by a servo motor, makes fine adjustments of 10°±0.5°, generating a small lateral force to fine-tune the bicycle's direction. The liquid metal counterweight unit, driven by an electromagnetic coil array, achieves precise displacement of ±5cm, adjusting the bicycle's center of gravity. This, in conjunction with the movement of the pneumatic flaps, enables path corrections of 5-10cm. The differential clutch unit switches to balancing mode, adjusting the torque distribution ratio to 1:1.2, ensuring more even torque distribution between the left and right wheels and smooth riding. Simultaneously, the system activates brake kinetic energy recovery and piezoelectric ceramic energy recovery, converting and storing mechanical energy during braking into electrical energy, ensuring energy consumption remains within 5W and achieving low-energy operation.

[0077] In a low-friction road scenario, when the road friction coefficient μ is detected to be less than 0.3, the system immediately takes action to ensure the bicycle's stability on the low-friction road. The pressure sensor, speed sensor, or visual sensor in the perception module monitors road conditions in real time and transmits road friction coefficient data to the edge computing unit.

[0078] The edge computing unit rapidly adjusts the operating state of the multi-actuator coordinated control module based on changes in the road friction coefficient. The differential clutch unit is locked to a 1:1 torque distribution ratio, ensuring equal torque distribution between the left and right wheels and preventing wheel slip due to uneven torque distribution. The liquid metal counterweight unit preloads the weight within ±2mm of the center of the vehicle, lowering the bicycle's center of gravity and increasing stability. These adjustments enable the bicycle to maintain stable operation on low-friction surfaces, reducing the risk of loss of control due to slippery conditions and ensuring rider safety.

[0079] Therefore, the multimodal sensing and active pneumatic control bicycle intelligent obstacle avoidance system provided by this application has the advantages of significantly improving displacement accuracy, optimizing response speed, and enhancing scene adaptability, and this is achieved through multi-dimensional technical means. In terms of improving displacement accuracy, a three-modal sensor array is used to monitor temperature, magnetic field, and deformation in real time, combined with a micro-temperature control system to maintain the liquid metal temperature at 25±0.5°C. The displacement deviation is dynamically corrected through a multi-physics field model and an improved Kalman filter, which improves the displacement accuracy from ±1mm to within ±0.3mm at extreme temperatures.

[0080] In terms of optimizing response speed, a two-stage compensation mechanism is constructed. The first-stage compensation quickly adjusts the driving current, and the second-stage compensation starts the temperature control unit. The improved BP neural network is used to predict temperature trends and correct driving parameters in advance. The response delay during temperature changes is shortened to within 200ms, effectively avoiding the risk of secondary collisions.

[0081] In terms of enhancing scenario adaptability, the collaborative work of hardware composite sensing and algorithm multi-physics field modeling, combined with a four-element closed-loop control chain, realizes dynamic feedback of temperature, magnetic field, displacement and current, enabling the system to operate stably in the temperature range of -20℃~80℃, fully covering extreme scenarios such as summer rainstorms and winter indoor conditions.

[0082] In addition, the multimodal sensing and active pneumatic control bicycle intelligent obstacle avoidance system provided by this application achieves higher detection accuracy in obstacle detection through the integration of multimodal sensing technologies. The data fusion of millimeter-wave radar and ultrasonic array improves the obstacle distance solution accuracy from ±3cm (ultrasonic) + ±0.1m (radar) before fusion to a unified accuracy of ±2cm across the entire range, which can more accurately detect the location of obstacles and provide more reliable data support for obstacle avoidance decisions. In a complex urban street environment, the system can accurately detect a trash can with a diameter of 0.5m 20m ahead. Traditional obstacle avoidance systems based on a single sensor may have large detection errors and cannot accurately identify the location and distance of the obstacle.

[0083] Obstacle avoidance response time is a key performance indicator for obstacle avoidance systems. This system's edge computing unit utilizes an ARM Cortex-M7 and NPU heterogeneous computing architecture, combined with efficient data processing algorithms and dynamic mapping models, to achieve an obstacle avoidance response time of less than 200ms in all scenarios. In actual testing, when a bicycle traveling at 25 km / h suddenly encountered an obstacle 5 meters ahead, the system was able to react and initiate obstacle avoidance within 150ms. In contrast, some existing obstacle avoidance systems may have a response time exceeding 300ms, resulting in delayed obstacle avoidance and an increased risk of collision.

[0084] In terms of control precision, the design of a multi-actuator coordinated control module enables more precise control. The pneumatic flap unit is driven by a high-pressure air pump and a servo motor, achieving a continuously adjustable deployment angle of 0-45°, with an angle control accuracy of ±0.5°. The liquid metal counterweight unit is electromagnetically driven, achieving a counterweight displacement accuracy of ±15cm±1mm. The gyroscopic flywheel unit's speed is adjustable between 20,000 and 23,000 rpm, with a speed control accuracy of ±100 rpm. These high-precision control capabilities enable the bicycle to more stably and accurately adjust its attitude and direction during obstacle avoidance, effectively preventing obstacles such as obstacle avoidance failure or vehicle loss of control due to insufficient control accuracy. In high-speed obstacle avoidance scenarios, the system can precisely control the aerodynamic flaps to 45°, generating 30N of lateral thrust, while simultaneously increasing the gyroscopic flywheel speed to 23,000 rpm, enabling the bicycle to quickly and stably avoid obstacles. Traditional systems may not be able to control the actuators with such precision, resulting in poor obstacle avoidance results.

[0085] The system of the present invention has demonstrated excellent stability and reliability in complex environments. Under adverse weather conditions, such as rainy and foggy days, the sensor noise layering suppression strategy and the data asynchronous interference calibration module play an important role. On rainy days, millimeter-wave radars are easily interfered with by raindrops, resulting in increased ranging errors and the detection accuracy of the ultrasonic array will also be affected. This system reduces the interference of raindrops on the signal by integrating an anti-electromagnetic interference shield for the millimeter-wave radar and the ultrasonic array; at the same time, algorithms such as median filtering and extended Kalman filtering are used to process sensor data, effectively improving the accuracy and stability of the data. Experimental data show that in rainy environments, the detection accuracy of this system for obstacles can still be maintained at above 95%, while the detection accuracy of traditional obstacle avoidance systems may drop below 70%.

[0086] When driving on bumpy roads, the roll angle and yaw rate detected by the IMU are easily affected by mechanical vibration noise, resulting in zero-drift errors accumulating over time. This system secures the IMU with a rubber shock-absorbing mount and incorporates a hardware low-pass filter to attenuate high-frequency vibration noise. It also processes the IMU data using algorithms such as sliding average filtering and complementary filtering, effectively reducing noise interference and improving attitude detection accuracy. During testing on a bumpy road, this system accurately detected changes in the bicycle's roll angle and yaw rate, and promptly adjusted the actuators to maintain stable driving. Traditional systems, however, can cause the bicycle to lose balance due to inaccurate attitude detection.

[0087] Through a series of technical measures, this system effectively reduces the frequency of actuator misoperation. Under adverse conditions such as rain, fog, and bumps, the frequency drops from 0.3 times per kilometer to 0.05 times per kilometer. This not only improves system reliability but also ensures rider safety. During actual riding, the system accurately determines the threat level of obstacles and precisely controls actuator action based on the actual situation, avoiding unnecessary dangers caused by misoperation.

[0088] Although preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they are aware of the basic creative concepts. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the embodiments of the present invention. Finally, it should be noted that, in this document, relational terms such as first and second, etc., are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or terminal device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article, or terminal device. In the absence of further restrictions, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or terminal device comprising the element.

[0089] The above is a detailed introduction to the multimodal sensing and active pneumatic control bicycle intelligent obstacle avoidance system provided in the embodiments of the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the technical solutions and core ideas of the present application. Ordinary technicians in this field should understand that they can still modify the technical solutions recorded in the aforementioned embodiments, or replace some of the technical features therein with equivalents; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A bicycle intelligent obstacle avoidance system with multi-modal sensing and active pneumatic control, characterized in that: It includes a perception module, an edge computing module, a multi-actuator collaborative control module, a communication module, a power supply module, and a data preprocessing and calibration module; The sensing module is used to collect real-time data on the bicycle's driving status and external environment. The sensing module includes at least a millimeter-wave radar, an ultrasonic array, and an inertial measurement unit. The millimeter-wave radar is used to detect the distance and speed of obstacles within a range of 30m±0.1m; the ultrasonic array is used to obtain the position information of near-field obstacles with an accuracy of ±3cm; and the inertial measurement unit is used to detect the bicycle's roll angle and yaw angular velocity. The edge computing unit uses an ARM Cortex-M7 and NPU heterogeneous computing architecture to fuse and process the multi-source data output by the perception module, calculate the obstacle threat level in real time, and dynamically generate control instructions for the multi-actuator collaborative control module based on vehicle speed, obstacle distance, road friction coefficient, and vehicle body posture parameters; The multi-actuator coordinated control module includes a pneumatic vane unit, a differential clutch unit, a liquid metal counterweight unit, and a gyro flywheel unit. The pneumatic vane unit is driven by a high-pressure air pump and a servo motor to achieve a continuously adjustable deployment angle of 0-45° for generating lateral thrust or attitude compensation force. The differential clutch unit supports a dynamically adjustable torque distribution ratio of 1:1 to 1:

3. The liquid metal counterweight unit is electromagnetically driven by an electromagnetic coil to achieve a counterweight displacement with an accuracy of ±15cm±1mm. The gyro flywheel unit is made of carbon fiber and has an adjustable speed within the range of 20,000-23,000 rpm. Among them, a three-modal sensor array is embedded in the liquid metal counterweight unit, including a platinum resistance temperature sensor PT1000, a micro Hall magnetic field sensor and a fiber Bragg grating strain sensor, forming a composite sensing structure and achieving a high degree of coupling with the temperature control unit; the platinum resistance temperature sensor PT1000 and the micro Hall magnetic field sensor can monitor the working temperature of the liquid metal and the magnetic field strength of the electromagnetic coil in real time; the sensing data of the platinum resistance temperature sensor PT1000, the micro Hall magnetic field sensor and the fiber Bragg grating strain sensor are connected to the edge computing unit through a high-speed bus through a synchronous clock acquisition interface, and the sampled data are fused into the temperature-magnetic field-deformation three-dimensional state matrix to update the operating state vector of the liquid metal counterweight unit in real time; The communication module adopts a dual CAN bus redundant architecture to achieve real-time data interaction between the perception module, the edge computing unit and the multi-actuator collaborative control module; The power supply module includes a supercapacitor emergency power supply unit, which can maintain the core functions for 60 seconds when the main power supply fails; The data preprocessing and calibration module is used to solve the problem of asynchronous interference between sensor noise and data. The data preprocessing and calibration module at least includes a sensor noise layer suppression unit, a preprocessing and model linkage unit, and a temperature-magnetic field coupling compensation unit; Among them, by implanting semiconductor refrigeration sheets and flexible heating films in the temperature-magnetic field coupling compensation unit, a micro-scale temperature control unit with an accuracy of ±0.5°C is constructed, which is used to precisely adjust the working temperature of the liquid metal counterweight medium to suppress the displacement error and control hysteresis caused by thermal expansion and contraction effects; and an algorithm is used to establish a multi-physics field mapping model based on the magnetohydrodynamic equation; when the edge computing unit detects that the temperature fluctuation exceeds ±1°C, a two-stage compensation mechanism is automatically triggered: the first-stage compensation adjusts the driving current through the PID algorithm (compensation coefficient 0.8A / °C) to maintain the stability of the magnetic field; the second-stage compensation starts the temperature control unit and activates the newly added temperature-current mapping hidden layer (16 neurons) of the BP neural network, and optimizes the electromagnetic coil drive parameters in real time through the dual CAN bus, ultimately maintaining the counterweight displacement accuracy within ±0.3mm under extreme temperatures, ensuring the dynamic consistency of the body roll compensation during high-speed obstacle avoidance.

2. The multi-modal sensing and active pneumatic control bicycle intelligent obstacle avoidance system according to claim 1 is characterized in that: The liquid metal is a gallium-indium-tin alloy.

3. The multi-modal sensing and active pneumatic control bicycle intelligent obstacle avoidance system according to claim 1 is characterized in that: In the liquid metal counterweight unit, the platinum resistance temperature sensor PT1000 is mounted on the high thermal conductivity path of the counterweight cavity wall through precision positioning tooling to ensure the thermal conductivity efficiency between the platinum resistance temperature sensor PT1000 and the liquid metal alloy contact interface. A sampling frequency of 100 Hz is used to obtain real-time temperature data inside the liquid metal counterweight cavity. Through temperature readings with an accuracy of ±0.05°C, the thermal field distribution of the liquid metal alloy is accurately portrayed, providing real-time feedback for the temperature control loop.

4. The multi-modal sensing and active pneumatic control bicycle intelligent obstacle avoidance system according to claim 1, characterized in that: The micro-Hall sensors are arranged at both ends of the liquid metal flow path in the liquid metal counterweight unit. The micro-Hall sensors monitor the changes in local magnetic field intensity generated by the excitation of the electromagnetic drive coil. The 0.5mT resolution of the micro-Hall sensors can capture tiny magnetic field disturbances, and then judge the consistency between the driving magnetic flux distribution and the electromagnetic force field response, and assist in controlling the closed-loop regulation of the driving current of the electromagnetic coil.

5. The multi-modal sensing and active pneumatic control bicycle intelligent obstacle avoidance system according to claim 1, characterized in that: The fiber Bragg grating strain sensor is arranged in the high strain area of ​​the liquid metal counterweight cavity. The fiber Bragg grating strain sensor measures the elastic deformation of the structure of the liquid metal counterweight cavity under electromagnetic excitation in real time through the wavelength drift signal. The accuracy of the fiber Bragg grating strain sensor reaches 1με, which can effectively identify the changes in viscous resistance and wall friction characteristics of the alloy medium under different displacement conditions.

6. The multi-modal sensing and active pneumatic control bicycle intelligent obstacle avoidance system according to any one of claims 3 to 5, characterized in that: The semiconductor refrigeration chip achieves cooling function when powered on through the Peltier effect. The cold end of the semiconductor refrigeration chip is directly attached to the heat conduction enhanced area on the outer wall of the liquid metal counterweight cavity. The interfacial gap is filled with high thermal conductivity silicone grease to reduce thermal resistance. At the same time, the hot end of the semiconductor refrigeration chip is connected to the graphite heat conductive sheet and aluminum heat sink array for efficient heat dissipation. The flexible heating film is attached to the symmetrically distributed area outside the liquid metal counterweight cavity. The structure of the flexible heating film has low thermal inertia and planar heating characteristics, and can supplement heat in an instantaneous response to offset the effects of ambient cooling or overcooling of the refrigeration end.

7. The multi-modal sensing and active pneumatic control bicycle intelligent obstacle avoidance system according to claim 6, characterized in that: The micro-scale temperature control unit uses the cavity temperature value fed back in real time by the platinum resistance temperature sensor PT1000 as a control reference input, and compares it with a preset target temperature. The cooling current of the semiconductor refrigeration plate and the heating power of the flexible heating film are adjusted through a PID closed-loop control algorithm. The specific proportional coefficient is set to Kp equal to 2.5, the integral time is 100 milliseconds, and the differential time is 20 milliseconds, achieving a dynamic balance between rapid response to temperature changes and steady-state accuracy. In the initial heating stage, when the feedback temperature of the platinum resistance temperature sensor PT1000 is lower than the preset target temperature, the microscale temperature control unit outputs a PWM signal through the PID controller to drive the flexible heating film to gradually heat up. As the temperature approaches the preset target temperature, the PID controller gradually reduces the heating power according to the deviation change trend until the preset target temperature is reached; when the external ambient temperature rises or the electromagnetic excitation generates Joule heat causing the cavity to heat up, the system starts the semiconductor refrigeration plate, controls the cold end heat flow of the semiconductor refrigeration plate to be extracted from the inside of the liquid metal counterweight cavity, and discharged to the heat dissipation system through the hot end of the semiconductor refrigeration plate, and adjusts the cooling current amplitude in real time to maintain temperature stability.

8. The multi-modal sensing and active pneumatic control bicycle intelligent obstacle avoidance system according to claim 1, characterized in that: Based on the magnetohydrodynamic equation, a multi-physics field mapping model is established with electromagnetic driving force as the core driving quantity, and its expression is: ； in, It represents the total driving force per unit volume of liquid metal, is the current density vector, is the magnetic field intensity vector, ρ is the density of liquid metal, is the gravitational acceleration vector, μ is the dynamic viscosity of the liquid metal, is the flow velocity vector; The mapping model constructs a three-dimensional physical field based on the geometric structure of the liquid metal counterweight cavity, and forms the current density distribution by setting the current input boundary conditions of the electromagnetic coil. , and at the same time calibrate the magnetic field distribution using the measured data of the micro Hall magnetic field sensor , to ensure the spatial accuracy of magnetic field modeling; After the platinum resistance temperature sensor PT1000 outputs a real-time temperature signal, the mapping model dynamically corrects the density ρ and dynamic viscosity μ according to the thermophysical property response curve of the liquid metal material, capturing the impact of temperature changes on the fluid mechanics properties; in the edge computing unit, the real-time sensing data is input into a pre-built coupling model database or a fast solver is used to obtain the force field distribution and flow velocity solution at the current moment, thereby inferring the driving current, electromagnetic waveform and frequency modulation required for compensation; this compensation mechanism ultimately realizes the dynamic coupling closed-loop adjustment of the control signal and the liquid metal state, so that the counterweight displacement process has higher accuracy, response speed and environmental robustness, thereby significantly improving the stability and execution efficiency of the obstacle avoidance system.

9. The multi-modal sensing and active pneumatic control bicycle intelligent obstacle avoidance system according to claim 8, characterized in that: In the process of sensor data fusion processing, the temperature-magnetic field coupling compensation unit uses an improved extended Kalman filter to accurately estimate the state of multimodal signals, and constructs an enhanced state vector by introducing temperature and magnetic field state components. , where x, y, z represent the position of the liquid metal in the three-dimensional coordinate system, T represents the current temperature of the liquid metal cavity, and B represents the local magnetic field intensity under the excitation of the electromagnetic coil; Based on this state vector, the state transition equation of the extended Kalman filter is updated as: ； in, is the predicted state vector for the next moment, is the current state, F is the state transfer matrix, which represents the natural evolution law of the system without control input, G is the control matrix, is the control input, which represents the adjustment instruction of the electromagnetic drive parameters by the edge computing unit. is the process noise term, which is used to model disturbance factors and dynamic errors that are not explicitly considered; This state prediction mechanism embeds the viscosity drift caused by temperature changes and the nonlinear changes in the force field caused by magnetic field intensity fluctuations into the state update process, enabling the extended Kalman filter to actively perceive the impact of the physical environment on displacement behavior; During the observation phase, the temperature-magnetic field coupling compensation unit constructs an observation vector from the three-modal data collected from the platinum resistance temperature sensor PT1000, the micro Hall magnetic field sensor and the optical fiber strain sensor. , and its observation model is expressed as: ； Among them, H is the observation matrix, which defines the mapping relationship between the sensor observation value and the internal state vector. is the observation noise term, taking into account sensor accuracy and sampling error; The extended Kalman filter calculates the Kalman gain matrix based on the residual between the current predicted state and the actual observation , and updates the state estimate accordingly, thereby achieving fine compensation for the actual displacement state of the liquid metal counterweight; the real-time introduction of temperature and magnetic field states enables the extended Kalman filter to actively capture the impact of environmental disturbance factors on position prediction, thereby maintaining a high-precision estimate of the counterweight position under conditions of significant or rapid fluctuations in magnetic and thermal interference.

10. The multi-modal sensing and active pneumatic control bicycle intelligent obstacle avoidance system according to claim 1, characterized in that: In the bipolar compensation mechanism of the temperature-magnetic field coupling compensation unit, the first-level compensation is activated when the temperature fluctuation is detected to be between ±1 and ±2 degrees Celsius. Current compensation is performed only for the electromagnetic drive module. Based on the feedback temperature deviation ΔT, the temperature-magnetic field coupling compensation unit adjusts the driving current value of the electromagnetic coil according to a linear proportional coefficient of 0.8 amperes per degree Celsius. The compensation formula is: ； in is the adjusted driving current, is the current setting current, and ΔT is the temperature deviation. The first-level compensation strategy quickly responds to the fluctuation of magnetic field intensity caused by thermal drift, stabilizes the output of electromagnetic driving force, and thus maintains the accuracy of the liquid metal counterweight displacement trajectory; If the temperature deviation further expands to more than ±2 degrees Celsius, the secondary compensation mechanism is triggered, and the semiconductor refrigeration plate and the flexible heating film are activated at the same time. The PID temperature control strategy is used to restore the liquid metal working temperature to the set range within 100 milliseconds, and the predictive control capability is further enhanced.

Images