Full-working-condition self-balancing intelligent two-wheeled vehicle control system based on control moment gyroscope

Abstract

The invention relates to the technical field of two-wheeled vehicle control, and discloses a full-working-condition self-balancing intelligent two-wheeled vehicle control system based on a control moment gyroscope, and the system comprises a multi-source data center module which is used for converging data such as a satellite map and Internet of Vehicles crowdsourcing, generating a pavement friction coefficient distribution map, and transmitting the pavement friction coefficient distribution map as prior data. And the data analysis and decision module is used for receiving the prior data and the dynamic state of the vehicle, calculating a staged safety speed and a staged safety distance, evaluating risks based on reinforcement learning, and generating a staged intervention instruction. And the vehicle-mounted control module is used for monitoring the state of the vehicle in real time, cooperatively controlling self-balancing, electronic stabilization and a braking system to execute active intervention after receiving an instruction, and meanwhile, collecting process data and returning the process data to form a control closed loop. According to the invention, a collaborative architecture of cloud multi-source data fusion and vehicle end dynamic state perception is adopted, a predictive safety closed loop is constructed, and safety normal form transformation from passive response to active prediction is realized.

Description

Technical Field

[0001] This invention relates to the field of two-wheeled vehicle control technology, specifically to a full-condition self-balancing intelligent two-wheeled vehicle control system based on a control torque gyroscope. Background Technology

[0002] Two-wheeled vehicles, with their flexibility and convenience, have become an important part of modern urban transportation. However, their inherent dynamic instability places high demands on the driver's skills, especially when dealing with complex conditions such as high-speed curves, slippery roads, or sudden obstacles.

[0003] To improve the driving safety of two-wheeled vehicles, existing technologies have made some progress. For example, the Electronic Stability Control (ESC) system can effectively suppress the tendency to lose stability by applying precise braking force to the wheels when the vehicle sideslips. The Anti-lock Braking System (ABS) ensures that the wheels are not locked during emergency braking, maintaining the vehicle's steering ability. In addition, some high-end two-wheeled vehicles have begun to use independent gyroscope self-balancing technology to help the vehicle maintain an upright position when traveling at low speeds or standing still.

[0004] However, in-depth analysis reveals that existing technologies still have some fundamental limitations when dealing with complex and ever-changing driving environments. First, these safety systems are generally based on reactive mechanisms, relying on onboard sensors to detect danger signals such as instability and slippage before intervening. This intervention logic inherently has a delay, potentially missing the optimal control opportunity under rapidly changing extreme conditions. Second, the safety models of existing technologies are relatively rigid, typically employing preset, universal control thresholds that are difficult to dynamically adapt to changes in key variables such as road friction coefficients and road curvature. Finally, there is a lack of effective collaborative mechanisms between the various safety subsystems. Functional units such as self-balancing modules and electronic stability modules operate independently, failing to form a combined control force. This architecture can easily lead to insufficient vehicle safety redundancy and potential instability risks when a single module reaches its limit or malfunctions. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a full-condition self-balancing intelligent two-wheeled vehicle control system based on a control torque gyroscope, which solves the problems of response lag, rigid safety models, and low safety redundancy caused by insufficient coordination among various safety subsystems that are common in existing two-wheeled vehicle safety assistance systems.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a full-condition self-balancing intelligent two-wheeled vehicle control system based on a control torque gyroscope, comprising:

[0007] The multi-source data center module is used to aggregate and integrate satellite map data, road feature data, real-time meteorological data and vehicle network crowdsourced data. Based on the vehicle network crowdsourced data, it generates a road surface friction coefficient distribution map through reverse calibration and distributes the integrated data as prior data.

[0008] The data analysis and decision-making module is used to receive the prior data and the vehicle dynamic status, calculate the phased safe speed and graded safe distance under the current working conditions by combining the prior data and the vehicle dynamic status, run the reinforcement learning evaluation function based on the calculation results to assess the risk of the current driving behavior, and generate graded intervention instructions.

[0009] The vehicle control module is used to monitor the relative position of the vehicle and the curve ahead in real time using an integrated vehicle sensor group and obtain the dynamic status of the vehicle. It receives the graded intervention instructions and controls the human-machine interface, electronic stability control system, self-balancing control system or steerable braking system to perform early warning or active intervention operations according to the graded intervention instructions. It also collects and transmits vehicle status data, execution action data and environmental feedback data during the intervention process.

[0010] Preferably, the multi-source data center module generates a road surface friction coefficient distribution map based on reverse calibration of vehicle network crowdsourced data, including:

[0011] The data receiving unit receives real-time dynamic data reported by the vehicle control module from multiple different vehicles. The real-time dynamic data includes at least the peak or stable lateral acceleration reached by the vehicle in a specific road section.

[0012] The data integration unit extracts the road curvature radius corresponding to a specific road segment from the road feature database;

[0013] The data integration unit uses a dynamic back calibration model to calculate an estimate of the road surface friction coefficient based on the ratio of real-time lateral acceleration to gravitational acceleration.

[0014] After the data integration unit collects multiple estimated data points, it applies a Gaussian process regression algorithm to generate a continuous road surface friction coefficient distribution map. The road surface friction coefficient distribution map simultaneously provides the estimated road surface friction coefficient and the uncertainty of the estimated road surface friction coefficient.

[0015] Preferably, the vehicle control module utilizes an integrated vehicle sensor array to monitor the relative position of the vehicle to the curve ahead in real time and obtain the vehicle's dynamic status, including:

[0016] The curve monitoring unit performs fusion positioning by integrating measurement data from the Global Positioning System, Inertial Measurement Unit, and wheel speed sensors, and matches the positioning results with a high-precision map to calculate the curve distance from the vehicle's current position to the entrance point of the curve ahead.

[0017] The vehicle condition monitoring unit acquires the vehicle's longitudinal acceleration, lateral acceleration, and body roll angle through the inertial measurement unit;

[0018] The vehicle condition monitoring unit estimates the vehicle's true longitudinal speed by fusing wheel speed sensor data and longitudinal acceleration data from the inertial measurement unit.

[0019] The vehicle condition monitoring unit estimates the tire slip ratio based on the difference between the estimated true longitudinal velocity and the wheel linear velocity.

[0020] Preferably, the data analysis and decision-making module, combining the prior data and the vehicle's dynamic state, calculates the phased safe speed under the current operating conditions, including:

[0021] The safety distance and speed analysis unit calculates the basic safe speed based on the road surface friction coefficient, road curvature radius, road lateral slope, gravitational acceleration, load coefficient, and gyro coefficient, wherein the basic safe speed is proportional to the product of the road surface friction coefficient and the road curvature radius;

[0022] The safe distance and speed analysis unit dynamically corrects the basic safe speed based on the vehicle's entry into, middle of, and exit from the curve, generating a phased safe speed.

[0023] Preferably, the dynamic correction of the base safety speed includes:

[0024] During the cornering phase, based on the deviation between the vehicle's real-time steering angle and the theoretical target steering angle, a cornering correction coefficient is calculated, and the safe cornering speed is also calculated.

[0025] During the curve phase, the curve correction coefficient is calculated based on the current output power percentage of the self-balancing control system, and the safe speed in the curve is also calculated.

[0026] During the exit phase of a curve, the curve correction coefficient is calculated based on the vehicle's longitudinal acceleration, and the safe speed for exiting the curve is also calculated.

[0027] Preferably, the data analysis and decision-making module calculates the graded safety distance under the current working conditions, including:

[0028] The safe distance and speed analysis unit calculates the first safe distance based on the square difference between the real-time vehicle speed and the phased safe speed, the preset comfort deceleration, the driver's average reaction time, the dynamic safety factor, and the component of gravitational acceleration on the longitudinal slope of the road.

[0029] The safety distance and speed analysis unit calculates the second safety distance based on the square difference between the real-time vehicle speed and the phased safety speed, the vehicle's maximum physical deceleration, the preset safety buffer distance, the dynamic safety factor, and the component of gravitational acceleration on the longitudinal slope of the road.

[0030] The longitudinal slope component in the calculation formulas for the first and second safety distances is taken as negative in downhill conditions to extend the safety distance.

[0031] Preferably, the data analysis and decision-making module generates tiered intervention instructions including:

[0032] The reinforcement learning unit constructs a state vector, which includes at least the real-time vehicle speed, lateral acceleration, tire slip ratio, lateral trajectory deviation, gyroscope output power percentage, and the difference between the current vehicle's curve distance to the entrance of the next curve and the second safety distance.

[0033] The reinforcement learning unit uses a multi-objective reward function to calculate the instantaneous reward value at the current moment. The multi-objective reward function is used to balance safety, comfort, speed adaptability, trajectory tracking performance and energy consumption.

[0034] The reinforcement learning unit updates the value function of the state and action based on the temporal difference learning algorithm;

[0035] The vehicle control decision unit determines the final action as the graded intervention instruction based on the risk level represented by the updated state and action value function, combined with the physical constraints of the first and second safety distances.

[0036] Preferably, the on-board control module controls the self-balancing control system to perform active intervention operations according to the graded intervention command, including:

[0037] The graded early warning implementation unit applies a comprehensive control mode function. When the electronic stability control system has been triggered or the vehicle roll angle exceeds the preset safe roll threshold, the self-balancing system is activated.

[0038] The graded early warning implementation unit calculates the target output torque of the gyroscope based on the redundant torque control model, and controls the gyroscope frame motor to execute the target output torque;

[0039] The redundant torque control model calculates the basic stabilizing torque based on proportional and differential control laws, and provides directional optimization torque according to the change of steering angle to assist in cornering.

[0040] Preferably, the redundant torque control model further includes a redundancy adjustment factor;

[0041] When the vehicle status monitoring unit detects a fault code or response delay in the electronic stability control system or the brake-by-wire system, the graded early warning implementation unit automatically adjusts the redundancy adjustment factor to a preset positive value so that the gyroscope outputs a larger torque to compensate for the stability loss caused by the failure of other subsystems.

[0042] Preferably, the vehicle control module collects and transmits vehicle status data, action data, and environmental feedback data during the intervention process, including:

[0043] During the intervention event, the vehicle control module synchronously collects data from sensors and actuators using a unified clock signal.

[0044] The vehicle control module performs feature extraction and compression on the cached raw data. The feature extraction includes maximum slip rate, average gyro torque, and intervention response time.

[0045] The vehicle control module sends the processed data packets to the multi-source data center module via the vehicle network communication module;

[0046] The multi-source data center module uses the returned data as training samples for reinforcement learning units to iteratively optimize the global reward function weights and the value function table of initial state and action.

[0047] This invention provides a full-condition self-balancing intelligent two-wheeled vehicle control system based on a control torque gyroscope. It has the following beneficial effects:

[0048] 1. This invention adopts a collaborative architecture of cloud-based multi-source data fusion and vehicle-side dynamic state perception to construct a predictive safety closed loop, realizing a safety paradigm shift from passive response to proactive anticipation. Existing technologies mainly rely on real-time monitoring by on-board sensors, which can only trigger intervention after the vehicle enters a dangerous condition, resulting in significant response lag.

[0049] 2. This invention achieves refined assessment and progressive intervention of vehicle risks by constructing a phased and graded dynamic safety model. The control strategy is more in line with the actual needs of complex curves. Traditional two-wheeled vehicle stability control systems usually use fixed safety thresholds, resulting in abrupt intervention timing and difficulty in balancing the safety of extreme conditions with the driving experience of normal conditions.

[0050] 3. This invention deeply integrates the control torque gyroscope into the overall stability control system of the vehicle, which not only expands the application scope of the gyroscope, but also builds a redundant safety mechanism. In the prior art, the self-balancing system and the electronic stability system are often independent modules that lack coordination and functional complementarity. When a certain subsystem fails, the whole vehicle will face the risk of instability. Attached Figure Description

[0051] Figure 1 This is a schematic diagram of the system architecture according to an embodiment of the present invention;

[0052] Figure 2 This is a schematic diagram of the method flow according to an embodiment of the present invention.

[0053] in:

[0054] 10. Multi-source data center module; 20. Data analysis and decision-making module; 30. Vehicle control module. Detailed Implementation

[0055] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] See attached document Figure 1 and attached Figure 2 Appendix Figure 1 This is a schematic diagram of a system architecture according to an embodiment of the present invention. (Attached) Figure 2 This is a flowchart and typical application scenario diagram according to an embodiment of the present invention.

[0057] This invention provides a full-condition self-balancing intelligent two-wheeled vehicle control system based on a control torque gyroscope. Logically, this system may include:

[0058] Multi-source data center module 10;

[0059] Data Analysis and Decision Making Module 20;

[0060] And the vehicle control module 30.

[0061] The multi-source data center module 10 serves as the cloud data center, the data analysis and decision-making module 20 serves as the decision-making core, and the vehicle control module 30 serves as the vehicle execution end. The multi-source data center module 10, the data analysis and decision-making module 20, and the vehicle control module 30 interact through data to construct a real-time secure closed loop for cloud-vehicle collaboration.

[0062] See attached document Figure 2 , Figure 2 This is a flowchart and typical application scenario diagram according to an embodiment of the present invention. The present invention provides a full-condition self-balancing intelligent two-wheeled vehicle control method based on a control torque gyroscope. This method is implemented based on a cloud-vehicle collaborative architecture consisting of a multi-source data center module 10, a data analysis and decision-making module 20, and an on-board control module 30, and includes the following steps:

[0063] S1, the multi-source data center module 10 gathers and integrates satellite map data, road feature data, real-time meteorological data and vehicle network crowdsourced data, generates a road surface friction coefficient distribution map based on the vehicle network crowdsourced data, and sends the integrated data as prior data to the data analysis and decision module 20;

[0064] S2, the vehicle control module 30 utilizes an integrated vehicle sensor group to monitor the relative position of the vehicle to the curve ahead in real time through multi-sensor fusion positioning and vehicle dynamics model estimation, and obtains the vehicle dynamic status including tire slip ratio and lateral trajectory deviation.

[0065] S3, Data Analysis and Decision Module 20 combines prior data and vehicle dynamic status to calculate the phased safe speed and graded safe distance under the current working conditions;

[0066] S4, the data analysis and decision-making module 20 runs the reinforcement learning evaluation function to assess the risk of the current driving behavior, and generates graded intervention instructions based on the risk assessment results and the phased safe speed and graded safe distance.

[0067] S5, the vehicle control module 30 receives the graded intervention command and controls the human-machine interface, electronic stability control system, self-balancing control system or drive-by-wire braking system to perform warning or active intervention operations according to the graded intervention command;

[0068] S6, the vehicle control module 30 collects vehicle status data, execution action data and environmental feedback data during the intervention process and transmits them back to the multi-source data center module 10. The multi-source data center module 10 updates the reinforcement learning model and road friction coefficient distribution map based on the transmitted data.

[0069] The following section will provide a detailed explanation of the technical details involved in each of the above steps, combining specific logic and algorithms.

[0070] See attached document Figure 1 The multi-source data center module 10 of this embodiment of the invention serves as the cloud data center of the system. In terms of hardware, it may include a cloud server and a storage device, and in terms of logic, it may include a data integration unit, a data receiving unit, and a data transmission unit.

[0071] Cloud servers and storage are used to store static and semi-static data, including high-precision map data and road feature databases (such as curve curvature radii). longitudinal slope of the road lateral slope of the road (and historical meteorological data.)

[0072] The data receiving unit is used to receive the data transmitted back from the vehicle control module 30 in step S6. The transmitted data includes vehicle status data, executed action data, and environmental feedback data during the intervention process.

[0073] The data transmission unit is used to execute step S1, which sends the prior data processed by the multi-source data center module 10 to the data analysis and decision module 20.

[0074] The data integration unit is responsible for fusing the data received by the data receiving unit and the data stored in the memory. In one embodiment of the present invention, the data integration unit performs the calculation of the road surface friction coefficient involved in step S1. The dynamic estimation, which may include the following steps:

[0075] S101, the data receiving unit receives real-time dynamic data reported by the on-board control module 30 from multiple different vehicles (i.e., vehicle-to-everything crowdsourced data) while they are driving on a specific road segment. The real-time dynamic data includes at least the peak or steady-state lateral acceleration reached by the vehicle on that road segment. .

[0076] S102, The data integration unit extracts the road curvature radius corresponding to a specific road segment from the road feature database. (Note: This step extracts...) Used to send data to the data analysis and decision-making module 20 for subsequent steps (such as S3) to calculate the safe speed. use).

[0077] S103, the data integration unit applies a dynamic inverse calibration model, based on real-time dynamic data ( ), calculate the estimated value of the road surface friction coefficient :

[0078] ;

[0079] in, This refers to the vehicle's real-time lateral acceleration. This is gravitational acceleration. The model is based on the principle of lateral acceleration of a vehicle near its sideslip limit. With gravitational acceleration The ratio represents the coefficient of friction that the current road surface can provide.

[0080] S104, Data integration unit aggregates multiple estimates After obtaining the data points, a spatial interpolation algorithm is applied to process these sparsely distributed estimated data points in space to generate a continuous road friction coefficient. Distribution diagram. In a specific implementation, the spatial interpolation algorithm can be Gaussian Process Regression (GPR). The advantage of using GPR is that it not only provides a smooth estimate of the friction coefficient based on sparse crowdsourced data points, but also simultaneously provides the uncertainty (i.e., confidence interval) of this estimate. This uncertainty information can be sent along with the data analysis and decision-making module 20 to dynamically adjust the conservatism of the safety model (e.g., adjusting the safety factor). , For the Gaussian process regression algorithm, those skilled in the art can implement it using existing mature technologies. The core of its implementation lies in using the geographic coordinates collected in S103 as input. )of (As a noisy observation) Using this as training data, the hyperparameters of the covariance function are optimized by maximizing the marginal likelihood function, ultimately leading to the optimal results at any new geographic coordinates (query point). Output on The predicted mean and variance (i.e., confidence interval) of the values. This implementation is a conventional technique in this field, so the specific coding will not be described in detail.

[0081] S105, the data integration unit will generate the road surface friction coefficient. The distribution map (and its corresponding uncertainty information) is handed over to the data transmission unit as part of the prior data and sent to the data analysis and decision-making module 20.

[0082] See attached document Figure 1 The data analysis and decision-making module 20 of this embodiment of the invention, as the decision-making core of the system, can be deployed in the vehicle-mounted high-performance computing unit. Logically, the data analysis and decision-making module 20 may include a safe distance and speed analysis unit, a reinforcement learning unit, and a vehicle control decision-making unit. The data analysis and decision-making module 20 is mainly responsible for executing steps S3 and S4.

[0083] The safety distance and speed analysis unit is used to perform step S3 and calculate the dynamic safety model. Its specific implementation may include:

[0084] S301, the safe distance and speed analysis unit receives prior data from the multi-source data center module 10, including the road surface friction coefficient. Road curvature radius longitudinal slope of the road lateral slope of the road and dynamic safety factor , , Simultaneously, it receives real-time data from the vehicle control module 30, including real-time vehicle speed. Real-time steering angle Longitudinal acceleration gyroscope output power percentage Tire slip ratio and lateral trajectory deviation .

[0085] S302, Safety Distance and Speed ​​Analysis Unit calculates basic safe speed. This calculation is based on a dynamic model that comprehensively considers road surface adhesion conditions, road superelevation, and vehicle characteristics:

[0086] ;

[0087] in, The coefficient of friction of the road surface; It is the acceleration due to gravity; The radius of curvature of the road; For load factor; The gyroscope coefficient (where and (These are all calibration coefficients preset based on vehicle model and gyroscope performance). The height of the vehicle's center of gravity; The wheelbase of the vehicle; This refers to the lateral slope angle of the road.

[0088] In a preferred embodiment, the calibration coefficients are obtained as follows: The load factor can be obtained through two or more dynamic tests of the vehicle before it leaves the factory, under both no-load and full-load conditions, and is used to correct for changes in the center of mass under different loads. The gyroscope coefficient can be calibrated through bench testing of the gyroscope, which is a quantitative model characterizing the gyroscope's contribution to the vehicle's stability gain at different power levels.

[0089] S303, Safety Distance and Speed ​​Analysis Unit Based on Basic Safety Speed Calculate the phased safety speed This calculation is based on the vehicle's entry into, middle of, and exit from the curve. Perform dynamic correction:

[0090] a) Calculate the safe entry speed during the curve entry phase. :

[0091] ;

[0092] in, This is the entry correction factor, and its calculation method is as follows: . For the vehicle's real-time steering angle, Based on road curvature radius The calculated theoretical target steering angle, in a specific implementation, It can be calculated using well-known vehicle steady-state steering models (such as Ackermann steering geometry), for example: ,in This refers to the vehicle's wheelbase (preset parameter). This represents the current road curvature radius. This correction is used to limit the entry speed when steering deviation is large.

[0093] b) Calculate the safe speed in the curve during the mid-curve phase. :

[0094] ;

[0095] in, This is the correction factor for the bend, and its calculation method is as follows: . This represents the current percentage of output power of the self-balancing control system (gyroscope). This correction reflects the gyroscope's contribution to vehicle stability.

[0096] c) Calculate the safe exit speed during the exit phase of the curve. :

[0097] ;

[0098] in, This is the correction factor for exiting a curve, and its calculation method is as follows: . This refers to the vehicle's longitudinal acceleration. This correction is used to limit excessive acceleration when exiting a corner.

[0099] S304, Safety Distance and Speed ​​Analysis Unit Based on Phased Safety Speed (Right now , , (one of them) and real-time vehicle speed Calculate the graded safety distances:

[0100] a) Calculate the first safety distance (Warning distance):

[0101] ;

[0102] b) Calculate the second safety distance (Intervention distance):

[0103] ;

[0104] in, The preset comfort deceleration (e.g., 0.3g); The maximum physical deceleration of the vehicle (e.g., 0.8g) can be determined by... (Value calibration); The preset average driver reaction time (e.g., 1.0s). The preset safety buffer distance (e.g., 5m); , , Dynamic safety factor; The longitudinal slope angle of the road, in the formula The term is used to compensate for the longitudinal component of gravitational acceleration on the slope, where it is used when going downhill ( For negative values, take "−", when going uphill ( (For positive values) take the plus sign. This setting ensures that the denominator is positive during downhill driving conditions. Decrease, calculated and It automatically extends to match the longer braking distance required in downhill conditions.

[0105] In one embodiment, the dynamic security factor The calculations can be further aided by gyroscopes, for example: This setting enables higher gyroscope power. At higher levels, Reduce, thereby shortening the intervention distance The underlying technology is that the gyroscope provides additional stability, allowing the system to operate more smoothly closer to curves (i.e.,...). Active intervention is only carried out when the time is shorter (less than the required time).

[0106] The reinforcement learning unit is used to perform step S4, which involves assessing driving risks and optimizing the control strategy through online learning. To fully disclose the implementation details of this process, step S4 further includes the following sub-steps:

[0107] S401, State Space Construction. The reinforcement learning unit constructs the current state vector from the input data in step S301. State vector include: ,in For the present Intervention distance The difference.

[0108] Meanwhile, the reinforcement learning unit defines a discrete action space. Action space This refers to the intervention levels that the system can choose at each decision-making moment. For example, the action space can be defined as: . : No intervention (corresponding to "maintain the status quo" in S404); Level 1 warning (corresponding to "Trigger Level 1 warning" in S404); Moderate-intensity intervention (e.g., triggering S504 and setting target deceleration) (0.3g) High-intensity intervention (e.g., triggering S504 and setting target deceleration) (0.6g). Through this definition, the Q-value function in S403... The meaning then becomes: in the state Next, take action (For example The long-term expected return of ).

[0109] S402, Multi-objective reward calculation. Reinforcement learning units utilize multi-objective reward functions. Calculate the instantaneous reward value at the current moment. This function is designed to balance safety, comfort, and energy consumption. Its calculation formula is as follows:

[0110] ;

[0111] Among them, the first item (weight) Used to evaluate comfort, penalizing excessive lateral acceleration. The second item (weight) Used to evaluate stability and penalize excessive tire slip ratio. (in The range of values ​​is ); the third item (weight) Used to evaluate speed adaptability and reward actual speed. Approaching safe speed ; Fourth item (weight) Used to evaluate trajectory tracking performance and penalize lateral deviation. ; Fifth item (weight) The negative sign is used to evaluate energy efficiency, rewarding lower gyroscope power consumption. . , These are constant coefficients. In a specific implementation, the weighting coefficients... to The multi-source data center module 10 can distribute data according to different operating conditions, such as weight. It can be dynamically adjusted according to the curve stage (entering the curve, in the curve, exiting the curve).

[0112] S403, Q-value update. The reinforcement learning unit updates the state-action value function based on the Temporal Difference (TD) learning algorithm. The updated formula is:

[0113] ;

[0114] in, For evaluation functions; For the state space (e.g., containing S301) , , , wait); For action space; The learning rate; This is the discount factor.

[0115] S404, Strategy Selection and Command Generation. The vehicle control decision unit, based on the risk level represented by the updated Q-value, combines the physical constraints calculated by S304 (…). , ), determine the final action This action These are tiered intervention instructions, including: maintaining the status quo, triggering a Level 1 warning, or triggering a Level 2 / 3 physical intervention and corresponding control parameters (such as target deceleration). This logic ensures that the system adheres to both physical security boundaries (as defined by...). , It has both hard constraints and adaptive optimization capabilities (soft adjustment by RL).

[0116] See attached document Figure 1 The vehicle control module 30 of this embodiment of the invention, as the sensing and execution end of the system, may include, in hardware, an environmental perception sensor group (e.g., millimeter-wave radar, camera), a vehicle status sensor group (e.g., IMU inertial measurement unit, wheel speed sensor, steering angle sensor), and actuators (including human-machine interface HMI, electronic stability control system ESC, self-balancing control system, and brake-by-wire system BBW). The vehicle control module 30 is mainly responsible for executing steps S2, S5, and S6.

[0117] Logically, the vehicle control module 30 may include a curve monitoring unit, a vehicle status monitoring unit, and a graded early warning implementation unit.

[0118] The curve monitoring unit and the vehicle status monitoring unit are used to collaboratively execute step S2. In order to fully disclose the details of acquiring the sensing parameters, step S2 further includes the following sub-steps:

[0119] S201, Environmental Perception and Fusion Positioning. The curve monitoring unit performs high-precision positioning to calculate relative distance. and lateral deviation Provide benchmarks:

[0120] a) Fusion Positioning: The curve monitoring unit first fuses measurement data from the Global Positioning System (GPS), Inertial Measurement Unit (IMU), and wheel speed sensors using a Kalman filter to output the vehicle's real-time absolute coordinates (latitude and longitude), heading angle, and elevation. Accelerometer and gyroscope data from the IMU are used to compensate for the low-frequency (e.g., 10Hz) update rate of the GPS signal at high frequencies (e.g., 100Hz), achieving smooth and high-precision trajectory calculation.

[0121] b) Map matching: The curve monitoring unit matches the absolute coordinates obtained in a) with a high-precision map (HD Map) acquired from the multi-source data center module 10 or stored locally. This matching process positions the vehicle on the centerline of a specific lane on the map, which serves as the "reference path" for subsequent calculations.

[0122] c) Relative Distance Calculation: The curve monitoring unit calculates the curve distance from the vehicle's current position to the entrance point of the upcoming curve (defined by map data) along the "reference path" matched in b). This distance is the relative distance. .

[0123] d) Positioning Degradation Processing: In a preferred embodiment, when a weak GPS signal (e.g., in tunnels or mountainous areas) causes a decrease in the confidence level of the positioning result in a), the curve monitoring unit automatically switches to visual SLAM (Simultaneous Localization and Mapping) mode. This mode uses image feature points collected by the camera to match visual landmarks in a high-precision map, or calculates them through visual odometry, to maintain positioning continuity and estimate... .

[0124] S202, Vehicle Dynamic State Estimation. The vehicle condition monitoring unit uses data from the vehicle condition sensor array, combined with a dynamic model, to estimate the key dynamic parameters required for S3 and S4 in real time.

[0125] a) Basic State Acquisition: The vehicle state monitoring unit acquires the basic state from its sensor array. Specifically, the longitudinal acceleration of the vehicle is acquired via the IMU (Inertial Measurement Unit). Lateral acceleration Body roll angle yaw rate and yaw rate The driver's real-time steering angle is obtained via a steering angle sensor (the hardware is listed at the beginning of S2). .

[0126] b) Actual vehicle speed estimation: The vehicle condition monitoring unit estimates the vehicle's actual longitudinal speed by fusing wheel speed sensor data from both rear wheels (or non-drive wheels) and compensating for it with longitudinal acceleration data from the IMU. (Right now Using data from non-drive wheels can reduce the interference of slippage caused by driving / braking forces on speed estimation.

[0127] c) Tire slip ratio estimation: Tire slip ratio This is a core parameter reflecting tire adhesion and is crucial for S4 (RL bonus) and S5 (ABS). The vehicle condition monitoring unit calculates it using the following formula (taking braking conditions as an example):

[0128] ;

[0129] in, The actual vehicle speed estimated in b); Let be the linear velocity of the wheel, and its calculation formula is: This is the effective rolling radius of the wheel (preset calibration value). This is the real-time angular velocity of the front wheel (or brake wheel) (measured by a wheel speed sensor).

[0130] d) Lateral trajectory deviation estimation: Lateral trajectory deviation This is a key input for evaluating trajectory tracking performance in S4 (RL reward). The vehicle state monitoring unit calculates the shortest vertical distance from the vehicle center point to the "reference path" using the fused localization result (vehicle center point coordinates) from a) in S201 and the map matching result ("reference path") from b). This distance is the... .

[0131] e) Actuator Status Query: The vehicle status monitoring unit periodically queries the actuators (especially the self-balancing control system) in the on-board control module 30. By reading the drive controller status of the control torque gyroscope, it obtains the current real-time gyroscope output power percentage. This parameter characterizes the current work intensity of the gyroscope and is an important input for S3 and S4 to perform stability evaluation and decision-making.

[0132] The tiered early warning implementation unit is used to execute step S5, which involves coordinating the control of various implementing mechanisms based on the tiered intervention instructions sent by the data analysis and decision-making module 20. Its control logic may include the following sub-steps:

[0133] S501, when receiving a Level 1 risk warning instruction (corresponding to...) When a tiered early warning implementation unit activates, the human-machine interface (HMI) issues an alarm. Alarms include audible alarms (buzzer), visual alarms (blinking lights on the dashboard), or tactile alarms (vibrating handle motor pulses).

[0134] S502, when a Level 2 risk intervention instruction is received (corresponding to...) When the tiered early warning implementation unit activates the Electronic Stability Control (ESC) and the self-balancing control system (control torque gyroscope), the unit executes the coordinated triggering logic of the ESC and the self-balancing system.

[0135] a) ESC trigger determination: Calculate the vehicle's yaw rate deviation. ,in The current curve radius is obtained from the prior data issued by the multi-source data center module 10 in step S1. When Greater than the preset yaw threshold When, determine the ESC trigger condition. If true, the ESC system applies differential braking force to the wheels to correct the course.

[0136] b) Trigger determination of self-balancing system: applying integrated control mode function Determine whether to intervene. The self-balancing system will initiate intervention when any of the following conditions are met:

[0137] 1) The ESC system has been triggered (i.e.) (If true);

[0138] 2) Vehicle roll angle Exceeding the preset safe roll threshold (That is, a tendency to become unstable appears).

[0139] S503, when the self-balancing control system in S502 is activated, the graded early warning implementation unit calculates the target output torque of the gyroscope based on the redundant torque control model. The system controls the gyroscope frame motor to execute this torque. The calculation formula for the redundant torque control model is:

[0140] ;

[0141] in, Based on the fundamental stabilizing torque, the calculation is performed using the PD (proportional-derivative) control law: ,in , The preset gain parameters for the PD controller. For the real-time vehicle speed collected in step S202 The curve radius obtained in step S502a The calculated ideal roll angle, in a specific implementation, It can be calculated using well-known vehicle steady-state cornering roll models, for example: ,in It is the acceleration due to gravity; Optimize torque for steering, used to assist with cornering. When the vehicle is detected entering the cornering transition phase (i.e., steering angle adjustment),... When the sign changes, It outputs a pulse torque in the same direction as the steering to accelerate the vehicle's rollover; This is a redundancy adjustment factor, with a default value of 0. When the vehicle condition monitoring unit detects a fault code, degradation, or response delay in the ESC system or the brake-by-wire system, the graded warning implementation unit automatically... Adjust to a preset positive value (e.g., 0.3 to 0.5). This setting allows the gyroscope to output a larger torque to compensate for the stability loss caused by the failure of other subsystems, achieving functional redundancy.

[0142] S504, when receiving a Level 3 risk intervention instruction (corresponding to...) and When the tiered warning implementation unit maintains the gyroscope control in step S503, it further activates the brake-by-wire (BBW) system. The tiered warning implementation unit sends the target braking deceleration to the BBW actuator. instruction. The calculation is performed at the second safe distance. (or current) The vehicle speed will be reduced from ) Reduce to For the goal, and The value is defined and taken into account in step S304 by the data analysis and decision-making module 20. and Maximum physical deceleration The brake-by-wire system, when performing this deceleration, combines with anti-lock braking logic (ABS) to ensure that the applied braking force does not cause tire slippage. The slip ratio peak value exceeds the optimal adhesion coefficient.

[0143] After the intervention operation is completed, the system needs to complete the data loop closure. The vehicle control module 30 and the multi-source data center module 10 work together to execute step S6, which specifically includes the following sub-steps:

[0144] S601, Synchronous Data Acquisition and Buffering. During the intervention event, the onboard control module 30 synchronously acquires data from various sensors and actuators using a unified clock signal. The acquisition frequency can be dynamically adjusted, for example, increased to 200Hz during the S504 Level 3 intervention. The data is temporarily stored in the onboard high-speed buffer, forming a time-series data packet.

[0145] S602, Feature Extraction and Compression. To reduce transmission bandwidth, the vehicle control module 30 performs edge processing on the cached raw data. This processing includes extracting key feature values ​​(such as maximum slip rate, average gyro torque, and intervention response time) and performing lossless compression on the raw waveform data.

[0146] S603, Data Encryption and Backhaul. The vehicle control module 30 sends the processed data packet to the multi-source data center module 10 via a vehicle-to-everything (V2X) communication module (such as a 5G / V2X module). The data packet clearly indicates the geographical coordinates, environmental parameters (such as temperature and humidity), and intervention result labels (such as "successful avoidance" or "ABS triggered").

[0147] S604, Cloud Model Update. The data receiving unit of the multi-source data center module 10 receives this data packet. The data integration unit stores it in the historical database and triggers two update processes:

[0148] 1) Friction coefficient map update: using the actual slip ratio in the returned data. and lateral acceleration Correct this section of road Value confidence level;

[0149] 2) RL Parameter Evolution: This intervention process is treated as an "episode" sample and input into the cloud-based training cluster. The global reward function weights are optimized using batch gradient descent. and the initial Q value table ( ).

[0150] S605, parameters issued. Updated model parameters (such as weights optimized for this road segment). Or revised (Figure) When a vehicle requests data for this area again (i.e., step S1), it is sent as prior data, thereby enabling the continuous evolution of the control strategy.

[0151] To further illustrate the collaborative working process of the technical solution of this invention, a specific working scenario example will be used below.

[0152] Imagine a typical complex driving condition: a vehicle is traveling on a mountain road and is about to enter a series of S-shaped curves. This road section has the following characteristics: the road surface is slippery, resulting in a low coefficient of friction. The slope is relatively low; and the vehicles are traveling downhill continuously, resulting in a low longitudinal slope of the road. A negative value. The vehicle's current actual speed. The reading is too high, exceeding the safety threshold for this operating condition.

[0153] Before the vehicle enters the road section, the multi-source data center module 10 has executed step S1, sending the prior data of the road section to the data analysis and decision-making module 20 via the data transmission unit. This prior data includes: a lower road surface friction coefficient calibrated based on crowdsourced data (S103). The value, and the radius of curvature of continuous S-curves extracted from the road feature database. ( The value will change continuously and may reverse) and negative values ​​for road longitudinal slope. .

[0154] Simultaneously, the vehicle control module 30 executes step S2, and the curve monitoring unit (S201) monitors the relative distance between the vehicle and the entrance of the S-curve in real time. The vehicle condition monitoring unit (S202) reported a high vehicle speed. .

[0155] The data analysis and decision-making module 20's safety distance and speed analysis unit executes step S3. During step S304, when calculating the graded safety distance, due to... A negative value (downhill) results in a lower safe distance. and denominator in the calculation formula Decrease. This calculation result automatically reduces. and The calculated value is significantly extended to match the longer braking distance required for downhill conditions. Meanwhile, (S303) is based on a lower... Value calculated Calculate the lower phased safety speed ( The vehicle control decision unit (S4) is based on... With extension , Compare and determine the risk level.

[0156] Assuming the driver fails to anticipate road conditions and continues at a high speed Approaching the curve. When S201 monitors... Less than the value calculated by S304 At that time, the graded warning implementation unit of the vehicle control module 30 executes step S501, triggering a first-level risk warning and reminding the driver to slow down through the human-machine interface (HMI).

[0157] The vehicle continues to move even though the driver does not respond or responds insufficiently. Less than the value calculated by S304 At this time, the graded early warning implementation unit executes step S502, triggering a level-two risk intervention. Then, the self-balancing control system (gyroscope) intervenes, executing S503, based on the basic stabilizing torque. Provides roll stability.

[0158] The system further determines Still greater than the value calculated by S303 (For example The graded early warning implementation unit executes step S504, triggering a level-three risk intervention. The brake-by-wire system (BBW) is activated, applying the applied pressure. Constrained target braking deceleration Active braking is applied to forcibly reduce the vehicle speed.

[0159] As the vehicle enters the second reverse curve of the series of S-curves, the vehicle condition monitoring unit (S202) detects the steering angle. The sign changes. When executing the redundant torque control model of S503, the graded early warning implementation unit calculates the directional optimization torque. .Should The torque actively assists the vehicle body to tilt in the opposite direction, significantly improving the response speed and stability of the vehicle when switching S-curve attitudes in slippery downhill conditions.

[0160] After the entire intervention process is completed, the vehicle control module 30 executes step S6 (S505), which transmits the intervention data (including data from the wet and slippery S-curve downhill section) to the vehicle control module 30. , , gyroscope output torque and BBW braking deceleration The data is packaged and sent back to the multi-source data center module 10. The data integration unit of the multi-source data center module 10 receives this data and uses it for the iteration of S104 to fine-tune the road surface friction coefficient of this section. The confidence level of the distribution plot; on the other hand, this data serves as training samples for the reinforcement learning unit (S4), used to iteratively optimize the reward function. Weighting (e.g., increasing the weighting under wet conditions) For slip ratio The penalty weights are applied to continuously optimize the system decision model.

[0161] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A full-condition self-balancing intelligent two-wheeled vehicle control system based on a control torque gyroscope, characterized in that, include: The multi-source data center module is used to aggregate and integrate satellite map data, road feature data, real-time meteorological data and vehicle network crowdsourced data. Based on the vehicle network crowdsourced data, it generates a road surface friction coefficient distribution map through reverse calibration and distributes the integrated data as prior data. The data analysis and decision-making module is used to receive the prior data and the vehicle dynamic status, calculate the phased safe speed and graded safe distance under the current working conditions by combining the prior data and the vehicle dynamic status, run the reinforcement learning evaluation function based on the calculation results to assess the risk of the current driving behavior, and generate graded intervention instructions. The vehicle control module is used to monitor the relative position of the vehicle and the curve ahead in real time using an integrated vehicle sensor group and obtain the dynamic status of the vehicle. It receives the graded intervention instructions and controls the human-machine interface, electronic stability control system, self-balancing control system or steerable braking system to perform early warning or active intervention operations according to the graded intervention instructions. It also collects and transmits vehicle status data, execution action data and environmental feedback data during the intervention process.

2. The all-condition self-balancing intelligent two-wheeled vehicle control system based on a control torque gyroscope according to claim 1, characterized in that, The multi-source data center module generates a road surface friction coefficient distribution map based on reverse calibration using crowdsourced data from the Internet of Vehicles, including: The data receiving unit receives real-time dynamic data reported by the vehicle control module from multiple different vehicles. The real-time dynamic data includes at least the peak or stable lateral acceleration reached by the vehicle in a specific road section. The data integration unit extracts the road curvature radius corresponding to a specific road segment from the road feature database; The data integration unit uses a dynamic back calibration model to calculate an estimate of the road surface friction coefficient based on the ratio of real-time lateral acceleration to gravitational acceleration. After the data integration unit collects multiple estimated data points, it applies a Gaussian process regression algorithm to generate a continuous road surface friction coefficient distribution map. The road surface friction coefficient distribution map simultaneously provides the estimated road surface friction coefficient and the uncertainty of the estimated road surface friction coefficient.

3. The all-condition self-balancing intelligent two-wheeled vehicle control system based on a control torque gyroscope according to claim 1, characterized in that, The vehicle control module utilizes an integrated onboard sensor array to monitor the relative position of the vehicle to the curve ahead in real time and obtain the vehicle's dynamic status, including: The curve monitoring unit performs fusion positioning by integrating measurement data from the Global Positioning System, Inertial Measurement Unit, and wheel speed sensors, and matches the positioning results with a high-precision map to calculate the curve distance from the vehicle's current position to the entrance point of the curve ahead. The vehicle condition monitoring unit acquires the vehicle's longitudinal acceleration, lateral acceleration, and body roll angle through the inertial measurement unit; The vehicle condition monitoring unit estimates the vehicle's true longitudinal speed by fusing wheel speed sensor data and longitudinal acceleration data from the inertial measurement unit. The vehicle condition monitoring unit estimates the tire slip ratio based on the difference between the estimated true longitudinal velocity and the wheel linear velocity.

4. The all-condition self-balancing intelligent two-wheeled vehicle control system based on a control torque gyroscope according to claim 1, characterized in that, The data analysis and decision-making module, combining the prior data and the vehicle's dynamic state, calculates the phased safe speed under the current operating conditions, including: The safety distance and speed analysis unit calculates the basic safe speed based on the road surface friction coefficient, road curvature radius, road lateral slope, gravitational acceleration, load coefficient, and gyro coefficient, wherein the basic safe speed is proportional to the product of the road surface friction coefficient and the road curvature radius; The safe distance and speed analysis unit dynamically corrects the basic safe speed based on the vehicle's entry into, middle of, and exit from the curve, generating a phased safe speed.

5. The all-condition self-balancing intelligent two-wheeled vehicle control system based on a control torque gyroscope according to claim 4, characterized in that, The dynamic correction of the base safety speed includes: During the cornering phase, based on the deviation between the vehicle's real-time steering angle and the theoretical target steering angle, a cornering correction coefficient is calculated, and the safe cornering speed is also calculated. During the curve phase, the curve correction coefficient is calculated based on the current output power percentage of the self-balancing control system, and the safe speed in the curve is also calculated. During the exit phase of a curve, the curve correction coefficient is calculated based on the vehicle's longitudinal acceleration, and the safe speed for exiting the curve is also calculated.

6. The all-condition self-balancing intelligent two-wheeled vehicle control system based on a control torque gyroscope according to claim 1, characterized in that, The data analysis and decision-making module calculates the graded safety distances under the current operating conditions, including: The safe distance and speed analysis unit calculates the first safe distance based on the square difference between the real-time vehicle speed and the phased safe speed, the preset comfort deceleration, the driver's average reaction time, the dynamic safety factor, and the component of gravitational acceleration on the longitudinal slope of the road. The safety distance and speed analysis unit calculates the second safety distance based on the square difference between the real-time vehicle speed and the phased safety speed, the vehicle's maximum physical deceleration, the preset safety buffer distance, the dynamic safety factor, and the component of gravitational acceleration on the longitudinal slope of the road. The longitudinal slope component in the calculation formulas for the first and second safety distances is taken as negative in downhill conditions to extend the safety distance.

7. The all-condition self-balancing intelligent two-wheeled vehicle control system based on a control torque gyroscope according to claim 1, characterized in that, The data analysis and decision-making module generates tiered intervention instructions, including: The reinforcement learning unit constructs a state vector, which includes at least the real-time vehicle speed, lateral acceleration, tire slip ratio, lateral trajectory deviation, gyroscope output power percentage, and the difference between the current vehicle's curve distance to the entrance of the next curve and the second safety distance. The reinforcement learning unit uses a multi-objective reward function to calculate the instantaneous reward value at the current moment. The multi-objective reward function is used to balance safety, comfort, speed adaptability, trajectory tracking performance and energy consumption. The reinforcement learning unit updates the value function of the state and action based on the temporal difference learning algorithm; The vehicle control decision unit determines the final action as the graded intervention instruction based on the risk level represented by the updated state and action value function, combined with the physical constraints of the first and second safety distances.

8. The all-condition self-balancing intelligent two-wheeled vehicle control system based on a control torque gyroscope according to claim 1, characterized in that, The on-board control module controls the self-balancing control system to perform active intervention operations according to the graded intervention instructions, including: The graded early warning implementation unit applies a comprehensive control mode function. When the electronic stability control system has been triggered or the vehicle roll angle exceeds the preset safe roll threshold, the self-balancing system is activated. The graded early warning implementation unit calculates the target output torque of the gyroscope based on the redundant torque control model, and controls the gyroscope frame motor to execute the target output torque; The redundant torque control model calculates the basic stabilizing torque based on proportional and differential control laws, and provides directional optimization torque according to the change of steering angle to assist in cornering.

9. The all-condition self-balancing intelligent two-wheeled vehicle control system based on a control torque gyroscope according to claim 8, characterized in that, The redundant torque control model further includes a redundancy adjustment factor; When the vehicle status monitoring unit detects a fault code or response delay in the electronic stability control system or the brake-by-wire system, the graded early warning implementation unit automatically adjusts the redundancy adjustment factor to a preset positive value so that the gyroscope outputs a larger torque to compensate for the stability loss caused by the failure of other subsystems.

10. The all-condition self-balancing intelligent two-wheeled vehicle control system based on a control torque gyroscope according to claim 1, characterized in that, The vehicle control module collects and transmits vehicle status data, action data, and environmental feedback data during the intervention process, including: During the intervention event, the vehicle control module synchronously collects data from sensors and actuators using a unified clock signal. The vehicle control module performs feature extraction and compression on the cached raw data. The feature extraction includes maximum slip rate, average gyro torque, and intervention response time. The vehicle control module sends the processed data packets to the multi-source data center module via the vehicle network communication module; The multi-source data center module uses the returned data as training samples for reinforcement learning units to iteratively optimize the global reward function weights and the value function table of initial state and action.

Images