Wheel-legged robot real-time working condition identification method based on fusion of bus steering engine and imu

Abstract

A wheel-legged robot working condition real-time identification method based on bus steering engine and IMU fusion belongs to the technical field of wheel-legged mobile robot control. Firstly, wheel speed-IMU fusion vehicle speed estimation and wheel ground slip rate calculation are carried out based on adaptive anti-disturbance Kalman filtering. Secondly, a six-dimensional working condition characteristic vector containing global slip rate, pitch-torque correlation degree, vertical impact intensity, abnormal joint torque rise rate, roll-lateral slip abnormality and vibration frequency spectrum roughness is constructed based on the steering engine-IMU physical deviation model. Finally, based on the TSK fuzzy inference machine, eight working conditions such as skidding, uphill, downhill, step obstacle crossing, joint overload, lateral slip, severe jolt and normal flat ground are comprehensively judged and the confidence is output. The application greatly reduces the dependence on external sensing sensors, significantly reduces the system cost and complexity, and has high real-time performance, strong robustness and platform migration ability.

Description

Technical Field

[0001] This invention belongs to the field of wheeled mobile robot control technology, and in particular, it is a method for identifying the working condition of a wheeled robot based on the fusion of bus servo multi-parameter and IMU feedback. Background Technology

[0002] Wheeled mobile robots are core equipment in fields such as field inspection, emergency rescue, and unstructured terrain exploration. Their control systems must possess precise attitude perception, condition recognition, and adaptive adjustment capabilities. Existing wheeled robot control systems generally rely on a single IMU for attitude perception. However, the IMU can only provide information on the overall angular velocity and acceleration of the robot body and cannot directly measure detailed condition parameters such as wheel-ground contact state, joint load characteristics, wheel-end speed, and slippage. The system struggles to distinguish whether changes in body attitude are caused by terrain undulations, wheel-end slippage, or joint jamming, resulting in a lack of critical condition perception capabilities such as slippage recognition, load detection, and obstacle crossing determination. To compensate for this deficiency, existing solutions have to add a large number of external sensing units, such as foot force sensors, joint torque sensors, contact sensors, and wheel-end encoders. This not only significantly increases the overall cost and structural complexity of the machine, but also leads to extremely cumbersome wiring. Furthermore, external sensors are susceptible to interference from environmental factors such as vibration, humidity, and dust in the field, resulting in poor data reliability, high failure rates, and difficult maintenance. At the same time, traditional wheeled robots mostly use ordinary PWM open-loop servo motors for driving, which can only send position control commands downwards and cannot transmit back core operating parameters such as real-time joint position, speed, and temperature. The main controller only... While relying on a single IMU to acquire global attitude data of the fuselage, it cannot accurately identify detailed operating conditions such as wheel slippage, joint jamming, sudden terrain changes, and overload. This results in issues such as control lag, insufficient closed-loop control accuracy, and poor terrain adaptability. Furthermore, the multi-sensor fusion algorithm of the existing control system is highly complex, places stringent demands on the main control computing power, has poor real-time performance, and lacks electrical isolation in the power supply system. The large current surges during servo start-up and shutdown, as well as fluctuations in the host computer load, can easily interfere with the stable operation of the main control and IMU sensors. As a result, the overall system reliability is insufficient, making it difficult to meet the long-term stable use requirements of complex and unstructured scenarios.

[0003] The robot is a modular four-wheeled robot driven by a fully bus servo motor. It adopts a symmetrical four-wheeled layout and consists of five core parts: the main body, four sets of modular wheel and leg drive devices, an end effector wheel drive device, a power module, and a main control module. Relying on a fully bus serial drive architecture, the same actuator has both wheeled high-speed passage and leg obstacle crossing functions. It does not require additional external force / torque sensors, has a simple structure, high integration, and is easy to disassemble and maintain. It is suitable for inspection and operation scenarios in unstructured terrain.

[0004] Modular wheel-leg drive unit: Each wheel-leg is a multi-link articulated structure driven by a dual-bus servo motor. The bus servo motor, in conjunction with the transmission servo disc, transmission shaft, and angular contact bearing, forms the hip joint transmission chain. The leg rods are hinged together by bearings and rivets. Each leg is an independent and complete module that can be quickly disassembled and replaced, resulting in low maintenance costs.

[0005] Wheel drive unit: Integrated at the end of the wheel leg, it adopts single-bus servo motor coaxial drive, and forms a complete transmission chain through transmission servo disc, transmission shaft, angular contact bearing, positioning spacer, and shaft end retaining ring. The end is connected to the walking wheel through a spline structure. The same actuator has both wheel drive and foot joint support functions, and can achieve smooth switching between wheel and foot modes without additional switching mechanism. Summary of the Invention

[0006] To fundamentally address the shortcomings of existing wheeled robot control systems—which rely heavily on external sensors for terrain perception and operational condition identification—and the resulting problems of high cost, complex structure, susceptibility to environmental interference, and poor real-time control, this invention proposes a sensorless real-time operational condition identification method based on deep fusion of bus servo intrinsic parameter feedback and IMU (Integrated Device Unit). This method abandons the traditional stacked perception paradigm of "adding sensors first and then fusing them," and creatively constructs a three-level progressive, closed-loop collaborative joint perception framework of "kinematic estimation, multi-dimensional feature mining, and fuzzy operational condition reasoning." This method utilizes only two types of native data—rotation speed, position, and temperature—returned from the bus servo and attitude, acceleration, and angular velocity from the onboard IMU—through the coordinated operation of the following three core steps. For the first time in the field of wheeled robots, this method achieves accurate real-time identification of all operational conditions without the need for additional external sensors for foot force, joint torque, and contact, while also possessing high dynamic accuracy and strong model robustness.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: A real-time identification method for the working conditions of a wheeled robot based on the fusion of bus servo motor and IMU includes the following steps: Step 1: Vehicle speed estimation and wheel slip ratio calculation based on adaptive disturbance rejection Kalman filter: A longitudinal vehicle speed estimator is established, jointly driven by IMU horizontal acceleration integral and servo wheel speed feedback. An adaptive Kalman filter with time-varying observation noise covariance is adopted. The average linear velocity of the driving wheel is used as the observation and IMU acceleration integral is used as the process prior. The sensor zero bias is dynamically compensated and the contamination of speed estimation by slip conditions is suppressed. On this basis, the slip ratio of each wheel is calculated in real time, providing an accurate longitudinal motion reference for subsequent feature extraction. Step 2: Construction of multi-dimensional working condition feature vectors based on the servo IMU physical deviation model: Starting from the "consistency / deviation relationship between the internal state of the servo and the motion state of the fuselage", six interpretable and non-redundant normalized features are designed and extracted, including: global slip rate feature, pitch torque correlation feature, vertical impact intensity feature, joint torque abnormal rise rate feature, roll lateral slip abnormality feature, and vibration spectrum roughness feature. Each feature is generated by the physical deviation mathematical model between the servo position / speed / temperature and the IMU attitude / acceleration. No offline training or precise dynamic parameter model is required. It has natural robustness to unmodeled factors such as load changes and joint wear. Step 3: Multi-condition integrated decision and confidence assessment based on TSK fuzzy inference: The six-dimensional feature vector is input into a regularized TSK fuzzy inference engine. Seventeen expert rules are used to comprehensively map eight operating conditions: slippage, uphill, downhill, obstacle crossing, joint overload, lateral slippage, severe bumps, and normal flat ground. The confidence score and auxiliary control parameters (such as estimated climbing angle, slip wheel ID, and overload joint ID) are output. This inference engine is interpretable, has extremely low computational requirements, and can complete single-cycle operations within sub-milliseconds, fully meeting the real-time control requirements above 100Hz.

[0008] The beneficial effects of this invention are mainly reflected in: This invention completely eliminates the reliance on external sensing sensors, significantly reducing system cost and structural complexity. Existing wheeled robots require numerous external sensing units, such as force sensors, torque sensors, contact sensors, and encoders, to achieve terrain perception and working condition identification. This invention completely replaces these external sensors by deeply exploring the physical deviation relationship between the inherent feedback capability (speed, position, temperature) of the bus servo motor and the onboard IMU data. This improvement directly brings three benefits: First, a significant reduction in single-unit hardware cost, eliminating all external sensors and their signal conditioning circuits; second, a simplified robot structure, eliminating the need for reserved space and wiring channels for sensor installation at joints and feet, greatly increasing the freedom of mechanical design; and third, a sharp reduction in system assembly and maintenance workload, eliminating the need for repeated calibration of external sensors and avoiding cable breakage due to pulling and bending during field operations.

[0009] In current technology, traditional wheel speed IMU fusion schemes introduce significant biases in wheel speed observations when severe wheel-to-ground slip occurs, leading to divergent speed estimates. The adaptive Kalman filter used in step one of this invention dynamically adjusts the observation noise covariance by real-time monitoring of the slip indication between wheel speed and predicted vehicle speed. Upon detecting slippage, it automatically reduces the confidence weight of abnormal wheel speeds, relying more on IMU acceleration integrals for speed inference. This mechanism ensures that vehicle speed estimation remains convergent even under extreme conditions where the drive wheels are completely spinning, providing a reliable longitudinal motion benchmark for subsequent slip rate calculation and condition classification, fundamentally solving the technical problem of kinematic estimation failure under slippage conditions.

[0010] Existing solutions integrating multiple external sensors place stringent demands on the computing power of the main control chip for multi-sensor data alignment, heterogeneous signal synchronization processing, and complex fusion algorithms (such as particle filtering and deep network inference). Step three of this invention employs a zero-order TSK fuzzy inference machine, using 17 lookup table-based production rules to execute condition decisions, completely avoiding heavy computations such as iterative optimization or matrix inversion. On a 100MHz-level embedded ARM Cortex-M series chip, the entire process takes less than 0.9ms for a single update and can operate stably at a 200Hz control cycle. This means that this invention can be directly deployed on existing mainstream controllers for wheeled robots without requiring additional hardware upgrades, further reducing system modification costs.

[0011] This invention fully reuses the inherent parameters (speed, position, temperature) periodically transmitted by the onboard bus servo motor, requiring no modification to the servo motor firmware or customization of the communication protocol. For wheeled robot platforms that already use bus servos (such as TTL, RS485, CAN interface), functional upgrades can be achieved simply by burning the algorithm firmware onto the main control terminal, truly achieving "pure software deployment, zero hardware modification." This characteristic gives this invention extremely strong platform migration capabilities, allowing for rapid deployment to various heterogeneous wheeled robot systems.

[0012] The abnormal increase rate of joint torque in step two of this invention The joint torque value estimated based on position error and the temperature feedback are monitored and jointly alerted in parallel. It not only monitors abnormal resistance torque after the position tracking is in place, but also introduces a temperature forced reset mechanism—when the temperature of any joint servo exceeds the safety threshold, this feature is unconditionally set to its maximum value, directly triggering the joint overload protection rule in step three. This design integrates "force and heat" dual-dimensional information, covering both rapid detection of sudden mechanical jamming and slow heat accumulation caused by prolonged heavy loads, forming a safety mechanism with millisecond-level electrical protection and second-level thermal protection layered response, effectively preventing servo burnout or gearbox damage. Attached Figure Description

[0013] Figure 1This is a schematic diagram of the overall structure of a wheel-legged robot.

[0014] Figure 2 This is a flowchart of a method for real-time identification of the working conditions of wheeled robots based on the fusion of bus servo motors and IMU.

[0015] Figure 3 This is a schematic diagram of the drive control principle of a wheeled robot.

[0016] The attached diagram is labeled as follows: 1 is HTD-85H servo motor No. 1; 2 is HTD-85H servo motor No. 2; 3 is MicroSnow ST3215 servo motor; 4 is Intel N97 mini PC; 5 is a slave device with STM32F446RCT6 as the main control chip. Detailed Implementation

[0017] The invention will now be further described with reference to the accompanying drawings.

[0018] Reference Figures 1-3 This invention designs a wheel-legged robot structure where each leg is driven by three bus servos. HTD-85H servos 1 and 2 are responsible for the leg's planar movement, while a MicroSnow ST3215 servo 3 drives the wheel at the foot. An Intel N97 mini-PC 4 acts as the host computer for system calculations; a lower-level computer 5, with an STM32F446RCT6 as its main control chip, parses the host computer's instructions and sends servo drive commands or reads the servo's current parameters. The proposed wheel-legged robot working condition identification method based on bus servo and IMU fusion, without any external force sensors, contact sensors, or wheel-end encoders, uses only the periodically transmitted speed, position, and temperature data from the bus servos, and the acceleration, angular velocity, and attitude angle output from the onboard IMU to achieve real-time identification of eight working conditions: slippage, uphill, downhill, obstacle crossing, joint overload, lateral slippage, severe bumps, and normal flat ground.

[0019] A real-time identification method for the working conditions of a wheeled robot based on the fusion of bus servo motor and IMU includes the following steps: Step 1: Fusion of vehicle speed estimation and wheel slip ratio calculation based on rolling time-domain Kalman filtering wheel speed IMU; The longitudinal dynamics of the wheeled robot are simplified by treating the vehicle velocity as a slow variable, and a discrete state-space model is established. The state vector is defined as follows: ; in, For actual longitudinal vehicle speed, The system state is represented by a slow-to-zero error in the IMU forward acceleration channel (including installation error and sensor drift); a uniform-biased random walk model is used for the system state. ; in, The horizontal forward acceleration is calculated after gravitational acceleration and attitude compensation (from the IMU preprocessing module). The algorithm cycle time is 5ms. For process noise, its covariance matrix Characterizes the uncertainty of the rate of change of velocity and the zero-bias drift.

[0020] The observation equation uses the average flange linear velocity of all driving wheels as the observed quantity: ; in, For the first The rotational speed transmitted back by the drive servo motor Let be the effective radius of the wheel. The observation model is: ; in, To observe the noise (mainly caused by wheel slippage leading to deviations in wheel speed and vehicle speed), its covariance... Adaptive adjustments are made based on operating conditions; to avoid historical slippage data contaminating speed estimation, an exponentially decaying weighted rolling time-domain strategy is adopted. This is executed in each control cycle: First, the time update: ; ; In fact, it's adaptive observation noise adjustment: calculating the current slip rate indication. ; like If the threshold is exceeded, the increase is exponential. This reduces the reliability of observations on slipping wheel speed.

[0021] Finally, update the measurements: ; ; ; Calculate the slip ratio of each drive wheel using the estimated vehicle speed: ; When the robot is in wheeled mode, this formula reliably reflects the relative slippage between the wheels and the ground; when the robot is in legged mode, this slip ratio can be interpreted as an "ineffective propulsion ratio", which, after normalization, becomes the basis for the global slip ratio characteristics required for subsequent steps.

[0022] The HTD85H bus servo selected in this embodiment supports the transmission of angle, temperature, and voltage, but does not support the transmission of current and PWM commands; under the condition of no sensor, the joint load torque The calculation is based on the mapping relationship between servo position tracking error and joint equivalent stiffness.

[0023] The servo motor drivetrain (motor rotor → reducer → output shaft → load) is not an ideal rigid body. Under the action of load torque, it will undergo elastic torsional deformation, which macroscopically manifests as the actual angle of the output shaft. From the perspective of the target A deviation occurs. When the servo motor's closed-loop control is stable and has not reached its limit output, this deviation is approximately linearly related to the external load torque, i.e.: ; in, Position tracking error (unit: degrees). ; The equivalent stiffness coefficient of the joint (Nm / °); The fixed-end value of the joint load torque (Nm); The offline calibration of the equivalent stiffness coefficient was performed as follows: the robot was fixed to a horizontal platform, the servo motor of the joint under test was in closed-loop position holding mode, and the joint output shaft was placed in a horizontal orientation to eliminate gravitational interference; a known load torque was applied stepwise to the joint output end through a pulley-weight device. The load range covers from no-load to 80% of the stall torque; under each load level, wait for the servo to reach steady state (position error fluctuates less than 0.1° within 1 second), and record the steady-state position error. For all data points Perform least-squares linear fitting: ; The summation range is m = 1, 2, ..., M, where M represents the number of load levels under offline loading. The resulting slope is the equivalent stiffness coefficient of the joint. If the calibration curve exhibits significant nonlinearity near the stall zone, the operating range can be divided into low-load and medium-to-high-load segments, and the corresponding piecewise stiffness coefficient can be independently fitted to each segment. and Online estimation is based on the current error magnitude. Used when switching between zones.

[0024] For each control cycle, the following online estimation process is performed for each joint servo: First, the actual angle is parsed from the data packet returned from the bus servo. The target angle given by the current motion planner The position tracking error is obtained by subtracting the values. Set dead zone threshold (Take 0.5°, approximately 2-3 times the servo motor's no-load following noise), if Determine if the joint is in an unloaded or very lightly loaded state, and directly assign... And skip subsequent calculations; if Calculate the original torque estimate and apply saturation constraints ,in The stall torque is set for the servo motor nameplate; simultaneously, when the robot is moving at a low, constant speed on flat ground, the "Normal Flat Ground" condition label output in step three is used to automatically identify the light load window, and... The no-load torque benchmark is updated using an exponential moving average. The range of values ​​is 0 < <1, This embodiment takes The corresponding time constant This allows for a balance between tracking slowly changing factors and suppressing short-term noise.

[0025] Step 2: Constructing multi-dimensional operating condition feature vectors based on the servo IMU physical deviation model; From the kinematic quantities provided in Step 1, the joint load torque estimated in this step, and the temperature data transmitted back from the bus servo, six independent and distinguishable normalized features are extracted to form the feature space of this algorithm. Global slip ratio feature This is used to capture the overall wheel-to-ground slippage. The maximum absolute value of the slippage rate of all drive wheels is taken and obtained through non-linear mapping: ; in, The slip sensitivity threshold is set to 0.15 here. When both wheels experience severe free spin, Close to 1.

[0026] Pitch torque correlation characteristics This value reflects terrain undulations and obstacle-crossing dynamics. A higher value indicates a stronger consistency between torque changes and pitch motion, corresponding to climbing slopes or overcoming obstacles. The pitch angle is obtained from the IMU attitude calculation. Calculate its first difference The sum of the torques of all driving servo motors is denoted as ,in Estimated by the relationship between position error and equivalent stiffness (i.e. , For joint position tracking error, (Equivalent stiffness coefficients calibrated offline). At a length of... (correspond The Pearson correlation coefficient is calculated within a sliding window (whose length balances the statistical stability of the correlation coefficient and the dynamic response requirements to terrain changes). ; in, and These are the average of the difference in pitch angle within the window and the sum of the driving torque, respectively. and For the corresponding standard deviation, It is a very small positive number to prevent division by zero.

[0027] Vertical impact intensity characteristics Detect collisions with steps and impacts from deep pits. Calculate vertical acceleration. The short-time energy of k (20-point RMS) and long-term background energy (Ratio of 200 points RMS): ; The short-time and long-time energies are defined as follows: , ; Characteristics of abnormal increase rate of joint torque Real-time monitoring of overload or jamming of the servo motors in each joint. For the joints... When its position error satisfies (e.g., 2°) but actual torque Continuously exceeding no-load torque At that time, calculate the normalized overload exponent: ; It should be noted that this embodiment only focuses on the impact of load torque amplitude on joint safety; therefore, the absolute value of the torque is used when calculating the overload index. If practical applications require distinguishing between positive and negative loads (e.g., distinguishing between uphill and downhill braking conditions), the load can be... replace To redefine the corresponding no-load reference.

[0028] Take the maximum value for all joints and add the temperature penalty factor: ; in, This is an indicator function; if the temperature of any joint exceeds the warning value, then... Set it directly to 1 to force overload protection to be triggered.

[0029] Roll lateral slip anomaly characteristics Detect lateral slippage. Utilize the speed difference between the left and right drive wheels and the wheel track. Calculate the theoretical yaw rate: ; yaw rate measured by IMU Residuals are formed: ; When sideslip occurs, the wheel rotation cannot be fully converted into yaw, resulting in an abnormally large residual difference.

[0030] Vibration spectrum roughness characteristics To evaluate ground texture, an FFT was performed on a sliding window of vertical acceleration (128 points) to extract the energy percentage in the 50–100 Hz frequency band. ; in the denominator It represents the sum of the energies of all spectral lines in the frequency range from 0 to Nyquist.

[0031] It should be pointed out that features Taking the absolute value results in the loss of positive and negative information, and the distinction between uphill and downhill depends on the consistency of the direction of torque change and pitch angle change. Therefore, in this invention, the determination of downhill conditions does not directly use... Instead, it is done through the following two internal variables: (1) The first difference of the pitch angle is negative. (i.e., nose tilt); (2) the short-term average of the sum of driving torques is lower than the nominal torque on the flat surface. This decision logic is defined in fuzzy inference rule R15 as " The phrase "high (negative correlation)" actually means that the original correlation coefficient is negative and the absolute value reaches a high membership degree. This bivariate determination is independent of the feature vector and serves as an additional condition for rule activation.

[0032] Step 3: Multi-condition integrated decision and confidence assessment based on TSK fuzzy inference; The feature vectors generated in step two are mapped to specific robot operating condition labels, and confidence levels and auxiliary control parameters are given to achieve an interpretable and adjustable final decision.

[0033] Input fuzzification and membership function definition: for feature vectors For each component, three fuzzy sets are defined: low (L), medium (M), and high (H). To ensure smooth inference and reduce computational overhead, all membership functions adopt a combination of isosceles trapezoids and triangles, with specific parameters determined according to their respective physical distribution ranges.

[0034] Table 1 is a comparison table of fuzzy control inference rules;

[0035] Among them, R15 uses the combination of conditions that the first-order difference of the pitch angle is negative and the sum of the driving torques is lower than the nominal torque at ground level to determine the downhill condition; R10 is a mandatory safety rule: once High membership Or the temperature of any joint exceeds the warning threshold If it does not, it will directly output "Joint overload" and will no longer participate in the competition of other rules.

[0036] (2) Inference strength calculation, defuzzification and confidence generation: In each control cycle, the inference engine performs the following steps: ① Calculate the membership degree of the six-dimensional feature vectors on their respective L, M, and H fuzzy sets to form a membership degree matrix.

[0037] ② Calculate the activation strength of each rule individually, using the product implication operator: ; For dimensions marked "--" in the premise, their membership degree is 1.

[0038] ③ Perform overload priority detection: If or any If the current condition is not found, the current working condition will be set to "joint overload" with a confidence level of 1.0, and the subsequent process will be terminated.

[0039] ④ For the remaining working conditions c=1,…,8, calculate the representative strength as the maximum value of all regular strengths belonging to this class: ; ⑤ Output decision-making scenarios using a winner-takes-all strategy: ; ⑥ Calculate the normalized confidence level: ; in, Prevent division by zero. Confidence level reflects the significance of the current decision relative to other competing conditions. When When the value is below a preset threshold (e.g., 0.6), the upper-level controller can maintain the output of the previous cycle's operating condition to suppress false switching caused by transient interference. Furthermore, a median filter of 3 frames (15ms) can be applied to the output operating condition code to reduce the frequency of operating condition transitions with almost no increase in latency.

[0040] (3) Structured output of auxiliary control parameters, including auxiliary quantification parameters output synchronously with the operating condition label: Estimated climb angle: Based on the driving torque balance relationship ; in, For the total mass of the robot, It is the acceleration due to gravity. The effective radius of the wheel, Torque calibration for constant speed driving on flat ground; index of the most severe slippage and its slip ratio: and Overload joint indicator, current torque value, over-limit ratio, and joint temperature; lateral slip severity directly corresponds to the characteristic. The original value.

[0041] The above parameters, along with the working condition label, confidence level, and timestamp, are packaged into a unified data frame and sent to the motion control node at a frequency of 200Hz via serial port or shared memory space, directly participating in mode switching and joint-level command compensation.

[0042] The embodiments described in this specification are merely examples of implementations of the inventive concept and are for illustrative purposes only. The scope of protection of this invention should not be considered limited to the specific forms described in these embodiments; rather, it extends to equivalent technical means conceived by those skilled in the art based on the inventive concept.

Claims

1. A method for real-time identification of the working conditions of a wheeled robot based on the fusion of bus servo motor and IMU, characterized in that, The method includes the following steps: Step 1: Vehicle speed estimation and wheel slip ratio calculation based on adaptive disturbance rejection Kalman filter: A longitudinal vehicle speed estimator is established, which is jointly driven by IMU horizontal acceleration integral and bus servo wheel speed feedback. An adaptive Kalman filter with time-varying observation noise covariance is adopted. The average linear velocity of the driving wheel is used as the observation and IMU acceleration integral is used as the process prior. The sensor zero bias is dynamically compensated and the contamination of speed estimation by slip conditions is suppressed. The slip ratio of each wheel is calculated in real time. Step 2: Construction of multi-dimensional operating condition feature vector based on servo-IMU physical deviation model: Extract six interpretable and non-redundant normalized features from the speed, position, and temperature returned by the bus servo and the attitude, acceleration, and angular velocity output by the IMU. These features include global slip rate feature, pitch-torque correlation feature, vertical impact intensity feature, joint torque abnormal rise rate feature, roll-lateral slip anomaly feature, and vibration spectrum roughness feature. Step 3: Multi-condition integrated decision and confidence assessment based on TSK fuzzy inference: Input the six-dimensional feature vector generated in Step 2 into the zero-order TSK fuzzy inference engine, and use the preset expert rules to perform integrated mapping on eight conditions including slippage, uphill, downhill, step obstacle crossing, joint overload, lateral slippage, severe bumps and normal flat ground, and output the current condition label, confidence level and auxiliary control parameters.

2. The real-time identification method for the working condition of a wheeled robot based on the fusion of bus servo motor and IMU as described in claim 1, characterized in that, In step one, the observation noise covariance of the adaptive Kalman filter is dynamically adjusted based on the current slip rate indication, and the slip rate indication is calculated: ； like If the observed noise covariance exceeds a preset threshold, it increases exponentially. This reduces the reliability of wheel speed observations during slippage.

3. The real-time identification method for the working condition of a wheeled robot based on the fusion of bus servo motor and IMU as described in claim 1 or 2, characterized in that, In step two, the joint load torque is indirectly estimated based on the linear mapping relationship between the servo motor position tracking error and the joint equivalent stiffness coefficient: ； in, For position tracking error, The equivalent stiffness coefficient is calibrated offline; and a dead zone threshold is set. ,when When the load is determined to be no-load, the torque is set to zero.

4. The real-time identification method for the working condition of a wheeled robot based on the fusion of bus servo motor and IMU as described in claim 3, characterized in that, The equivalent stiffness coefficient The position was obtained through offline calibration. The robot was fixed to a horizontal platform, and the tested joint servo motor was in closed-loop position holding mode. Known load torques were applied stepwise to the joint output end, and the steady-state position error was recorded. Least square linear fitting was performed on the load torque and position error data points to obtain the position. The working range is divided into low-load and medium-high-load segments based on the degree of nonlinearity, and the corresponding segment stiffness coefficients are fitted respectively.

5. The real-time identification method for the working condition of a wheeled robot based on the fusion of bus servo motor and IMU as described in claim 1 or 2, characterized in that, In step two, the global slip ratio characteristic The calculation formula is: ； in, For the slip rate of each drive wheel, This is the slip sensitivity threshold; The pitch-torque correlation characteristic is the absolute value of the Pearson correlation coefficient, which is calculated within a sliding window as the sum of the pitch angle difference and the drive torque. The vertical impact intensity characteristic is the ratio of short-time vertical acceleration energy to long-time background energy. This is a characteristic of abnormal increase rate of joint torque, i.e., the maximum value of the normalized overload index of each joint, plus a temperature penalty factor. When the temperature of any joint exceeds the warning value... Set it directly to 1; The roll-lateral slip anomaly characteristic is the residual between the theoretical yaw rate calculated based on the speed difference between the left and right wheels and the yaw rate measured by the IMU, which is then mapped by hyperbolic tangent. This represents the roughness characteristics of the vibration spectrum, specifically the energy percentage in the 50-100Hz frequency band after the vertical acceleration sliding window FFT.

6. The real-time identification method for the working condition of a wheeled robot based on the fusion of bus servo motor and IMU as described in claim 5, characterized in that, In step three, the TSK fuzzy inference engine employs 17 expert rules. The premise of each rule corresponds to the low, medium, and high fuzzy set membership degree of the six-dimensional feature vector, while the conclusion corresponds to the operating condition category. Rule activation strength is calculated using the product implication operator, and a winner-takes-all strategy is used to output the decision operating condition, along with a normalized confidence score. An overload priority detection rule is set: if... If the membership degree of a joint exceeds the threshold or the temperature of any joint exceeds the warning threshold, the "joint overload" condition will be output directly with a confidence level of 1.

0.

7. The real-time identification method for the working condition of a wheeled robot based on the fusion of bus servo motor and IMU as described in claim 6, characterized in that, The expert rules include: Rule R1: For H, Let L be the number of L. L → Soft and slippery; Rule R3: Let L be the number of L. For H, L → Uphill; Rule R15: Let L be the number of L. For H, For L, combined with the first-order difference of the pitch angle being negative and the sum of the driving torques being lower than the nominal torque on the flat ground → downhill; Rule R10: H → Joint overload; Rule R13: All features are low values ​​→ Normal flat land.

8. The real-time identification method for the working condition of a wheeled robot based on the fusion of bus servo motor and IMU as described in claim 1 or 2, characterized in that, In step three, the auxiliary control parameters include the estimated climbing angle, the index of the most severe slipping wheel and its slip rate, the overload joint indicator and its current torque value and temperature, and the severity of lateral slippage; the estimated climbing angle is based on the drive torque balance relationship. ； in, For the total mass of the robot, The effective radius of the wheel, The torque is calibrated for constant speed driving on flat ground.

Images