A method for sand power control of a two-wheeled vehicle

Abstract

The application discloses a two-wheeled vehicle sand power control method. A controller forms a current authorized torque Tauth according to driving requirements or a verified propulsion authorization, and makes all targets and outputs not exceed Tauth; longitudinal acceleration and sinking speed responses are identified in the same complete observation window, and are synthesized with an offline envelope covering a limited torque increment domain. Only candidates that match a unique resistance calibration unit, a future increment resistance one-side lower tolerance limit, a non-negative and one-side upper tolerance limit, are tolerated, and a sinking margin and a traction margin formed by a candidate driving force lower limit and a future resistance upper limit are respectively constructed to form a feasible set Omega, and then a planning time domain reachable set, a phase resolution set and a planned point reachable screening are selected to select a thrust target, and non-negative bridging, maintaining, unloading and observation are executed. A preset zero torque event set E0 is zeroed in a current period, and a torque reduction event set ED is safely zeroed according to Tauth and an event upper limit.

Description

Technical Field

[0001] This invention relates to the fields of vehicle power control, wheel-ground interaction state identification, and intelligent control technology for off-road vehicles. In particular, it relates to a control method for constructing a feasible torque set on soft sand by using the longitudinal acceleration response and sinking velocity response of the same observation window, and implementing unidirectional phase-separated power output in the non-negative drive domain. Background Technology

[0002] When a vehicle starts, escapes from a stuck situation at low speed, or continues to move forward on soft sand, the increased torque of the drive wheels can improve the tangential force between the wheels and the ground, but it may also exacerbate shearing damage, digging, and sinking. Tire load, tire pressure, sand particle size distribution, moisture content, and existing ruts all change the torque-acceleration relationship and the torque-sinking relationship, and a single fixed slip ratio or fixed torque threshold is unlikely to cover these changes.

[0003] Existing feedback control systems often use a single response from acceleration, slip ratio, or sinking. Acceleration reflects the net propulsion effect but cannot distinguish between driving force and the increasing drag with sinking; sinking reflects the risk of sinking but cannot directly prove that the torque is sufficient to overcome the drag. If the net acceleration response is directly used as the driving force gain, or if the upper limit of drag is derived from the lower limit of the reference driving force, it is easy to cause errors in the boundary direction.

[0004] In addition, low-speed operation is prone to occur in sandy areas. Numerical instability, zero-reference torque causing relative error division to zero, an empty feasible set, detection period interrupted by a safety event, and identification boundaries exceeding their validity period can all lead to problems. If the target feasible set after the response phase ends is directly filtered using only the set of changes in a single sampling period, the controller may misjudge an empty set when the target feasible interval is more than the distance from the reference torque in a single step, even though it could be reached through multi-step safe bridging. Therefore, it is necessary to clearly distinguish between target feasibility, time-domain reachability planning, and the sample-by-sample bridging process, and to provide definite branches for computational hold, computational timeout, and invalid observation.

[0005] Chinese patent applications CN114655028A and CN114407673A disclose a control method for inter-axle torque transfer in sand driving mode based on wheel speed and driving state characterized by yaw angle and lateral and longitudinal acceleration; CN102880051A discloses a control method for soft ground that uses slip ratio as a state variable and adjusts wheel torque through fuzzy sliding mode to indirectly suppress slippage and sinking. However, none of these solutions provide a mechanism that integrates paired identification of dual responses, finite-domain offline envelope, tolerance limit under non-negative future incremental resistance, dual-margin set, phase-separated reachability screening, and graded safety recovery into a point-by-point executable closed loop within the authorized torque range. Summary of the Invention

[0006] The technical problem this invention aims to solve is that, on soft sand, the net acceleration response of a two-wheeled vehicle cannot be directly equated to the driving force gain, nor can propulsion capability be proven solely by the amount of subsidence. Furthermore, low-speed measurements, finite calibration domains, state changes during bridging, the time domain of the hold-up evaluation, empty sets, and safety events can all invalidate the calculation results. To address this, this invention provides a control chain from authorized and measurement inputs, through peer-to-peer identification, offline envelope and drag tolerance limit verification, to feasible sets, phase-by-phase reachability targets, actual hold-up evaluation, and safety output.

[0007] This invention provides a two-wheeled vehicle dynamic control method for sand terrain, comprising the following steps executed by the vehicle's main controller: acquiring the current authorized torque upper limit, the actual driving torque of the controlled wheel set, longitudinal acceleration, and subsidence measurement values; determining the upper limit of subsidence depth and the upper limit of subsidence velocity based on the subsidence measurement values; performing reference acquisition, torque excitation, and response acquisition within the same observation time window; determining the conservative lower limit of longitudinal acceleration response and the conservative upper limit of subsidence velocity response based on the changes in the actual driving torque of the controlled wheel set relative to the reference torque, and the changes in longitudinal acceleration and subsidence velocity relative to their respective reference values, and determining the lower limit of reference driving force and the upper limit of reference resistance; generating candidate torques between the reference torque and the current authorized torque upper limit, and determining the conservative lower limit of subsidence depth, the upper limit of subsidence velocity, and the conservative upper limit of subsidence velocity response based on the aforementioned upper limit of subsidence depth, upper limit of subsidence velocity, and upper limit of subsidence velocity response. The predicted subsidence margin is determined by the upper bound and the increment of candidate torques relative to the reference torque. The predicted traction margin is determined by the conservative lower bound of the longitudinal acceleration response, the lower bound of the reference driving force, the upper bound of the reference resistance, and the pre-calibrated future resistance increment boundary with a non-negative lower bound and a finite upper bound. Candidate torques whose predicted subsidence margin and predicted traction margin are both not lower than their respective preset margin thresholds are combined into a feasible torque set. Based on the current torque command, the preset torque increase rate, the preset torque decrease rate, and the planned duration, a target torque that can be reached within the planned duration is selected from the feasible torque set. The torque command is then transitioned to the target torque and maintained. After the actual driving torque reaches the target torque, the target torque is re-verified based on the updated subsidence measurement value. If the re-verification fails, the torque is reduced.

[0008] Preferably, determining the predicted subsidence margin includes: determining the upper bound of the predicted subsidence depth corresponding to the candidate torque based on the upper bound of the subsidence depth, the upper bound of the subsidence velocity, the increment of the candidate torque relative to the reference torque, the conservative upper bound of the subsidence velocity response, and the preset subsidence prediction time, and determining the predicted subsidence margin by the difference between the critical subsidence amount and the upper bound of the predicted subsidence depth.

[0009] Preferably, determining the predicted traction margin includes: under the condition that the pre-calibrated lower bound of the future resistance increment is not less than zero, determining the lower bound of the candidate longitudinal driving force based on the lower bound of the reference driving force, the increment of the candidate torque relative to the reference torque, the conservative lower bound of the vehicle equivalent mass, and the conservative lower bound of the longitudinal acceleration response; determining the upper bound of the candidate future resistance based on the upper bound of the reference resistance and the pre-calibrated upper bound of the future resistance increment; and determining the predicted traction margin by the difference between the lower bound of the candidate longitudinal driving force and the upper bound of the candidate future resistance.

[0010] Preferably, selecting the target torque from the feasible torque set includes: determining the achievable torque range based on the current torque command at the start of bridging, the preset torque increase rate, the preset torque decrease rate, and the planned duration, and selecting the target torque from the intersection of the achievable torque range and the feasible torque set; and transitioning the torque command to the target torque under the condition that both the target torque and the torque command are non-negative and do not exceed the current authorized torque limit.

[0011] Preferably, the recalibration of the target torque includes: after the actual driving torque enters the target torque allowable range, recalculating the predicted settlement margin and predicted traction margin using the updated upper bound of settlement depth and upper bound of settlement speed; maintaining the target torque and subsequently performing unloading and observation when both predicted margins are still not lower than their respective preset margin thresholds; and reducing the torque when either predicted margin does not meet the corresponding preset margin threshold.

[0012] Preferably, the step of performing benchmark acquisition, torque excitation, and response acquisition within the same observation time window includes: determining a benchmark reference torque before the start of the observation time window; determining the benchmark torque, benchmark longitudinal acceleration, and benchmark sinking speed based on the median of the effective actual driving torque, longitudinal acceleration, and sinking speed of the controlled wheel set during the benchmark acquisition phase; and maintaining these three values ​​until the end of the observation time window; the target torque during the torque excitation phase is higher than the benchmark reference torque; and the target torque during the response acquisition phase returns to the benchmark reference torque.

[0013] Preferably, the determination of the conservative lower bound of the longitudinal acceleration response and the conservative upper bound of the sinking velocity response includes: resampling the actual driving torque, longitudinal acceleration, and sinking velocity of the controlled wheel assembly onto the same time axis, without interpolating missing or invalid sampling points, and using the change of the actual driving torque of the controlled wheel assembly relative to the reference torque and the changes of the longitudinal acceleration and sinking velocity relative to their respective reference values ​​to determine the conservative lower bound of the longitudinal acceleration response and the conservative upper bound of the sinking velocity response, respectively.

[0014] The beneficial effects of this invention are: the controller outputs candidate torque only when the current authorization, measurement quality, boundary version, and applicable domain are all valid; the dual response limits the risks of under-propulsion and sinking, respectively; This is to ensure that the net acceleration increment is not mistakenly taken as an excessively large driving force increment. The potential increase in driving resistance under the candidate state is then included in the upper bound of future resistance, so that the traction margin is calculated in the same conservative direction as "lower bound of candidate driving force minus upper bound of future resistance". Time-based re-verification and overwriting The continued consistency ensures that the prediction time domain aligns with the actual execution history; Locking zeros Limited torque reduction provides a defined failure path. Attached Figure Description

[0015] Figure 1 This is a complete control closed-loop flowchart of one embodiment of the present invention. Detailed Implementation

[0016] The method of this invention sequentially performs authorization and synchronization, benchmark-excitation-response, dual-response identification within the same window, online-offline boundary synthesis, multi-dimensional resistance element, and verification of the lower and upper tolerance limits of the future resistance increment on one side. structure, intersect, Phase resolution, Point-by-point reachability filtering, target bridging and actual retention, unloading, observation and safe recovery. Constrain all stage objectives and outputs; when the candidate actual torque completes in-band confirmation, re-verify using the fresh settlement state. Future resistance tolerance limit and This avoids ignoring state changes during bridging. Specific steps are described in detail in the implementation details below.

[0017] The following examples further illustrate the input, calculation order, boundary conditions, state transitions, and related parameters of the present invention. (Subscript) Indicates a conservative lower bound, subscript This indicates a conservative upper bound; the subsidence depth and velocity are positive with respect to the interior of the undisturbed terrain surface, and the longitudinal driving direction is also positive. This applies when the propulsion authorization function is not configured or is explicitly inactive via the state machine. When the authorization function is activated and the authorization value passes the permission, freshness, and integrity checks, The authorization value for advancing the pilgrimage permission verification; if the advancement authorization function is activated but the authorization is invalid, it is not allowed to roll back to the previous state. And directly classified into Then enter SAFE_ZERO, where invalid authorization calculations are not allowed. For valid driver authorization . Determine before the observation window begins; , and Freeze after the reference phase ends and use it for the complete window. Measure the torque at each electric drive wheel torque measurement position. The equivalent total actual torque of the controlled wheel assembly is obtained by summing the results after calculating the equivalent torque of the same motor according to a pre-defined torque measurement. ; , , , and All use this torque caliber. Meanwhile, for the first... Each electrically driven wheel, with the wheel effective within the reference phase. Determining the median of the sample This is used for wheel-by-wheel conversion of the lower bound of the reference driving force; the actual torque of each wheel in the reference synchronization sample is denoted as... This is used for wheel-by-wheel conversion of the upper bound of the reference driving force. When converting from this equivalent lateral wheel-end driving force, the total transmission ratio, efficiency, and effective radius boundary of the wheel-end torque for the corresponding wheel are used separately; the wheel-end force is not calculated by directly adding the unconverted motor torques under different transmission ratios. Safety arbitration is performed first in each sampling period: Output zero directly. Perform safety torque reduction; the main controller only executes normal bridging, holding, unloading, or observation commands when the safety channel is not under control.

[0018] Taking the example of a vehicle advancing at low speed on soft sand, the driver requests a positive torque. While increased torque is beneficial for propulsion, it may exacerbate wheel-to-ground shear and sinking. Longitudinal acceleration only reflects the net propulsion effect and cannot directly distinguish between changes in driving force and driving resistance. Furthermore, the sinking state alone cannot prove that the torque is sufficient to overcome the resistance. Therefore, it is necessary to simultaneously assess propulsion capability and sinking risk within the current authorized range, and to handle bridging, changes in state during the holding period, and safety events.

[0019] After the control cycle begins, the main controller acquires the driving demand torque, propulsion authorization, vehicle safety torque limit, actual driving torque, longitudinal acceleration, and subsidence measurement values. If the propulsion authorization function is activated but the authorization is invalid, zero torque output is directly performed; otherwise, the current authorized torque limit is determined. Reference acquisition, torque excitation, and response acquisition are completed within the same complete observation window. The conservative lower bound of the longitudinal acceleration response and the conservative upper bound of the subsidence velocity response are identified based on the actual torque change, and synthesized with the offline conservative envelope within the finite torque increment domain. Candidates are generated between the reference torque and the current authorized torque limit. Each candidate participates in the calculation only when it uniquely matches an effective resistance calibration unit, the lower tolerance limit of the future resistance increment is non-negative, and the upper tolerance limit is finite. The subsidence margin is predicted based on the subsidence state and subsidence velocity response, and the traction margin is predicted based on the reference driving force, longitudinal acceleration response, and candidate future resistance upper bound. Only candidates whose margins both reach the threshold are retained. In this embodiment, only 90 N·m out of 70, 80, 90, and 100 N·m is included in subsequent planning. The controller then performs reachability screening based on the current torque, torque increase / decrease rate, and planned duration, and gradually bridges to the target. After the actual torque is confirmed within the band, it recalculates two margins using the fresh sinking state and the still valid boundary. If it passes, it maintains and unloads and observes in sequence. If it fails, the boundary expires, or a safety event occurs, it executes limited torque reduction or zero torque safety output.

[0020] like Figure 1 As shown, in S101, the main controller collects the current authorized torque upper limit, the actual driving torque of the controlled wheel set, longitudinal acceleration, and settlement measurement values, and determines the upper limit of settlement depth and settlement velocity; in S102, within the same observation time window, the reference acquisition, torque excitation, and response acquisition are completed, and the conservative lower limit of longitudinal acceleration response, the conservative upper limit of settlement velocity response, the lower limit of reference driving force, and the upper limit of reference resistance are determined based on the actual torque changes; in S103, candidate torques are screened based on the predicted settlement margin and predicted traction margin, and a feasible torque set is constructed; in S1 04. Select the achievable target torque based on the current torque command, preset torque increase rate, preset torque decrease rate, and planned duration, and perform torque transition; in S105, update the settlement measurement value after the actual driving torque reaches the target torque; in S106, recalculate the predicted settlement margin and predicted traction margin using the updated upper bound of settlement depth and upper bound of settlement speed, and determine whether the re-verification passes; if it fails, proceed to S107 to reduce torque; if it passes, proceed to S108 to maintain the target torque and then perform unloading and observation, and then proceed to the next control cycle.

[0021] This forms a closed loop of "authorization constraints—dual-response identification—dual-margin feasibility assessment—planned reachability and bridging—re-verification after actual arrival—maintaining, unloading, and observing—safe recovery." This scheme uses two types of responses to constrain insufficient propulsion and subsidence risk, avoiding the misinterpretation of net acceleration as driving force gain; it utilizes lower and upper tolerance limits for future resistance increments to maintain a conservative direction for driving force derivation and traction margin; it avoids misjudging multi-step reachable targets as empty sets through planned reachability and point-by-point bridging; and it handles state changes, boundary failures, and safety events during bridging through re-verification and safe output after arrival, thereby coordinating propulsion capability and subsidence risk control within the same closed loop.

[0022] Example 1: Signal, State, and Boundary Generation

[0023] The main symbols, definitions, and constraints are as follows (only core and frequently used symbols are listed; local symbols are defined in the corresponding paragraphs).

[0024] symbol ; ; The definition is: driving torque demand; maximum authorized drive or propulsion torque; maximum safe torque for the entire vehicle. Propulsion authorization is triggered directly when it is activated but the authorization is invalid. , No rollback The unit or value constraint is N·m; it is a non-negative value when it is valid.

[0025] symbol The definition is: the current authorized torque; formed by taking the smaller value between the effective driving authorized upper limit and the vehicle safety torque upper limit in a non-negative direction; the unit or value constraint is: N·m; .

[0026] symbol ; ; The definition is: command torque; the total measured torque of the controlled wheel set after conversion based on the equivalent side of the torque measurement of the same motor; reference torque; the unit or value constraint is: N·m; and by constraint.

[0027] symbol ; ; ; The definitions are: the reference torque for freezing, the actual reference torque, the acceleration reference, and the settlement velocity reference; the units or value constraints are: N·m, m / s², and mm / s, respectively.

[0028] symbol ; ; ; The definition is as follows: sampling period; normal torque increase, torque decrease, and safe torque decrease rates; unit or value constraint: s; N·m / s; rate average .

[0029] symbol ; ; ; The definition of is: the increment of actual torque, acceleration and sinking velocity relative to the reference on the same resampled axis; the effective dead zone of torque increment; the unit or value constraint is: N·m, m / s², mm / s, N·m respectively.

[0030] symbol ; The definition is: conservative upper bound for subsidence depth; conservative upper bound for subsidence velocity; unit or value constraint: mm; mm / s.

[0031] symbol ; ; ; ; The definition is as follows: a finite set of candidate delays and the maximum delay; the number of points required for response correlation; the number of points for baseline confirmation; the maximum number of points for response status; and the unit or value constraint is: sampling points. The rest are positive integers.

[0032] symbol ; ; ; ; The definition of delay is: The following parameters are required: all valid pairing indexes; valid excitation pairings satisfying the torque dead zone; response type; selected delay; number of valid regression pairs; unit or value constraints. ; It is a positive integer.

[0033] symbol ; ; ; ; ; ; ; ; The definition is: normalized cross-correlation coefficient, signal-to-noise ratio, and sample number / correlation peak / peak difference. Follow-up error threshold; follow-up error denominator lower limit and target band tolerance; unit or value constraints are: , Dimensionless; It is a positive integer; , N·m and The remaining thresholds are fixed in advance.

[0034] symbol ; ; ; ; The definition of is: response The regression slope through the origin, the sum of squared residuals, and the standard error of the regression slope estimate; the boundary coverage coefficient of the two-response residuals; the unit or value constraint is: the unit varies with the origin. Sure; , .

[0035] symbol ; The definition is as follows: longitudinal acceleration response is online conservative lower bound; subsidence velocity response is online conservative upper bound; the unit or value constraint is: (m / s²) / (N·m) and (mm / s) / (N·m) respectively.

[0036] symbol ; The definition is: a conservative lower bound for the offline longitudinal acceleration response and a conservative upper bound for the subsidence velocity response within a finite calibration domain; the unit or value constraint is: the same as the corresponding online response.

[0037] symbol ; The definition is: the final conservative boundary related to the candidate torque after synthesizing the online and offline boundaries in the unfavorable direction; the unit or value constraint is: ; And it is limited.

[0038] Symbol N; ; ; ; The definition is as follows: number of full window points; length of the looping block; number of blocks per window; number of valid pairwise resamples of the target; maximum number of attempts; unit or value constraint: positive integer; ; .

[0039] symbol ; ; The definition is: the joint confidence missing value and the missing value assigned to the two responses; the unit or value constraint is: .

[0040] symbol ; ; ; The definitions are: acceleration evaluation duration and number of points; subsidence prediction duration and number of points; unit or value constraint: s; positive integer; .

[0041] symbol ; ; The definition is: the conservative lower bound of the reference driving force; the first The conservative upper bound of the driving force of each reference synchronous sample; the conservative upper bound of the reference resistance; the unit or value constraint is: N.

[0042] symbol ; The definition is: conservative lower bound for candidate longitudinal driving force; conservative upper bound for candidate future resistance; unit or value constraint: N.

[0043] symbol ; ; ; ; The definition is as follows: the non-negative upper bound of the lower / upper bound of the residual error in the driving force conversion; the reference resistance model error; the upper bound of the model error in the direction of the lower / upper tolerance limit of future resistance; the unit or value constraint is: N; .

[0044] symbol ; ; The definition is: synchronous longitudinal driving force; driving resistance; future resistance increment. Among these, based on longitudinal dynamics... ,have ; The variable representing the equivalent mass of the vehicle, in kg, is used for traversal. ; This represents the longitudinal acceleration of a vehicle, measured in m / s². The term represents the inertia term formed by the product of the vehicle's equivalent mass and longitudinal acceleration, and its unit is N. , and The unit for all values ​​is N.

[0045] symbol ; ; The definition is: the lower / upper bound of the online vehicle equivalent mass; the equivalent mass that is kept constant during independent verification; the unit or value constraint is: kg; ; .

[0046] symbol ; The definition is: the offline response envelope torque increment domain and its upper limit; the unit or value constraint is: .

[0047] symbol ; ; The definition is: multidimensional resistance calibration unit (c is the unit index); future resistance increment unilateral lower / upper tolerance limit; unit or value constraint is: candidate participation hour It must be unique. , Limited and .

[0048] symbol ; ; ; ; ; The definition is as follows: probability of one-sided tail exceeding the limit of the resistance increment; lower / upper confidence missing value; corresponding ordinal statistical index; number of independent validation samples; unit or value constraints are: ; ; .

[0049] symbol ; ; ; ; The definition is: critical subsidence amount; candidate subsidence margin and threshold; candidate traction margin and threshold; the unit or value constraint is: subsidence amount is mm; traction margin is N; , , .

[0050] symbol ; ; The definition is: a set of feasible torques with double margins; the range of achievable torques in the planning time domain; the intersection of the two; and the unit or value constraint is: a set of N·m.

[0051] symbol ; ; ; ; The definition is: the initial set after phase interval filtering, the set after point-to-point reachability filtering, and three sets of minimum phase intervals; the unit or value constraint is: the set is N·m; the interval is N·m and all are .

[0052] symbol ; The definition of is: planning duration; the current command torque frozen at the start of bridging; the unit or value constraint is: s; N·m.

[0053] symbol ; ; ; The definition is: thrust, unloading, observed target and phase target corresponding to the current state; the unit or value constraint is: N·m; .

[0054] symbol ; ; ; ; ; The definition is: the budget of points for arrival / in-band confirmation, push hold, unloading, observation and actual tracking; the unit or value constraint is: positive integer.

[0055] symbol ; ; ; The definition is as follows: the set of events where the current period is set to zero; the set of events where torque reduction is allowed; the upper limit of the event torque; the minimum value of the upper limit of the current effective event torque; the unit or value constraint is: a set; the torque quantity is N·m and .

[0056] symbol ; The definition is: the number of health confirmation points after safe cleanup; the restart requirement threshold; the unit or value constraint is: It is a positive integer; N·m and .

[0057] symbol ; ; The definition is: the end time of the complete observation window; the completion time of BOUND_CHECK; the end time of confirmation within the actual torque continuous band; the unit or value constraint is: s or a unified clock timestamp.

[0058] symbol ; ; ; ; The definition is as follows: number of stable gated continuous points; number of reference steady-state points; number of detected steady-state points; number of detected target reaching the budget point; and incremental detected reference torque. The unit or value constraint is that the first four are positive integers. .

[0059] symbol ; The definition is: the upper limit of the rate of change of driving demand and the upper limit of the reference torque for detection gate control; the unit or value constraint is: pre-fixed in the current boundary version.

[0060] symbol ; ; The definition is: the actual duration of the current activity cycle from the start of BASELINE to the end of OBSERVE; the maximum allowed duration of recovery during the activity cycle and week; the unit or value constraint is: s; Both maximum allowed durations .

[0061] symbol ; The definition is: the maximum single-cycle decrease in subsidence depth and subsidence velocity after complete observation, with conservative upper bounds; the unit or value constraint is: mm; mm / s; average. .

[0062] To avoid ambiguity in the correspondence of terms and symbols, the abbreviations used for the same physical quantity in different paragraphs in this application do not change the meaning of the symbols: zU represents the same parameter as "conservative upper bound of subsidence depth" and "upper bound of subsidence depth"; Mz(T) represents the same parameter as "candidate subsidence margin" and "predicted subsidence margin", where "candidate" only indicates that the margin is calculated for candidate torque T. The same rule applies to "conservative upper bound of subsidence velocity / upper bound of subsidence velocity" and "candidate traction margin / predicted traction margin"; the notation with sampling time subscripts or candidate torque independent variables is only used to indicate the corresponding time or candidate value, and does not indicate that additional technical parameters are set.

[0063] Settlement measurements are used to establish upper limits for settlement depth and settlement velocity. The upper limit for settlement depth referred to in this application is... The upper limit of the subsidence velocity is Let the first... The effective subsidence estimate after time synchronization and quality verification at each control sampling time is: The actual subsidence depth is The upper bound of the non-negative finite absolute error corresponding to the measurement uncertainty is: and satisfy ,Pick Timestamps on a uniform resampling timeline are selected and verified to be spaced apart. The moment and ,calculate .in, A pre-fixed positive integer, To control the sampling period, the uncertainties in the subsidence measurement at two time points and the residual errors of the velocity measurement model not yet covered by them are combined in the direction that increases the subsidence velocity. and take . To subtract the reference settlement velocity obtained by differential measurement at the same time interval from the calibrated independent reference settlement displacement measurement device, under the same sensor, time synchronization, resampling, differential, and filtering configuration as the online device. The one-sided upper bound of the obtained residuals; this upper bound is pre-formed by the independent validation residuals according to the one-sided upper bound order statistical rules described below. Its tail out-of-bounds probability, confidence missing amount, sample valid domain, configuration, version and validity period are fixed before online use and saved with the current boundary version. It is invalid if any item does not match, there is no valid order statistical index or it expires. , and The unit is mm. , , and The unit is mm / s. Any settlement measurement, time synchronization relationship, error upper bound, or model residual upper bound is invalid, or When, corresponding or invalid.

[0064] The aforementioned "merging in the direction that increases the subsidence velocity" specifically refers to: within the uncertainty range of each of the two subsidence depth measurements used for differentiation, taking the larger unfavorable endpoint for the subsidence depth at the later time and the smaller unfavorable endpoint for the subsidence depth at the earlier time, thereby achieving a larger difference in subsidence depth between the two time points. Therefore, the contributions of the upper bound of the absolute error of the subsidence depth at the two time points to the upper bound of the differential velocity are added together and divided by the two time intervals that have been verified to be positive; the residual error of the velocity measurement model not yet covered by this item is then included in the direction that increases the upper bound of the subsidence velocity.

[0065] When the one-sided upper bound of the residual error of the aforementioned velocity measurement model is formed according to the statistical rules for the one-sided upper bound order described below, each valid independent verification residual obtained with the same sensor, synchronization, resampling, differential, and filtering configuration is used as a verification sample, and they are arranged in ascending order of residual value. The number of valid samples is used as the sample number for this rule, and the pre-fixed one-sided tail excess probability and confidence missing amount are substituted into the upper-side index determination. The smallest valid integer index that satisfies the upper-side index condition is selected, and the sorted residual at that index position is used as the one-sided upper bound. If there is no valid upper-side index, the sample domain or version does not match, or the record expires, the upper bound is invalid.

[0066] Indicates the first The actual torque of the motor corresponding to each electric drive wheel at a specified torque measurement position is first converted to the equivalent torque measurement side of the same motor according to a specified benchmark. When converting the converted torque into the longitudinal driving force at the wheel end, the total transmission speed ratio from the equivalent side to the wheel end must be used simultaneously. Transmission efficiency and wheel end torque effective radius If it is hub direct drive, then If the wheel-end torque is measured directly, the already included transmission ratio and efficiency should not be considered again when calculating the longitudinal force between the wheel and the ground from the measured wheel-end torque. Effective radius of wheel-end torque. Not subject to sandy conditions Instead of direct replacement, the lower and upper boundaries are formed by independent calibration data of tire specifications, load, tire pressure, temperature, and wheel-ground conditions. Pre-defined ;when The kinematic ratio can be calculated at that time. ,in The rigid body projection of the vehicle's longitudinal velocity at the wheel position is used. Used only for low-slip effective domain consistency verification, anomaly detection, or offline effective radius calibration unit selection; only independent verification proves that the mapping error within the corresponding low-slip domain has been corrected. It is only allowed to participate in cell selection when it is covered; it should never be used directly under any circumstances. Alternative Perform torque-driving force conversion on soft sand. Among other things, and All directions are defined by the direction of the rolling of the corresponding wheel when the vehicle is moving forward; At that time, the so-called The reverse sign means , equivalent to Low-slip valid field verification does not use... Alternatively, the calibration results can be used as an index for tire specifications, load, tire pressure, temperature, drive status, and vehicle speed range, uniquely matching the low-slip calibration unit that has not yet expired. This unit stores the data that satisfies... Validation interval .when The values ​​must be finite, fall within this range, and the relevant timestamps, units, coordinate systems, and sensor health status must be valid, confirmed by the unit's independent verification records. and The verification is considered passed when the mapping error between the two is covered by the effective radius boundary of the wheel end torque; otherwise, it is considered failed. The above verification interval is obtained by a rolling test bench or equivalent calibration device with independent roller linear velocity measurement. Under each index condition, the wheel being verified is made to roll with the roller and the driving torque of the wheel is zero. The wheel position linear velocity and wheel speed verification samples are obtained. After calculating each verification sample according to the above rkin,j=vx,j / ωj, the following multi-dimensional grid, most unfavorable endpoint and single mismatch upper bound rule are formed. The calibration unit also saves the applicable domain, version and validity period of the verification sample. Extrapolation is prohibited when any dimension exceeds the boundary, the unit is missing, the version is inconsistent or the record expires.

[0067] when , If the value is less than 0 or its low-slip valid field verification fails, it is not used. The effective radius of wheel-end torque is derived from an offline table indexed by tire size, load, tire pressure, temperature, wheel-ground condition, and drive condition. The relevant boundary is invalid when there are no valid entries. Multi-motor thrust targets can be frozen. distribute, and ; This is used only for thrust target allocation and not for replacing the actual torque of each wheel with the total actual torque to form the driving force boundary. The driving force boundary is determined directly based on the most unfavorable endpoint of the actual torque of each wheel and its corresponding parameters. ; for the first One benchmark synchronous sample, .in For the first Each electric drive wheel is effective within the reference phase. The median of the sample. For the first In the first reference synchronization sample The actual torque of each electric drive wheel; , and These are the conservative lower / upper bounds for the corresponding total transmission ratio, efficiency, and effective radius of wheel-end torque, respectively. When used for calculating the lower bound of the reference driving force, the lower bound is formed only if the median of the actual torque of each electric drive wheel's effective sample within the reference phase is not less than zero; if any of the medians is less than zero, the lower bound of the reference driving force and subsequent sets dependent on this boundary are deemed invalid. The non-negative truncation in the above formula serves only as numerical protection when the non-negative valid condition is met and is not used to convert negative actual torque into a valid driving force lower bound input. If independent wheel-end force measurement is used, the independently verified lower / upper bound of the driving force confidence is directly adopted. If any gear, transmission path, or actuator limit is changed, the driving force boundary and all subsequent sets must be recalculated. reallocation upon change And re-verify all subsequent sets; the old boundary or old target must not be used before the recalculation or re-verification is completed.

[0068] The aforementioned offline table, indexed by tire specification, load, tire pressure, temperature, wheel-ground condition, and drive condition, is a multi-dimensional calibration table used to store the offline calibration results of the wheel-end torque effective radius. Each valid record includes at least the discrete values ​​or applicable ranges of the above index conditions, the corresponding lower and upper bounds of the wheel-end torque effective radius, the single mismatch upper bound used to form or correct this boundary, as well as the unit, applicable domain, calibration version, and validity period. The offline table outputs the wheel-end torque effective radius boundary for torque-drive force conversion, and does not directly replace this boundary with the kinematic ratio under sand conditions.

[0069] The overall transmission ratio, wheel-end torque effective radius, and efficiency offline tables all use a pre-fixed multi-dimensional grid. The mechanical fixed transmission ratio can be directly formed from its tolerance-checked lower and upper bounds. When multiple gears or variable transmission paths exist, the corresponding index is used based on the current gear or transmission status. The boundary is invalidated when shifting gears or when the state is uncertain. Continuous inputs are calculated only if all vertices falling into the grid cell have unexpired calibration records; Take the minimum of the lower bounds of each vertex and then subtract the independently verified upper bound of the single mismatch. Take the maximum value of the upper bounds of each vertex and add it to the upper bound of the mismatch. and Each dimension is obtained according to the same worst-case endpoint rule. Extrapolation is prohibited if any dimension exceeds its bounds, vertices are missing, versions are inconsistent, or there is no valid record for the mismatched upper bound, and the corresponding driving force conversion boundary is marked as invalid. Continuous dimensions are left-closed and right-open, and the last element contains the right endpoint.

[0070] When querying offline tables, the corresponding multidimensional grid cell is first determined based on the current tire specification, current load, tire pressure, temperature, wheel-ground condition, and drive status. Continuous input only participates in the calculation if all vertices of the cell contain valid records that have not yet expired. The lower bound of the effective radius of wheel-end torque is formed by subtracting the aforementioned single mismatch upper bound from the minimum value of the lower bounds of all valid vertices, and the upper bound is formed by adding the aforementioned single mismatch upper bound to the maximum value of the upper bounds of all valid vertices. This query result is then used for the aforementioned conversion of actual torque to longitudinal driving force at the wheel end. No valid query result is generated if any index dimension is out of bounds, vertices are missing, versions are inconsistent, or there are no valid records for the mismatch upper bound.

[0071] Before the observation window begins, take the most recent The median of the effective actual torque samples is generated. and limit it to The stability window must also meet gating requirements for torque ripple, slope, and rate of change in driving demand. After the reference phase ends, that phase becomes effective. Determining the median of the sample to be effective and Determining the median of the sample and Once these three elements are determined, they are frozen until the end of the complete window. All boundaries retain numerical values, units, coordinate systems, original time ranges, versions, applicable domains, generation times, and expiration times.

[0072] Residual uncertainties in the reference torque measurement, driving force conversion, and execution that are not yet covered by the transmission ratio, efficiency, effective radius boundaries, or independent wheel-end force confidence boundaries are incorporated into the upper bound of the non-negative error frozen before the start of the observation window. and The driving force is calculated once in each direction that decreases the lower bound and increases the upper bound. If the independent wheel-end force path has already directly formed a reference driving force confidence lower and upper bound that covers the corresponding residual error, then it is directly used as the reference driving force confidence lower and upper bound. and and the corresponding or Set to zero; do not include errors already covered by existing boundaries.

[0073] Zero torque can be considered a safe output, but online identification only applies to... ,according to able to Within the budget, the formation of greater than The system is activated when the actual torque increment meets the following error condition. Bridging samples are not included in the 80-point steady-state observation window described later; steady-state sample counting for this phase only begins after entering the target's allowable zone.

[0074] Example 2: Dual-Response Identification and Online Conservative Boundary

[0075] Soft gating only determines whether to allow probing to begin; it does not replace hard safety arbitration. An executable gating mechanism is defined as follows: the attitude and rate of attitude change do not exceed a calibrated threshold. No more than , , The key sensor timestamps, units, coordinate systems, and health status are valid, and the excitation energy can be [contained / excited]. Arrive within the designated time, and the expected activity period is no more than [time period missing]. . , , , , and All conditions are fixed by the current boundary version before the observation window begins. A new window will not begin if any condition is not met.

[0076] Key sensors include at least , , Source, wheel speed, vehicle speed, longitudinal acceleration, subsidence measurement signal, attitude, braking, and grounding status; when the propulsion authorization function is configured and activated, the propulsion authorization source is also a critical input. The absence of a configured or explicitly inactive propulsion authorization function does not constitute an input failure; if the propulsion authorization function is activated but authorization data permission verification fails, the timestamp expires, integrity is abnormal, or the communication health status does not meet requirements, it is directly classified as... It outputs zero torque in the current cycle and must not be changed to... Continue with normal control. This includes issues such as brake override, wheel lift-off or ineffective ground contact, motor overspeed, and critical sensor failure. Failures and valid safety shutdown requests; other hard safety events or algorithm failures that, after safety analysis, allow the slope to return to zero are classified as... .by This represents the current valid security event index, with each event storing the clearing conditions, debouncing time, and event torque limit. ;when When multiple valid events exist simultaneously, take and Currently valid.

[0077] When triggered, the safety channel outputs zero in the current cycle and latches SAFE_ZERO, while simultaneously invalidating the current cycle. , , , Online paired boundaries, timers, and counters. Only available in... Release, Brake release The system is valid, the authorization status is consistent with the current configuration, and the key inputs are continuous. Only when the input is fresh and valid, and an independent reset confirmation is received, will it transition to IDLE with zero output; where the authorization status is consistent with the current configuration, it means that the advance authorization function is not configured or activated, or that the advance authorization function is activated and the authorization input is valid again. Subsequently, it must be re-identified from BASELINE and PROBE. Press when triggered Reduced to zero, Release and basic inputs are continuous Once the point is valid, it will switch to IDLE with zero output; neither of the two recovery methods should directly enter PUSH.

[0078] exist Point-by-point definition on the unified resampling axis , and ,in To standardize the resampling point index, and The first The longitudinal acceleration and subsidence velocity of each resampling point. Among them, when hour, ;when hour, . The actual torque, not the commanded torque, must be used. Each It is stored together with the original valid flags, timestamps, and resampling sources of the three channels; missing or invalid samples are not imputed. The response phase is allowed to start acquisition from the first return to the reference from the actual torque. There are 10 response samples, of which the first Each sample is used simultaneously for continuous baseline verification; the command has been returned. The process ends when both the total number of samples and the confirmation criteria are met. (Exceeding this threshold...) When the activity period expires, the response window is determined to be an algorithm failure and classified as ED, and reduced to zero according to the aforementioned ED safety torque reduction rules; when there is an E0 event in the same sampling period, E0 is executed first.

[0079]

[0080] In the formula, the symbol "√" indicates taking the non-negative square root of the product of the sums of the two squares within the brackets, and the square root applies to the entire bracket; the normalized cross-correlation coefficient is calculated only when both sums of squares are positive finite values. For each , It consists only of all pairs where both endpoints are valid and the index is not out of bounds, not by... Remove small increment samples from the numerator or denominator of cross-correlation; define another... As an effective incentive pairing set, first eliminate pairs with cross-correlation denominators of zero or non-finite values, or... Candidate delay; if fewer than two candidates remain, the response is invalid. All remaining candidates are treated as a whole... calculate Select the delay corresponding to the largest correlation value, and take the smallest delay when they are in parallel; it is also required that the maximum value is not less than And the difference between it and the second largest value is not less than . The resolution and uncertainty of the actual torque are predetermined and used only for gating the number of effective excitation pairs, selecting the regression after the delay, and screening the regression structure in the resampling window. It is not used to change the cross-correlation summation set. Calculate based on all valid points of the complete steady-state window, without applying the dead zone.

[0081] Normalized following error of complete steady-state window ,in The number of valid points for this window. The lower limit of the denominator is fixed in advance.

[0082] The square root in the above normalized following error formula only applies to the result of "the sum of squared following errors divided by the number of effective points in the window" within the square brackets. That is, first calculate the non-negative square root of the result, and then divide it by the normalized denominator formed by taking the larger value of the probe torque increment and the preset lower limit of the denominator. This normalized denominator is outside the square root, and the square root does not apply to the entire quotient.

[0083] The original complete window also requires the number of effective excitation pairs, two correlated peaks, peak difference, and so on. and Each condition meets the preset threshold and there are no hard security events. The joint loop moving block resampling phase only repeats the pairing structure validity screening, and does not repeat... Peak difference and A threshold is set to avoid selectively truncating the resampled distribution after the fact.

[0084] Explanation of key sensor and window validity.

[0085] Timestamp freshness, range, jump rate, hardware diagnostics, coordinate system, and units are fixed within a boundary version. Each invalid flag enters the security arbitration within the same sampling period, replacing the freshness invalidation value with the old value that is not retained. A complete window of ternary samples, validity masks, and baseline quantities are archived as indivisible records.

[0086] Explanation of dual-response regression, signal-to-noise ratio, and online conservative boundary.

[0087] Under their respective selected delays, only use Effective incentive pairs were subjected to origin least squares regression; peak cross-correlation was selected using... Regression slope estimation uses The purpose of the samples for both remains fixed within the same observation window.

[0088]

[0089]

[0090] in, The least squares regression slope estimate passing through the origin. The standard error, For effective regression pairings and .

[0091] The residual directional boundary is , The number of effective regression pairs , and This is the positive coverage coefficient fixed before the observation window begins. For the response... Definition of regression explanation of sum of squares and take ; zero and Time definition positive infinity and All responses are invalid at time zero.

[0092] Cyclic moving block resampling resamples the ternary samples and the validity mask together. A complete window is set. Each attempt to extract Each of the zero-based cyclic blocks starts from a continuous loop. Connect the points by wrapping them around the N-th point, and take the first N points to form a resampling window. Each resampling window is then re-formed. and Only when the cross-correlation denominator is finite and non-zero, Not less than When candidate delays are retained, the candidate with the largest correlation value is selected; if there is a tie, the minimum delay is taken, and then only the corresponding delay is used. Calculate the slope through the origin , ,in This is the sequence number of the valid joint resampling result. Pairwise rejection occurs when no structurally valid candidate delay is found in any response; the joint resampling stage is not repeated. Peak difference and threshold.

[0093] The block length, extraction rules, pseudo-random seed or seed derivation rules, and boundary versions for each window are fixed before the observation window begins. The number of valid paired results for the target is... At most try Next; in Unable to obtain within one time When there are valid paired results, the online paired boundary of this window is invalid. The implementation stores sufficient data to verify the extraction index, validity mask, acceptance or rejection reason, selected delay, and... The state and the result.

[0094] Pre-allocation and and satisfy .Will One valid and Sort them in ascending order, starting with 1 and without interpolation, and take... and Ultimately, , The seed or its derived algorithm is part of the boundary version and is fixed within the same version; it must not be selectively changed after observations.

[0095] Example 3: Offline envelope, resistance tolerance limit, and dual-margin feasible torque set

[0096] The offline envelope uses the same baseline-excitation-response-history maintenance as the online probe. Offline response nodes and future resistance increments are determined according to the following one-sided sequential statistical rules: A number of independent verification samples Sort in ascending order Given the probability of one-sided tail overflow. and confidence missing quantity Conservative lower bound index takes the condition that satisfies The largest Conservative upper bound index takes the condition that satisfies The smallest ,in Let be the number of combinations; corresponding boundaries are statistically analyzed in the corresponding order, and a side is invalid if no valid index exists on either side. For the offline response envelope, let It is a positive integer. This represents the torque increment interval between adjacent nodes in the offline response envelope, in N·m, and satisfies... ,node , For range indexes ,make and increase the candidate torque increment .when lie in The interval boundaries are formed by corresponding adjacent nodes, and the last interval is formed by... Therefore, it includes the right endpoint. Each node uses Each independent validation history is obtained by aligning with the online process in terms of time, effective sample gating, and regression caliber. and And before sorting, the measurement, synchronization, and regression residual errors are sorted according to... Decrease, make The direction of increase is taken into account. Following the aforementioned unified rule, the nodal acceleration response is statistically analyzed using the lower-order boundary values. The node subsidence velocity response is taken from the upper-side sequential statistical boundary. ; Node usage , and Fix before verification. and It covers changes between adjacent nodes, interpolation, working condition migration, excitation migration to continuous maintenance, and prediction time domain changes, and only counts the unfavorable direction once.

[0097] The aforementioned "unfavorable direction" refers to the conservative direction relative to the double-margin determination: for longitudinal acceleration response, it is handled in a direction that reduces the lower bound of the response to avoid overestimating the propulsion capability brought by the candidate torque; for subsidence velocity response, it is handled in a direction that increases the upper bound of the response to avoid underestimating the subsidence risk brought by the candidate torque. Therefore, the coverage corresponding to changes between adjacent nodes, interpolation, working condition transition, transition from excitation to sustained operation, and changes in the predicted time domain are used to reduce the conservative lower bound on the acceleration response side and to increase the conservative upper bound on the subsidence velocity response side.

[0098] The specific process of "including one instance in the unfavorable direction" is as follows: First, based on the node verification history, a lower or upper bound for the nodes that already includes node measurement, synchronization, and regression residual errors is formed. Then, for the interval where the candidate torque is located, adjacent nodes are selected. On the acceleration response side, the corresponding migration coverage is deducted only once based on the smaller value of the lower bound of the adjacent nodes, and on the subsidence velocity response side, the corresponding migration coverage is added only once based on the larger value of the upper bound of the adjacent nodes. After the offline interval boundary is formed, the same migration coverage will not be repeatedly deducted or added when synthesizing it with the online boundary in the unfavorable direction; the same verification residual already covered by the node boundary will also not be included again as migration error.

[0099]

[0100]

[0101] For candidates Take directly If the result is less than 0 or not finite, then the paired boundary is invalidated and set... Empty. The direction of the subsidence velocity is taken as... .thus and All are conservative boundary functions determined by the increment of the candidate torque, and no temporary variables are set only for intermediate propagation. To complete the observation window at the end of time, The completion time of BOUND_CHECK. To meet the actual torque requirements and continuously The final time after each valid sampling point is confirmed. Boundary packets are specified separately. and Maximum age allowed; arrival Used no earlier than fresh , Recalculation Forecast margin and , No longer belongs to the new Reduce torque immediately.

[0102] Longitudinal dynamics Therefore The online vehicle equivalent mass is defined as an interval determined and verified based on the current load condition. express, ,in In order to be in The mass ergodic variable takes values ​​within the range; it does not form a feasible set when the load state or mass boundary is invalid. This can be directly calculated from the most unfavorable endpoint of the aforementioned transmission ratio, efficiency, and effective radius boundary of wheel-end torque, or formed by independent wheel-end force measurement. For the first [unclear] within the frozen reference window... One valid synchronized sample, This represents a conservative lower bound for the longitudinal acceleration of the synchronized sample, taken as... Take again This quality range can be maximized to cover The most unfavorable mass direction when it is positive or negative. and Measurement uncertainties are included separately in the directions of increasing driving force and decreasing acceleration. It has been formed according to the direct conversion rules of Example 1. Numerical values, sources, units, coordinate systems, mass ranges, and error terms must be stored in pairs and must not be mixed. The predicted settlement margin and predicted traction margin are calculated item by item according to the following formulas; the settlement state, response boundary, reference boundary, and resistance increment boundary used must all belong to the same valid boundary version. For satisfying the unique... , and Finite and not less than The candidate list has the following boundary:

[0103] in Indicates candidate torque The corresponding conservative upper bound for the predicted subsidence depth; ; The difference is the predicted subsidence margin.

[0104]

[0105]

[0106]

[0107] in, To predict traction margin.

[0108] Used to verify the direction of the lower bound of the driving force increment derived from the net acceleration increment: only when the conservative lower bound of the future drag increment is non-negative. This prevents overestimation of the increase in driving force simply because of a decrease in resistance. Conversely, A conservative upper bound used to give the future resistance of candidate states. ,thereby The comparison is between the lower bound of the candidate driving force and the upper bound of the candidate future resistance. , Non-finite When a cell is missing or a candidate crosses / repeatedly matches a cell, the candidate is directly excluded without truncating negative values ​​or expanding the feasible set by using adjacent cells.

[0109] by , structure It also requires... , , only, , Limited and Furthermore, all boundary versions, units, coordinate systems, time domains, and applicable domains must be consistent. Candidates are excluded if any condition is not met; otherwise,... Empty.

[0110] The offline error handling is divided as follows: and Coverage of response nodes, interpolation, operating conditions, preservation of history, and prediction of temporal migration; in the sinking boundary and Uncertainties in subsidence measurement and errors in velocity measurement models; Covering the upper bound model of the benchmark resistance; Only from The lower tolerance limit is deducted downwards. Only to The upper tolerance limit is increased, and both cover synchronization, maintenance, and benchmark matching errors at future evaluation times. The same verification residual is only included in one of the error terms, and the source identifier is stored in the boundary bag.

[0111] Multidimensional resistance calibration unit At least by candidate , , Vehicle speed, load, tire pressure, sand type, place , and The division is as follows: continuous dimensions are left-closed and right-open, with the last element containing its right endpoint. For multidimensional resistance calibration cell indexing; each candidate must uniquely match a fully covered, unexpired one. .

[0112] exist Within, the s-th independent verification process is maintained in Obtain synchronization point estimation and And at the same time origin, and matching vehicle speed, sinking, load, tire pressure, attitude and reference torque , Pairing. Vehicle equivalent mass verified independently each time. Determined independently by weighing or calibrated load measurement and fixed during the holding process, its residual uncertainty is incorporated in the unfavorable direction. and ;calculate Use only Each comes from a mutually independent verification and maintenance process. And sorted in ascending order as According to the aforementioned unified one-sided order statistical rule, let , Use on the bottom Top side ,get and ;Pick , There is no valid index on either side, the upper bound is not finite, or... When this happens, the calibration unit is invalid. , Synchronization and benchmark matching errors are all conservatively accounted for only once. Candidate torque Unique Match season , This uniquely maps the two-sided tolerance limit of the unit to the candidate .

[0113] , Both calibration and online execution are performed using actual torque. The last point within a continuous band is ;since Keep until ,exist Acquired simultaneously and To verify the direction of the driving force, in The settlement state is obtained to verify the predicted direction. If there is a break or missing evaluation sample, that sample and the current candidate are invalid. The representative data below are only listed from multiple dimensions. Projection in the torque dimension.

[0114] Example of calculating tolerance limit under incremental resistance

[0115] The representative projection of the multidimensional resistance element on the candidate torque dimension is as follows.

[0116] Representative data are presented in the following order: Unit , , Sort samples / N , , .

[0117] The corresponding data for the first representative resistance calibration unit are as follows: , 72, 1.10, 1.40, 1.60, …, 1.80, 1 / 72, 0.60, 2.30.

[0118] The corresponding data for the second representative resistance calibration unit are as follows: , 72, 1.35, 1.70, 1.90, …, 2.00, 1 / 72, 0.85, 2.50.

[0119] The following representative data provides two representative resistance calibration units. (Take...) , , and order At that time, according to the aforementioned unified one-sided order statistical rules, the results were obtained. , The listed units Neither of them are negative. All values ​​are finite and not less than the corresponding lower tolerance limit; actual deployment requires recalculation using independent measured verification samples for each unit. Candidate torques are uniquely assigned to a single predefined calibration unit according to preset operating condition zones, and not to multiple units. The optimal choice should be selected from among the available options; when using cross-unit adaptive selection, a separate simultaneous confidence correction should be performed.

[0120] The final candidate set is:

[0121] only,

[0122] Limited and

[0123]

[0124] Empty, boundary expired, version or domain inconsistent, time age exceeded, result not finite, future resistance tolerance invalid or After recalculation No longer belong If at that time, Then latch zero output; otherwise press of and Safety torque reduction. Extrapolation, cross-unit interpolation, or expansion using old conditions are prohibited. .

[0125] Boundary Packet Calibration Example

[0126] Domain package C06-DOM-1.0 in Store the 0, 10, 20, 30, and 40 N·m nodes. 0.00805, 0.00795, 0.00786, 0.00780, and 0.093000, 0.101000, 0.106000, 0.110000. Each node uses 72 independent verification histories, taking... , The lower / upper boundaries of nodes are formed according to the aforementioned unified one-sided sequential statistical rules. Independent verification covers all endpoints, adjacent open intervals, and the period from detection to persistence. History; taking and Later obtained , If any node, interval, hold period, version, or validity period is missing, Overall invalid.

[0127] Same boundary packet storage The authorization function is not activated. Therefore ;save , , , , The efficiency lower / upper bound is 0.86 / 0.88, where For subsequent use only Distribution. Take the total reference torque. Single motor case , The result is obtained by direct conversion based on the aforementioned most unfavorable endpoint. Representative synchronous sample acquisition of the reference window , ,have to .Pick , , ,but Therefore Boundary packets are saved simultaneously. , ,each corresponding Sources of error terms. to The maximum permissible boundary age is 0.10 s. Use no earlier than The fresh, sunken state; beyond that age, the torque decreases directly.

[0128] Following the boundary packet example above, in Take from the only resistance calibration unit and take , , The effective online boundary was taken this time. , ; because the corresponding offline boundary is , After synthesizing in the unfavorable direction, select from the candidate interval. , Based on this, get ,Depend on get Therefore, the continuously feasible torque set is Of the discrete candidate torques {70, 80, 90, 100}, only 90 N·m is retained. Thus, the two margins respectively constrain the lower and upper limits of the candidate torques.

[0129] Discrete candidates For example, the corresponding unique resistance calibration unit satisfies and , and have Calculate according to the above formula: , ; , , Therefore, this candidate satisfies both the conditions for predicting subsidence margin and the conditions for predicting traction margin.

[0130] Example 4: Planning Reachability, Phased Execution, and Secure Recovery

[0131] BOUND_CHECK in the current command Completed and frozen when bridging begins. .make , , . and Freeze before selection, and satisfy .

[0132]

[0133]

[0134]

[0135] The worst-case command bridging point number between any two non-negative torque targets is determined by the following point function: For candidates ,Require , , And each command falls into The group of satisfied individuals Then from The candidates are selected according to pre-fixed candidate selection rules. The selection rules can be set with fixed priority or fixed weight based on traction margin, sink margin, torque increment and number of arrival points, and are determined before the start of the cycle and remain unchanged during the current cycle.

[0136] When the safety passage is not under control, the normal torque increase command is: The normal torque reduction command is Once any sampling period is completed... or Established, meaning the latch is first set to zero or contains... and The safety torque reduction formula is no longer included in ordinary commands. The target criterion is explicitly defined as follows. ; It's not a pure command bridging step count, but rather the number of steps accumulated from the first command point of PUSH until the actual torque meets the target criterion. The worst-case confirmation point index when there are consecutive valid in-band points, and the final confirmation point is... If the count is interrupted at an inner point, the continuous counting restarts. Therefore, it still fully covers the area after confirmation. and . The record will only be maintained after the re-verification passes.

[0137] The observation feedback is only allowed to reduce the conservative state when the BASELINE to OBSERVE are completed as planned, there are no safety events or lost samples, the versions are consistent, and the follow-up error is acceptable. and These represent the conservative upper bounds of the settlement depth and settlement velocity, respectively, formed based on the latest settlement measurements for a complete and valid OBSERVE stage. The two are determined separately, and one valid quantity is not used to replace the other.

[0138] For conservative states that require allowable descent, a uniform limit update rule is adopted: ,in Take respectively and ,corresponding Take respectively and The two types of states are determined separately, and if either observation is invalid, it cannot be replaced by another observation.

[0139] When the observation period is complete but individual samples are missing and the previous boundary is still valid, only maintaining or expanding the conservative boundary is allowed. Timestamp out-of-tolerance or invalid settlement measurements are considered critical sensor failures and are categorized as... According to the aforementioned The rule outputs zero torque and latches SAFE_ZERO in the current sampling period; boundary version mismatch or the previous boundary expiring indicates algorithm boundary failure, and in the absence of concurrency... Event time classification According to the aforementioned Rules and Maximum and press Safety measures reduce torque to zero. (This may be affected by version incompatibility.) , If the input is authorized or any of the aforementioned key sensor inputs, it will still be classified as... ; Same sampling period and When two processes coexist, the one that takes precedence will be executed first. . Even after a new boundary is reached that matches the version and has not expired, the event is still lifted only according to the aforementioned ED recovery rules; The release and reset will still be performed according to the aforementioned SAFE_ZERO recovery rules. The current activity cycle terminates at the end of OBSERVE. During the week's preparation phase, it will first follow the rules... Consistent directional speed limiting rules reduce torque from Restore to the most recent effective date :when Press at time Rise and twist, when Press at time Reduce torque; then continue The points satisfy the stability gating; under the example parameters below, the recovery and stability wait totals 24 points, not included in the previous activity cycle. It is also not included in the next observation window and is subject to the independent recovery cap. constraint.

[0140] The example values ​​and explanations for each core example parameter are as follows.

[0141] Example values ​​for the parameter group "Sampling and Full Window": ; ; ; ; ; ; The 80-point identification window does not include command bridging / in-band confirmation points; these additional points are counted in the activity cycle. The worst-case scenario for the return-to-base phase, from the end of PROBE until the actual torque first enters the base tolerance band, is four points, excluding the subsequent RESPONSE period used for continuous return-to-base confirmation. One sample; BOUND_CHECK is calculated based on 1 control cycle.

[0142] Example values ​​for the parameter group "Target and Rate": Target detected: 70.4; ; Torque (N·m), speed (N·m / s); maximum normal variation at each point is 12 N·m.

[0143] Example values ​​for the parameter group "Delay, Dead Zone, and Quality Threshold": ; ; ; ; ; ; ; ; Candidate delays are sampling points; For N·m; correlation, peak difference, The following error is only used for the original window qualification determination.

[0144] Example values ​​for the parameter group "Joint Resampling": , , , and , , Pre-fixed according to the current boundary version; ; ; The cyclic block rules and seed rules are fixed in advance.

[0145] Example values ​​for the parameter group "Evaluation Time Domain": ; ; ; ;Continuously maintain coverage of both evaluation moments.

[0146] Example values ​​for the parameter group "Number of Phase Split Points and Target": ; ; ; ; ; ; ; ; Points are calculated uniformly. Rule verification; torque values ​​are all subject to constraint.

[0147] Example values ​​for the parameter group "Time Limit": ; The activity cycle and the resumption during the week are subject to independent time limits.

[0148] Example values ​​for the parameter group "sinking and margin" are: ; ; ; ; Offline domain .

[0149] Example values ​​for the parameter group "Observe Updates": ; Only complete, high-quality observations can reduce the conservative boundary by limiting the single-cycle decrease.

[0150] Example values ​​for the parameter group "Safe Reboot": ; ; Independent reset confirmation was then required; the old online boundaries, sets, and counters were all invalidated and rebuilt.

[0151] Example values ​​for the parameter group "resistance tolerance limit": ; ; Lower / upper tolerance limits are formed according to a unified one-sided sequential statistical rule.

[0152] Example values ​​for the parameter group "Offline Response Statistics": ; ; Each node uses an independent verification history, and the lower / upper boundaries of the node are obtained according to a unified one-sided sequential statistical rule.

[0153] Taking the above parameters as an example, the worst-case number of BASELINE, probe bridging and confirmation, PROBE, return to baseline, RESPONSE, BOUND_CHECK, PUSH, UNLOAD, and OBSERVE is calculated as 20+20+20+4+40+1+106+10+20=241 points, which is 2.41 seconds, less than ; PUSH Complete in-band confirmation within budget and continue coverage. and The recovery and stabilization wait period after the observation period ends is counted separately as 0.24 s, and is not included in the previous activity cycle. The above point counts, according to the aforementioned non-repeating counting caliber, meet the constraints of the planning reachability, maintenance time domain, and activity cycle time limit.

Claims

1. A method for dynamic control of a two-wheeled vehicle in sandy terrain, characterized in that, The vehicle main controller performs the following steps: obtains the current authorized torque limit, the actual driving torque of the controlled wheel set, the longitudinal acceleration and the subsidence measurement value, and determines the upper limit of the subsidence depth and the upper limit of the subsidence speed based on the subsidence measurement value; Within the same observation time window, baseline acquisition, torque excitation, and response acquisition are performed. Based on the changes in the actual driving torque of the controlled wheel set relative to the baseline torque, as well as the changes in longitudinal acceleration and sinking speed relative to their respective baseline values, conservative lower bounds for longitudinal acceleration response and sinking speed response are determined, along with lower bounds for baseline driving force and upper bounds for baseline resistance. Candidate torques are generated between the baseline torque and the current authorized torque upper limit. The predicted sinking margin is determined based on the upper bounds for sinking depth, sinking speed, and sinking speed response, as well as the increment of the candidate torque relative to the baseline torque. The predicted traction margin is determined based on the conservative lower bounds for longitudinal acceleration response, baseline driving force, and baseline resistance, as well as a pre-calibrated future resistance increment boundary with a non-negative lower bound and a finite upper bound. Candidate torques whose predicted sinking margin and predicted traction margin are both not lower than their respective preset margin thresholds are combined into a feasible torque set. Based on the current torque command, preset torque increase rate, preset torque decrease rate, and planned duration, a target torque that can be reached within the planned duration is selected from the feasible torque set, and the torque command is transitioned to and maintained at the target torque. After the actual driving torque reaches the target torque, the target torque is recalibrated based on the updated subsidence measurement value. If the recalibration fails, the torque is reduced.

2. The method according to claim 1, characterized in that, The determination of the predicted subsidence margin includes: determining the upper bound of the predicted subsidence depth corresponding to the candidate torque based on the upper bound of the subsidence depth, the upper bound of the subsidence velocity, the increment of the candidate torque relative to the reference torque, the conservative upper bound of the subsidence velocity response, and the preset subsidence prediction time, and determining the predicted subsidence margin by the difference between the critical subsidence amount and the upper bound of the predicted subsidence depth.

3. The method according to claim 2, characterized in that, The determination of the predicted traction margin includes: under the condition that the pre-calibrated lower bound of the future resistance increment is not less than zero, determining the lower bound of the candidate longitudinal driving force based on the lower bound of the reference driving force, the increment of the candidate torque relative to the reference torque, the conservative lower bound of the vehicle equivalent mass, and the conservative lower bound of the longitudinal acceleration response; determining the upper bound of the candidate future resistance based on the upper bound of the reference resistance and the pre-calibrated upper bound of the future resistance increment; and determining the predicted traction margin by the difference between the lower bound of the candidate longitudinal driving force and the upper bound of the candidate future resistance.

4. The method according to claim 1, characterized in that, The step of selecting a target torque from the feasible torque set includes: determining the achievable torque range based on the current torque command at the start of bridging, the preset torque increase rate, the preset torque decrease rate, and the planned duration, and selecting a target torque from the intersection of the achievable torque range and the feasible torque set; and transitioning the torque command to the target torque under the condition that both the target torque and the torque command are non-negative and do not exceed the current authorized torque limit.

5. The method according to claim 4, characterized in that, The recalibration of the target torque includes: after the actual driving torque enters the target torque allowable range, recalculating the predicted settlement margin and predicted traction margin using the updated upper bound of settlement depth and upper bound of settlement speed; maintaining the target torque and then performing unloading and observation when both predicted margins are still not lower than their respective preset margin thresholds; and reducing the torque when either predicted margin does not meet the corresponding preset margin threshold.

6. The method according to claim 1, characterized in that, The process of performing baseline acquisition, torque excitation, and response acquisition within the same observation time window includes: determining a baseline reference torque before the start of the observation time window; determining the baseline torque, baseline longitudinal acceleration, and baseline sinking speed based on the median of the effective actual driving torque, longitudinal acceleration, and sinking speed of the controlled wheel set during the baseline acquisition phase; and maintaining these three values ​​until the end of the observation time window; the target torque during the torque excitation phase is higher than the baseline reference torque; and the target torque during the response acquisition phase returns to the baseline reference torque.

7. The method according to claim 6, characterized in that, The determination of the conservative lower bound of the longitudinal acceleration response and the conservative upper bound of the sinking velocity response includes: resampling the actual driving torque, longitudinal acceleration, and sinking velocity of the controlled wheel set onto the same time axis, without interpolating missing or invalid sampling points, and using the change of the actual driving torque of the controlled wheel set relative to the reference torque and the changes of the longitudinal acceleration and sinking velocity relative to their respective reference values ​​to determine the conservative lower bound of the longitudinal acceleration response and the conservative upper bound of the sinking velocity response, respectively.

Images