A method and system for controlling a roll attitude of an electric vehicle

Abstract

The application discloses a kind of based on landing constraint reverse transmission and air accessibility gate's electric vehicle flying slope cross phase posture collaborative control method and system.The application determines safe landing terminal constraint by landing area terrain, reversely determines take-off condition feasible region under the condition of aerial attitude dynamics, wheel torque limit and wheel angular momentum budget, controls the torque of multiple electric drive wheels in ground travel phase to drive actual take-off state into the feasible region, makes attitude tend to terminal constraint by wheel reaction torque during aerial period, and executes safety action when feasible region is empty or unreachable;Ground, aerial, landing three phases are uniformly distributed torque by the same optimization distribution framework, and transmission connection is not disconnected throughout the process;Three phase control subsystems are isolated from each other, and are only coupled by terminal constraint and feasible region boundary conditions interface.The application expands the speed and terrain envelope that can be safely flown, reduces landing impact, and ensures that the quality of ground phase control is not deteriorated by adding aerial attitude capability.

Description

Technical Field

[0001] This invention relates to vehicle dynamics control and distributed electric drive chassis control technology, specifically to a method, system, electronic device, and computer-readable storage medium for coordinated speed and attitude control of electric vehicles during takeoff (flying over ramps) based on ground constraint back-transmission and air accessibility gating, for use in autonomous driving and driver assistance scenarios. Background Technology

[0002] When high-speed off-road vehicles traverse terrain features such as slopes and dips, they may experience airborne motion (flying off the road). During this airborne phase, the vehicle's center of gravity trajectory is solely governed by gravity, and its translational motion is uncontrollable. However, the vehicle's attitude can be controlled through the reaction torque generated by the acceleration and deceleration of the wheels (the exchange of angular momentum between the wheels and the vehicle body). At the moment of landing, whether the vehicle's attitude is in close contact with the ground, and whether its angular and vertical velocities fall within the suspension's absorption range, directly determines the landing impact, instability, and rollover risk. Existing technologies can be categorized into four types, all of which have shortcomings.

[0003] The first category is reactive attitude control in mid-air. Representative works utilize the wheel reaction torque generated by throttle, braking, or steering after takeoff to adjust the vehicle body towards the target landing attitude (such as academic work on mid-air attitude control for high-speed off-road vehicles, and existing publicly available methods for adjusting pitch and roll in mid-air using distributed drive / braking torque). Its shortcomings are: it treats the attitude, angular velocity, and speed at the moment of takeoff as given and uncontrollable boundary conditions, only passively correcting them after takeoff; there is no solution when the takeoff conditions exceed the correction capability of the in-air actuators, and it provides no reverse guidance on the ground phase before takeoff.

[0004] The second category involves pre-planning of trajectory or velocity before takeoff. Representative works combine learned terrain-aware dynamics models with sampled model predictive control to pre-plan the entire trajectory, including the air segment, before takeoff. Alternatively, they jointly optimize position and velocity with velocity-conditional passability costs to determine when to jump and when to slow down. The drawbacks are: treating the air segment as a gravity-dominated open-loop trajectory without active air attitude control; incorporating velocity risk as a soft cost term into the shared cost function, allowing for compromises in global trade-offs; and not considering whether the air actuators can correct the vehicle to a safe landing attitude as a verifiable hard constraint or gating condition.

[0005] The third category is power system protection during takeoff. Existing methods disclose disengaging the clutch during takeoff to separate the power source from the drive wheels, and then re-engaging it before landing by matching the wheel speeds. The drawback is that, while aiming to protect the power system, it actively relinquishes the ability to adjust attitude using the reaction torque of the drive wheels during takeoff, contradicting the goal of active aerial attitude control.

[0006] The fourth category is retraining methods based on a unified learning model, which integrates skills such as takeoff and obstacle clearance into the same strategy or dynamics model through demonstration-based aggregated retraining. The applicant's simulation control experiments (high-fidelity vehicle dynamics simulation environment) show (see Comparative Examples 1 to 4 in the specific implementation details): Individual aggregated retraining introduces takeoff maneuver dynamics and contaminates the ground start and low-speed prediction submanifolds. The local gains of the target skill are offset by newly added start instability, and overall robustness is degraded on out-of-distribution initial states; global knowledge distillation pullback is overcorrected and generally too conservative; additive domain representation and data downsampling are insufficient to eliminate contamination. In other words, the seemingly natural approach of using a single learning model to cover the three phases has empirically proven failure modes, constituting a counter-instruction to existing technologies.

[0007] The common drawback of the aforementioned technologies lies in the fact that pre-jump planning and in-flight attitude control are either unidirectional modular series or coupled with single-model systems, lacking closed-loop coupling through verifiable boundary conditions—neither propagating landing constraints back to the pre-jump ground phase control nor using in-flight accessibility to reverse-gated takeoff conditions. As a result, the safe speed and terrain envelope for ramps is narrow, leading to significant landing impact and a high risk of instability when encountering large drops or irregular slopes.

[0008] In summary, there is an urgent need to design a method that can determine the safe landing terminal constraint from the terrain of the landing area, determine the feasible domain of takeoff conditions in reverse budget, and use the wheel reaction torque during takeoff to make the attitude tend towards the terminal constraint, thereby reducing the landing impact. This is the core challenge currently facing this field. Summary of the Invention

[0009] The purpose of this invention is to overcome the problem of lack of closed-loop coupling between pre-takeoff planning and in-flight attitude control in the prior art, specifically solving the following technical problems:

[0010] 1. Under the conditions of uncontrollable translational motion in the air and strict limitation of the ability to correct attitude in the air by wheel angular momentum budget, how to transform the safety landing terminal constraint into an executable constraint for ground phase torque control before takeoff, so that the speed and attitude of the vehicle when it arrives at the takeoff point fall into the range that the air actuator can correct in advance;

[0011] 2. How to determine online and verifiable whether a jump slope is safe to execute, and automatically perform safety actions such as deceleration, abandonment, rerouting, or slowing down when it is unreachable, avoiding reliance on unreliable extrapolation of learning models on out-of-distribution initial states;

[0012] 3. How to achieve joint optimization of the ground, takeoff, and landing phases without contaminating each other, avoid cross-phase negative transfer caused by coupling of the unified learning model, and ensure that adding takeoff attitude capability does not degrade the control quality of the ground phase.

[0013] In a first aspect, embodiments of this application provide a method for controlling the attitude of an electric vehicle during a ramp. The electric vehicle includes multiple electrically driven wheels, and the driving torque and braking torque of each electrically driven wheel are independently controlled by a controller. The method includes:

[0014] Based on forward-looking perception data, detect the characteristics of the ramp ahead of the electric vehicle's driving path, and predict the take-off point and the airtime of the electric vehicle after passing the take-off point.

[0015] Determine the terminal constraints for safe landing based on the terrain information of the landing area;

[0016] Based on the attitude dynamics of electric vehicles during takeoff, the torque limits of multiple electric drive wheels, and the takeoff duration, the feasible region of takeoff conditions is determined in reverse with terminal constraints as terminal conditions.

[0017] During the ground driving phase before the electric vehicle reaches the take-off point, the torque of multiple electric drive wheels is controlled so that the actual take-off state of the electric vehicle when it reaches the take-off point falls into the feasible region of take-off conditions.

[0018] During the airborne phase of an electric vehicle, the torque of multiple electrically driven wheels is controlled to generate a reaction torque, causing the electric vehicle's posture to tend to meet the terminal constraints.

[0019] Optionally, the method for controlling the attitude of electric vehicles during ramp crossings also includes:

[0020] Terminal constraints are used to limit the target attitude range and target velocity range when an electric vehicle lands safely.

[0021] The feasible domain of take-off conditions is the set of take-off states as follows: Starting from the take-off state, during the airborne period, the electric vehicle can satisfy the terminal constraints in attitude through the reaction torque generated by the torque of multiple electric drive wheels. The take-off state includes at least the speed and attitude parameters of the electric vehicle at the take-off point.

[0022] Optionally, the method for controlling the attitude of electric vehicles during ramp crossings also includes:

[0023] When the feasible region of take-off conditions is an empty set, or when the actual take-off state cannot fall into the feasible region of take-off conditions within the actuator capability and remaining travel distance of the electric vehicle, a safety action is performed. The safety action includes at least one of the following features: deceleration to abandon take-off, changing the travel path, and decelerating to pass through the flyover.

[0024] Optionally, the method for controlling the attitude of an electric vehicle during a jump includes: setting the trigger threshold for a safety action and the contraction margin of the feasible region for the take-off condition according to the operating mode, which includes at least one driver-in-the-loop mode and an autonomous driving mode; in at least one driver-in-the-loop mode, in response to the driver's confirmation of the safety prompt, no forced deceleration is performed, but attitude control and landing buffer are retained during the airborne period; in the autonomous driving mode, the safety action is automatically selected by the electric vehicle's behavior planning device between decelerating and abandoning the take-off, changing the driving path, and decelerating to pass through, and does not respond to the driver's confirmation.

[0025] Optionally, the feasible region for the take-off condition is determined by at least one of the following methods: analytical reachability criterion based on saturation switching control, Hamilton-Jacobi backward reachability solution, ellipsoidal or multicellular approximation, sampling approximation based on inverse numerical integration, and conservative inner approximation of the learning model output after shrinking according to the calibration residual; wherein, when using the conservative inner approximation, the take-off condition is determined only when both the conservative inner approximation and the analytical reachability criterion are passed.

[0026] Optionally, when determining the feasible region of takeoff conditions, wheel angular momentum budget constraints are also applied. The wheel angular momentum budget is determined by the moment of inertia of each electrically driven wheel and the margin between its current speed and the speed limit, in order to limit the changeable amount of the vehicle body angular momentum during takeoff.

[0027] Optionally, the wheel torque during the ground driving phase, the airborne phase, and the landing phase is allocated to the same group of multiple electric drive wheels by the same optimization allocation framework. The optimization allocation framework only switches the equality constraints and cost weights in different phases: the ground driving phase is allocated according to the target longitudinal force, the airborne phase is allocated according to the target vehicle body torque, and the landing phase is allocated according to the target wheel speed and buffer torque.

[0028] Optionally, the airborne phase also includes a neutral wheel speed rearrangement step: applying a torque combination to each wheel with a sum of zero torque to minimize the deviation between the rotational speed of each electrically driven wheel and the rotational speed corresponding to the expected landing speed without changing the vehicle's attitude.

[0029] Optionally, during the entire period when the electric vehicle is airborne, the transmission connection between the power source and the multiple electric drive wheels is kept uninterrupted, and attitude control is achieved solely through the bidirectional torque of the multiple electric drive wheels.

[0030] Optionally, the torque control during the ground driving phase is solved by roll time domain optimization. The roll time domain optimization takes the actual take-off state falling into the feasible region of the take-off conditions as a constraint, including model predictive control or sampled model predictive control. The roll time domain optimization adjusts the speed in the take-off state through the total torque, and adjusts the pitch angle and pitch velocity in the take-off state through the load transfer caused by the torque difference between the front axle and the rear axle.

[0031] Optionally, torque control during the ground driving phase involves correcting the torque requested by the driver, with the correction being the minimum intervention amount required to bring the actual take-off state into the feasible region of take-off conditions.

[0032] Optionally, the control of the ground driving phase, the takeoff phase and the landing phase are executed by mutually isolated control subsystems. The control subsystems are coupled only through terminal constraints, the feasible domain of takeoff conditions and the state continuity conditions at the phase switching time, so that the attitude control during takeoff does not change the control characteristics of the ground driving phase.

[0033] Optionally, at least one control subsystem includes a learning model, wherein the trainable parameters of the learning model related to takeoff and the trainable parameters of the learning model related to the ground driving phase are isolated from each other.

[0034] Optionally, the learning model includes a shared backbone model and a residual adapter: the residual adapter outputs a non-zero residual only for input samples with a preset stage domain identifier, and zero for the rest of the input samples. The shared backbone model and domain representation parameters are frozen during the training of the residual adapter. The residual adapter is a low-rank bottleneck structure, and its projected weights are initialized to zero at the start of training. The preset stage domain identifier is stored as a persistent parameter in the model weight file. At runtime, based on the precise matching results between the driving scene identifier and the preset scene set, the matching scene is routed to the preset stage domain identifier, and the unmatched scene is routed to the unconditional domain. When the scene domain identifier exceeds the number of domains supported by the learning model, it is treated as an unconditional domain.

[0035] Optionally, it also includes at least one of the following features: terminal constraints include at least one of the following: target pitch angle range, target roll angle range, upper limit of landing angular velocity, vertical landing velocity range, and wheel contact sequence; at least some of the multiple electric drive wheels have independently controllable steering angles, changing the wheel spin axis direction during takeoff through a combination of torque and steering angle, so that reaction torque is generated in at least two directions: pitch axis and roll axis; before the electric vehicle reaches the takeoff point, the damping and / or height of the semi-active or active suspension are preset to offset the actual takeoff state and / or relax the terminal constraints; the rotational speed of each electric drive wheel is maintained in the middle range of its rotational speed range to simultaneously retain wheel angular momentum adjustment margin in both directions during takeoff. The determination of takeoff and ground contact is based on at least one of the following: vertical acceleration threshold, suspension travel, sum of estimated wheel loads for each wheel, wheel acceleration change rate, and a calibration lookup table or learning classification model indexed by vehicle speed, suspension travel, and vertical acceleration. When the motor of any electric drive wheel fails or is limited in torque, regenerative braking capability is limited, or perception confidence decreases, the corresponding actuator limit contraction or prediction interval expansion is substituted into the determination of the takeoff condition feasible region, thereby reducing the takeoff condition feasible region accordingly. During takeoff, the remaining flight time is updated online based on the measured motion state, and accessibility is re-determined. When the terminal constraint becomes unreachable, the attitude control target during takeoff is switched to the attitude with the smallest deviation from the terminal constraint in the set of reachable attitudes.

[0036] Secondly, embodiments of this application provide a ramp attitude control system for an electric vehicle, employing the ramp attitude control method for electric vehicles as described above. The system includes: multiple electric drive wheels, each of which has independently controllable driving torque and braking torque; a forward-looking perception device configured to acquire terrain information ahead of the electric vehicle's driving path; and a controller configured to include:

[0037] Terrain Prediction and Fly Slope Prediction Module: Based on forward-looking perception data, detect the characteristics of fly slopes ahead of the electric vehicle's driving path, and predict the take-off point and the airtime of the electric vehicle after passing the take-off point.

[0038] Landing Terminal Constraint Determination Module: Determines the terminal constraints for safe landing based on the terrain information of the landing area;

[0039] Aerial backward reachability analysis module: Based on the attitude dynamics of the electric vehicle during takeoff, the torque limits of multiple electric drive wheels, and the takeoff duration, the feasible region of takeoff conditions is determined in reverse with terminal constraints as terminal conditions.

[0040] Pre-jump rolling time domain optimization module: During the ground driving phase before the electric vehicle arrives at the take-off point, the torque of multiple electric drive wheels is controlled so that the actual take-off state of the electric vehicle when it arrives at the take-off point falls into the feasible region of take-off conditions.

[0041] Airborne attitude control and landing buffer module: During the airborne period of the electric vehicle, the torque of multiple electric drive wheels is controlled to generate a reaction torque, so that the attitude of the electric vehicle tends to meet the terminal constraints.

[0042] Also includes:

[0043] Take-off determination and safety action module: When the feasible region of take-off conditions is an empty set, or when the actual take-off state cannot fall into the feasible region of take-off conditions within the actuator capability and remaining travel distance of the electric vehicle, a safety action is executed.

[0044] Three-phase unified torque distribution module: The ground, airborne, and landing phases are allocating torque by the same group of distributed electric drives according to the same optimized distribution framework;

[0045] Phase isolation architecture constraint module: The three-phase control is implemented by mutually isolated control subsystems, and is coupled only through terminal constraints, the feasible domain of the start-up condition and the continuous condition of the phase switching state.

[0046] Thirdly, embodiments of this application provide an electronic device, including:

[0047] A memory on which computer programs or instructions are stored;

[0048] A processor is configured to execute the computer program or instructions in the memory to implement the above-described method for controlling the attitude of an electric vehicle during a ramp.

[0049] Fourthly, embodiments of this application provide a vehicle, which includes a plurality of electrically driven wheels, wherein the driving torque and braking torque of each electrically driven wheel are independently controlled by a controller, and further includes: electronic equipment as described above, or includes a ramp attitude control system for electric vehicles as described above.

[0050] Fifthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of a method for controlling the attitude of an electric vehicle during ramp crossing.

[0051] Compared with existing technologies, it has the following outstanding advantages:

[0052] 1) The method and system of this invention propose a step-by-step backpropagation of landing constraints (cross-phase joint optimization). The landing terminal constraint is determined by the terrain of the landing area, and the feasible region of the takeoff conditions is derived in reverse. Then, using this feasible region as the terminal constraint, the motor torque sequence of the ground phase is solved. That is, the landing constraint, takeoff conditions, and ground torque are propagated step-by-step, and the three phases are combined into a single optimal control problem. Technical effects: The takeoff conditions are changed from passively given to actively shaped, significantly expanding the safe speed and terrain envelope for ramps, and reducing landing impact.

[0053] 2) The method and system of this invention propose aerial backward reachability gating and takeoff determination. Under the constraints of airborne attitude dynamics, torque limits, and wheel angular momentum budget, the feasible region of takeoff conditions is determined in reverse with the landing terminal constraint as the terminal; if the feasible region is empty or unreachable, a safety action is executed. Technical effect: The takeoff determination is based on analytically verifiable safety boundaries rather than learning extrapolation, and does not trigger unverified airborne maneuvers in the initial state outside the distribution, thus avoiding uncontrollable rollover and bottoming out from the source. (Simulation closed-loop experiment: In the dimensionality reduction implementation on the learning dynamics model and sampling model predictive control stack in a high-fidelity vehicle dynamics simulation environment, the cross-section reachability gating, the pre-step reverse transmission, and the step-over braking hold improve the recovery rate of the continuous step-over scenario from 16.7% to 88.9% (approaching the theoretical upper limit of about 96% of the ceiling probe), which is purely software implemented and does not rely on additional actuators. See Implementation Example 7 for details.)

[0054] 3) The method and system of this invention propose a unified three-phase torque allocation. The ground, takeoff, and landing phases are allocating torque using the same distributed electric drive system within the same optimized allocation framework. Different phases only switch between equality constraints and cost weights. During takeoff, a combination where the sum of the torques of all wheels is zero can be applied, eliminating wheel speed differences without changing the vehicle's attitude and aligning the wheel speeds with the landing speed (due to conservation of angular momentum, the zero-sum operation only rearranges the speed difference modulus components, without changing the inertia-weighted average speed; the residual deviation of the average speed is balanced with the attitude control objective within the pitch angular momentum budget). Technical effects: Attitude adjustment and landing wheel speed matching are executed concurrently in time, reducing the required takeoff time, and full driving capability is restored immediately upon landing.

[0055] 4) The method and system of this invention propose a continuous transmission connection throughout the entire airborne process. During airborne operation, the transmission connection between the power source and the drive wheels is maintained, with attitude adjustment achieved solely through bidirectional torque from the wheel-end motors. Technical benefits: Preserves the ability to adjust attitude in mid-air and maintains drive continuity upon landing.

[0056] 5) The method and system of this invention propose phase isolation and boundary condition interface coupling. The three-phase control is implemented by mutually isolated control subsystems, coupled only through terminal constraints, the feasible domain of take-off conditions, and the continuous conditions of phase switching states; in the learning embodiment, the take-off related model adopts a structure of frozen shared backbone plus stage domain gated low-rank residual adapter, and the forward calculation of the non-take-off domain is strictly consistent with the baseline. Technical effect: While the target phase is improved, the non-target phase can be proven to have zero downlink (in the simulation experiment of Comparative Example 5, 21 random seed experiments were conducted, the target route completion rate increased from 1 / 21 to 5 / 21, the average route progress increased by about 130 meters, and the output of the non-target scenario was consistent with the baseline), overcoming the cross-phase negative migration failure mode shown in Comparative Examples 1 to 4. Attached Figure Description

[0057] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0058] Figure 1 This is a schematic diagram of the slope attitude control method for electric vehicles according to the present invention. Figure 1 ;

[0059] Figure 2 This is a schematic diagram of the slope attitude control method for electric vehicles according to the present invention. Figure 2 ;

[0060] Figure 3 This is a flowchart of the slope attitude control method for electric vehicles according to an embodiment of the present invention;

[0061] Figure 4 This is a schematic diagram of the connection conditions between the three-phase time axis and the phase switching of the present invention;

[0062] Figure 5 This is a schematic diagram illustrating the backpropagation of the sequence from the landing constraint to the feasible region of the take-off condition to the ground torque sequence in this invention.

[0063] Figure 6 This is a schematic diagram of the feasible region for the jump conditions of the present invention;

[0064] Figure 7 This is a schematic diagram showing the relationship between the dynamics of the airborne posture and the reaction torque input of each wheel in this invention;

[0065] Figure 8 This is a timing diagram of the landing phase wheel speed preset, attitude neutral wheel speed rearrangement, and landing buffer of the present invention;

[0066] Figure 9 This is a schematic comparison diagram of the landing pitch angle error and peak vertical acceleration of the present invention and reactive air control under large drop conditions;

[0067] Figure 10 This is a schematic diagram of the distributed electric drive topology of the present invention;

[0068] Figure 11 This is a schematic diagram comparing two cross-phase coupling architectures;

[0069] Figure 12 Comparison data charts for examples one through six;

[0070] Figure 13 Side view group for airborne directional control case;

[0071] Figure 14 This is a schematic diagram of the ramp attitude control system for the electric vehicle of the present invention;

[0072] Figure 15 This is a general block diagram of the ramp attitude control system for an electric vehicle according to an embodiment of the present invention;

[0073] Figure 16 A schematic diagram showing the sensor layout and controller module labeling for the entire vehicle;

[0074] Figure 17 This is a schematic diagram of a human-computer interface. Detailed Implementation

[0075] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0076] It should also be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0077] It should also be understood that, in various embodiments of the present invention, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0078] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0079] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0080] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0081] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0082] To make the above-mentioned features and effects of the present invention clearer and easier to understand, specific embodiments are described below in conjunction with the accompanying drawings. This specification discloses one or more embodiments incorporating the features of the present invention. The disclosed embodiments are merely illustrative. The scope of protection of the present invention is not limited to the disclosed embodiments, but is defined by the appended claims.

[0083] This invention addresses the problems of existing fly-up control systems that treat takeoff conditions as given boundaries, passively correct attitude only after takeoff, and cause cross-phase negative migration due to the coupling of three phases in a unified learning model. This invention determines the safe landing terminal constraint based on the terrain of the landing area. It then reversely determines the feasible region of takeoff conditions under the conditions of takeoff attitude dynamics, wheel torque limits, and wheel angular momentum budget. During the ground travel phase, it controls the torque of multiple electrically driven wheels to drive the actual takeoff state into this feasible region. During takeoff, the wheel reaction torque is used to guide the attitude towards the terminal constraint. If the feasible region is empty or unreachable, a safety action is executed. The ground, takeoff, and landing phases are uniformly allocated torque by the same optimization allocation framework and are continuously connected. The three-phase control subsystems are isolated from each other and coupled only through the interface of the terminal constraint and the feasible region boundary conditions. This invention expands the speed and terrain envelope of safe fly-ups, reduces landing impact, and ensures that adding takeoff attitude capability does not degrade the quality of ground phase control.

[0084] The physical symbols involved in this application are defined as shown in Table 1.

[0085]

[0086] Table 1

[0087] The methods and systems of this application are described in detail below with reference to specific embodiments:

[0088] Example 1

[0089] like Figure 1 As shown, Figure 1 This is a schematic diagram of the slope attitude control method for electric vehicles according to the present invention. Figure 1 This invention provides a method for controlling the attitude of an electric vehicle during a ramp climb, the method comprising:

[0090] Terrain Pre-aiming and Fly Slope Prediction Step 101: Based on forward-looking perception data, detect the fly slope features ahead of the electric vehicle's driving path, and predict the take-off point and the airtime of the electric vehicle after passing the take-off point.

[0091] Specifically, in this embodiment of the invention, the above-mentioned terrain pre-aiming and slope prediction step 101 includes: constructing an elevation map based on forward-looking perception. Detect the characteristics of the jump ahead of the driving path and predict the take-off point and the angle of the slope. And the duration of takeoff. This step includes the following sub-steps:

[0092] S1011, Slope Edge Detection: Detects abrupt changes in elevation gradient and convex edges along the planned path; Slope edge detection can be achieved by any one or a combination of geometric detection, learning-based terrain classification, and high-precision map priors.

[0093] S1012, Ballistics and Airborne Duration Prediction: Assuming the takeoff velocity is... The x-position at the take-off point is The z-position at the take-off point is Where g is the acceleration due to gravity and t is the current time; then The center of mass trajectory is levitation time Depend on The numerical root is used to obtain the result; the takeoff time can also be given by a calibration lookup table or a learning model, and recursively corrected online by inertial measurements after takeoff; for A confidence interval is given based on the slope of the landing surface, and the lower bound of the interval is used for subsequent accessibility assessment to ensure conservatism.

[0094] Landing terminal constraint determination step 102: Determine the terminal constraints for safe landing based on the terrain information of the landing area; the terminal constraints are used to limit the target attitude range and target speed range when the electric vehicle lands safely; the feasible domain of the take-off conditions is the set of the following take-off states: starting from the take-off state, during the airborne period, the electric vehicle can satisfy the terminal constraints in attitude through the reaction torque generated by the torque of multiple electric drive wheels, wherein the take-off state includes at least the speed and attitude parameters of the electric vehicle at the take-off point.

[0095] Specifically, such as Figure 5 and Figure 6 As shown in the specific embodiment of the present invention, the above-mentioned landing terminal constraint determination step 102 includes: fitting the pitch and roll slope of the landing surface using the elevation of the neighborhood of the landing point to construct the landing terminal constraint. Its content includes at least one of the following: target pitch angle range, target roll angle range, upper limit of landing angular velocity, vertical landing speed range, and wheel contact sequence. It means that the vehicle body attitude fits the ground, multiple wheels contact the ground almost simultaneously, the landing angular velocity is close to zero, and the vertical landing speed falls within the suspension buffer window.

[0096] Step 103 of the aerial backward reachability analysis: Based on the attitude dynamics of the electric vehicle during takeoff, the torque limits of multiple electric drive wheels, and the takeoff duration, the feasible region of takeoff conditions is determined in reverse with the terminal constraints as the terminal conditions.

[0097] Specifically, in this embodiment of the invention, the above-mentioned aerial backward reachability analysis step 103 includes: the vehicle body attitude dynamics during takeoff... ,in, Let ω be the angular velocity of the vehicle body. The angular acceleration of the vehicle body, i.e., the reaction torque of the wheels' acceleration and deceleration, drives the vehicle body to rotate; the exchangeable angular momentum of the wheels is subject to budget constraints. .in, For the moment of inertia of the wheel, The angle at which the wheel jumps is... To determine the feasible region of takeoff conditions in reverse under the torque limit and angular momentum budget, for the terminal. Step 103 includes the following sub-steps:

[0098] S1031, Pitch Axis Analytical Criterion: Let , , thus obtaining a dual integrator ;remember , , , The pitch angle at landing is the angle of attack, which is determined by the duration of flight. The necessary and sufficient condition for the target to reach the landing pitch angle is: and and superimposed This criterion can be determined online in milliseconds and the jump determination boundary can be drawn directly.

[0099] S1032, Three-axis general case: For implementations with independent steering, backward reachability is solved according to the above dynamics. Any one of Hamilton-Jacobi backward reachability, ellipsoid or multi-cell approximation, or sampling approximation based on inverse numerical integration can be used.

[0100] S1033. Conservatism of the reachable region in learning: When the feasible region is given by the learning model, a conservative inner approximation after shrinking the calibrated residual is taken, and it forms a dual gating with the analytical criterion. Jumping can only be initiated when both criteria are simultaneously determined to be reachable.

[0101] Pre-jump rolling time domain optimization step 104: During the ground driving phase before the electric vehicle arrives at the take-off point, control the torque of multiple electric drive wheels so that the actual take-off state of the electric vehicle when it arrives at the take-off point falls into the feasible region of take-off conditions.

[0102] Specifically, in a specific embodiment of the present invention, the above-mentioned pre-jump rolling time-domain optimization step 104: using ground phase dynamics, torque limits, and adhesion constraints as constraints, and taking the actual take-off state at the take-off point as the landing point... Assuming terminal constraints, solve for the motor torque sequence; the objective function is: ;in, For speed weighting, For reference speed, As a torque weighting factor, the speed is adjusted by the total torque, and the takeoff pitch angle and pitch velocity are determined by the torque difference between the front and rear axles. After load transfer adjustment; the solution is obtained using model predictive control or sampled model predictive control, executed in a rolling manner from 50 to 100 Hz; in driver-in-loop mode, the output of this step is the minimum intervention correction amount for the torque requested by the driver.

[0103] Takeoff Attitude Control and Landing Buffer Step 105: The inner loop of takeoff uses PID control at 200 to 500 Hz to track the attitude reference trajectory and outputs a unified secondary planning allocation; after takeoff, the remaining flight time is updated online based on the measured status and the reachability is recalculated on a rolling basis. When the terminal constraint becomes unreachable, it switches to the attitude with the smallest deviation in the set of reachable attitudes; the transmission connection between the power source and the drive wheels is kept continuous throughout the takeoff process; after landing, the torque of the wheels that touch the ground first is limited, while the remaining wheels are maintained in the air and the torque ramp is limited within a preset time to complete the landing buffer.

[0104] The following example illustrates the four-wheel independent drive pitch control method provided in this application:

[0105] In this embodiment, the control frequency of the electric vehicle is: 20 to 30 Hz for sensing and elevation mapping, 50 to 100 Hz for pre-jump roll time domain optimization, 200 to 500 Hz for the inner loop of the airborne posture, and torque distribution synchronized with the innermost loop. However, this invention is not limited to this, and other ranges of the above values ​​can also be set.

[0106] like Figures 4-8 As shown, for ease of reference in various embodiments, coordinates and a phase machine are first defined. The inertial frame has the z-axis pointing upwards, the vehicle system origin is at the center of mass, x forward, y left, and z upwards; the wheels are numbered left front, right front, left rear, and right rear. The phase machine contains three phases: P0 approach (ground) phase, the goal of which is to drive the take-off state upon arrival at the take-off point into the feasible region; P1 take-off phase, the goal of which is to use the wheel reaction torque to track the attitude to the landing end constraint; P2 landing phase, the goal of which is to preset wheel speed and torque to complete the landing buffer and restore ground control. The determination of takeoff and ground contact can be achieved using any of the following methods or any combination thereof: the difference between vertical acceleration and gravitational acceleration is less than a threshold and continues for more than a preset time window; the suspension travel reaches full extension or the back pressure exceeds a threshold (this can be determined wheel by wheel, supporting a transitional state where some wheels are off the ground); the sum of the estimated wheel loads of all wheels is less than a threshold; the wheel acceleration or its rate of change exceeds a threshold (wheel speed anomalies caused by a sudden drop in wheel end resistance after takeoff); a calibration lookup table or a learning classification model indexed by vehicle speed, suspension travel, and vertical acceleration. The takeoff time is taken as the time of the last wheel's takeoff.

[0107] (1) Vehicle configuration: Four-wheel independent wheel-end motor electric off-road vehicle, each motor can output bidirectional torque (driving and regenerative reverse towing), normal steering and the front wheels are almost straight when jumping uphill, and the self-rotation axis of each wheel is fixed along the y-axis in the vehicle system; equipped with inertial measurement unit, wheel speed encoder, suspension travel sensor and forward-looking lidar / camera; there is no active disengagement operation between the power source and the drive wheels.

[0108] (2) such as Figure 7 As shown, the dynamics and directionality of airborne attitude: by In a fixed-axis configuration, the resultant moment has only a pitch component, i.e. Only pitch is controllable; when the vehicle is in mid-air, the driving force (accelerating the wheels) causes the front of the vehicle to tilt upwards, and the regenerative drag (decelerating the wheels) causes the front of the vehicle to tilt downwards.

[0109] (3) Angular momentum budget: The upper bound of the exchangeable angular momentum of the wheel is... The change in the angular momentum of the corresponding vehicle body This is the hard upper limit of whether it can be corrected in the air.

[0110] (4) Analytical reachability criterion: Let , Pitch is a dual integrator ;remember , , The necessary and sufficient condition for attainment is and , and then superimpose Online millisecond-level determination, that is, as Figure 6 The jump determination boundary is shown.

[0111] (5) Pre-jump roll time-domain optimization: longitudinal dynamics Traction limit , Indicates resistance; For vehicle quality, Acceleration, takeoff pitch angle, and pitch velocity are modeled as follows: The difference in torque between the front and rear axles is transferred via load. , ,function , Identify the torque difference using the step torque difference test system (record the pitch response at the moment of leaving the slope and fit a low-order model for the torque difference step); solve the torque sequence with the condition of falling into the feasible region upon reaching the take-off point as the terminal constraint, and the objective function is the same as in step S104.

[0112] (6) Unified torque distribution and levitation inner loop: Unified secondary planning as , It is a symmetric matrix. It is a torque vector. The coefficient matrix (different for different phase switching) ),constraint Torque and speed limits; ground phase Equal to the target longitudinal force, airborne phase Equal to the target torque, landing phase constraint on the rotational speed of each wheel. , It is the wheel radius. That's the landing speed.

[0113] (7) Numerical Verification Example: Assuming a vehicle mass of 1800 kg, pitch inertia of 2600 kg·m², wheel inertia of 1.2 kg·m², wheel radius of 0.36 m, peak torque of a single motor of 1200 N·m, peak motor speed of 1200 rpm, slope angle of 18 degrees, takeoff speed window of 8 to 14 m / s, and airtime of 0.6 to 1.2 seconds, then determine the upper limit of pitch acceleration. In radians per square second, the upper limit of the takeoff pitch velocity that can be eliminated in terms of torque within 1 second of airtime is approximately 1.85 radians per second; however, the angular momentum budget... KJ·m² / s, which translates to a change in vehicle pitch velocity of only about 0.23 radians / s (about 13 degrees / s). This verification shows that the real bottleneck is the wheel angular momentum budget rather than the motor torque. The in-flight attitude correction authority is inherently very small, and it is necessary to rely on the feasible domain to compress the takeoff pitch velocity into the budget before takeoff. This is the quantitative necessity of the pre-takeoff shaping in this invention.

[0114] (8) Results and Comparison: Compared to the baseline of reactive attitude correction only after takeoff, this embodiment shows a reduction in landing pitch angle error and peak vertical acceleration under large drop conditions, such as... Figure 9 a and Figure 9 As shown in b, safety actions are performed in advance when accessibility is not met, rather than entering an uncontrollable levitation.

[0115] Example 2

[0116] Based on the first embodiment described above, the present invention also provides another preferred embodiment, such as... Figure 2 and Figure 3 As shown, Figure 2 This is a schematic diagram of the slope attitude control method for electric vehicles according to the present invention. Figure 2 , Figure 3 This is a schematic diagram of the method in an embodiment of the present invention; the fly-up attitude cooperative control method provided by the present invention consists of steps S201 to S208 executed sequentially (S201 is terrain pre-aiming and fly-up prediction, S202 is landing terminal constraint determination, S203 is aerial backward reachability analysis, S204 is pre-jump roll time domain optimization, S205 is jump determination and safety action, S206 is three-phase unified torque distribution, S207 is airborne attitude control and landing buffer, and S208 is phase isolation architecture constraint). An embodiment of the present invention provides a fly-up attitude control method for an electric vehicle, the method including:

[0117] Step S201: Detect the characteristics of the ramp ahead of the electric vehicle's driving path based on forward-looking perception data, and predict the take-off point and the airtime of the electric vehicle after passing the take-off point.

[0118] Step S202: Determine the terminal constraints for safe landing based on the terrain information of the landing area;

[0119] Step S203: Based on the attitude dynamics of the electric vehicle during takeoff, the torque limits of multiple electric drive wheels, and the takeoff duration, the feasible region of takeoff conditions is determined in reverse with the terminal constraints as the terminal conditions.

[0120] Step S204: During the ground driving phase before the electric vehicle arrives at the take-off point, control the torque of multiple electric drive wheels so that the actual take-off state of the electric vehicle when it arrives at the take-off point falls into the feasible region of take-off conditions.

[0121] Step S205: When the feasible region of take-off conditions is an empty set, or when the actual take-off state cannot fall into the feasible region of take-off conditions within the actuator capability and remaining travel distance of the electric vehicle, a safety action is performed. The safety action includes at least one of the following features: deceleration to abandon take-off, changing the travel path, and decelerating to pass through the flyover.

[0122] The trigger threshold for safety actions and the shrinkage margin of the feasible domain for take-off conditions are set in stages according to the operating mode. The operating modes include at least one driver-in-the-loop mode and one autonomous driving mode. In at least one driver-in-the-loop mode, in response to the driver's confirmation of the safety prompt, no forced deceleration is performed, but attitude control and landing buffer during the airborne period are retained. In the autonomous driving mode, the safety action is automatically selected by the electric vehicle's behavior planning device between decelerating and abandoning take-off, changing the driving path, and decelerating to pass through, and does not respond to the driver's confirmation.

[0123] Specifically, in the specific embodiments of the present invention, the above-mentioned take-off determination and safety action include: checking in each control cycle whether there is a feasible solution within the speed window and remaining distance that allows the take-off state to enter the feasible region; when the feasible region is empty or unreachable, executing a safety action (at least one of decelerating and abandoning the take-off, changing the driving path, or passing through at a reduced speed); the safety action trigger threshold and the feasible region shrinkage margin are set according to the operating mode, which includes the driver-in-the-loop mode and the automatic driving mode. In the automatic driving mode, the safety action is automatically selected by the behavior planning device and does not respond to the driver's confirmation.

[0124] Optionally, the feasible region for the take-off condition is determined by at least one of the following methods: analytical reachability criterion based on saturation switching control, Hamilton-Jacobi backward reachability solution, ellipsoidal or multicellular approximation, sampling approximation based on inverse numerical integration, and conservative inner approximation of the learning model output after shrinking according to the calibration residual; wherein, when using the conservative inner approximation, the take-off condition is determined only when both the conservative inner approximation and the analytical reachability criterion are passed.

[0125] Optionally, when determining the feasible region of takeoff conditions, wheel angular momentum budget constraints are also applied. The wheel angular momentum budget is determined by the moment of inertia of each electrically driven wheel and the margin between its current speed and the speed limit, in order to limit the changeable amount of the vehicle body angular momentum during takeoff.

[0126] Step S206: Optionally, the wheel torque during the ground driving phase, the airborne phase, and the landing phase are all allocated to the same group of multiple electric drive wheels by the same optimization allocation framework. The optimization allocation framework only switches the equality constraints and cost weights in different phases: the ground driving phase is allocated according to the target longitudinal force, the airborne phase is allocated according to the target vehicle body torque, and the landing phase is allocated according to the target wheel speed and buffer torque.

[0127] Specifically, in a specific embodiment of the present invention, the above-mentioned three-phase unified torque distribution includes: ,constraint Together with torque and speed limits, a unified secondary planning is formed, and ground phase is distributed according to the target longitudinal force and adhesion constraints are applied. , For the x-direction force of the i-th wheel, the airborne phase is allocated according to the target vehicle body torque, and the landing phase is allocated according to the target wheel speed and buffer torque. Different phases only switch. , and , The coefficient matrix (different for different phase switching) ).

[0128] Step S207: During the airborne phase of the electric vehicle, the torque of multiple electric drive wheels is controlled to generate a reaction torque, causing the electric vehicle's attitude to tend to meet the terminal constraints.

[0129] Optionally, the airborne phase also includes a neutral wheel speed rearrangement step: applying a torque combination to each wheel with a sum of zero torque to minimize the deviation between the rotational speed of each electrically driven wheel and the rotational speed corresponding to the expected landing speed without changing the vehicle's attitude.

[0130] Optionally, during the entire period when the electric vehicle is airborne, the transmission connection between the power source and the multiple electric drive wheels is kept uninterrupted, and attitude control is achieved solely through the bidirectional torque of the multiple electric drive wheels.

[0131] Optionally, the torque control during the ground driving phase is solved by roll time domain optimization. The roll time domain optimization takes the actual take-off state falling into the feasible region of the take-off conditions as a constraint, including model predictive control or sampled model predictive control. The roll time domain optimization adjusts the speed in the take-off state through the total torque, and adjusts the pitch angle and pitch velocity in the take-off state through the load transfer caused by the torque difference between the front axle and the rear axle.

[0132] Optionally, torque control during the ground driving phase involves correcting the torque requested by the driver, with the correction being the minimum intervention amount required to bring the actual take-off state into the feasible region of take-off conditions.

[0133] Step S208: Optionally, the control of the ground driving phase, the takeoff phase and the landing phase are executed by mutually isolated control subsystems. The control subsystems are coupled only through terminal constraints, the feasible domain of takeoff conditions and the state continuity conditions at the phase switching time, so that the attitude control during takeoff does not change the control characteristics of the ground driving phase.

[0134] Optionally, at least one control subsystem includes a learning model, wherein the trainable parameters of the learning model related to takeoff and the trainable parameters of the learning model related to the ground driving phase are isolated from each other.

[0135] Optionally, the learning model includes a shared backbone model and a residual adapter: the residual adapter outputs a non-zero residual only for input samples with a preset stage domain identifier, and zero for the rest of the input samples. The shared backbone model and domain representation parameters are frozen during the training of the residual adapter. The residual adapter is a low-rank bottleneck structure, and its projected weights are initialized to zero at the start of training. The preset stage domain identifier is stored as a persistent parameter in the model weight file. At runtime, based on the precise matching results between the driving scene identifier and the preset scene set, the matching scene is routed to the preset stage domain identifier, and the unmatched scene is routed to the unconditional domain. When the scene domain identifier exceeds the number of domains supported by the learning model, it is treated as an unconditional domain.

[0136] Specifically, in the specific embodiments of the present invention, the aforementioned phase isolation architecture constraints include: the control of the three phases of ground, takeoff, and landing is implemented by mutually isolated control subsystems, and the subsystems are coupled only through landing terminal constraints, takeoff condition feasible domains, and phase switching state continuity conditions; in the learning embodiment, the takeoff-related learning model consists of a frozen shared backbone model and a low-rank residual adapter gated to the takeoff phase domain. The adapter only outputs non-zero residuals for inputs and outputs with preset stage domain identifiers, and the projected weights are zero-initialized, while the gated domain identifiers are persistently stored in the model weight file; during runtime, routes are precisely matched according to the scene identifiers, and when the domain identifiers go out of bounds, the unconditional mode is rolled back.

[0137] Optionally, it also includes at least one of the following features: terminal constraints include at least one of the following: target pitch angle range, target roll angle range, upper limit of landing angular velocity, vertical landing velocity range, and wheel contact sequence; at least some of the multiple electric drive wheels have independently controllable steering angles, changing the wheel spin axis direction during takeoff through a combination of torque and steering angle, so that reaction torque is generated in at least two directions: pitch axis and roll axis; before the electric vehicle reaches the takeoff point, the damping and / or height of the semi-active or active suspension are preset to offset the actual takeoff state and / or relax the terminal constraints; the rotational speed of each electric drive wheel is maintained in the middle range of its rotational speed range to simultaneously retain wheel angular momentum adjustment margin in both directions during takeoff. The determination of takeoff and ground contact is based on at least one of the following: vertical acceleration threshold, suspension travel, sum of estimated wheel loads for each wheel, wheel acceleration change rate, and a calibration lookup table or learning classification model indexed by vehicle speed, suspension travel, and vertical acceleration. When the motor of any electric drive wheel fails or is limited in torque, regenerative braking capability is limited, or perception confidence decreases, the corresponding actuator limit contraction or prediction interval expansion is substituted into the determination of the takeoff condition feasible region, thereby reducing the takeoff condition feasible region accordingly. During takeoff, the remaining flight time is updated online based on the measured motion state, and accessibility is re-determined. When the terminal constraint becomes unreachable, the attitude control target during takeoff is switched to the attitude with the smallest deviation from the terminal constraint in the set of reachable attitudes.

[0138] like Figures 3-13 As shown in the specific embodiments of the present invention, the specific implementation flow of each embodiment of the method of the present invention will be described below with reference to specific implementation examples:

[0139] The following example, using Example 2, describes the three-axis attitude control method with four-wheel independent drive and independent steering provided in the embodiments of this application:

[0140] Based on Example 1, each wheel is given independent steering, and the spin axis unit vector is... Torque Weight contribution, elevation, meridian Roll is contributed by components; for example, the pure roll torque is obtained by symmetrically reversing the steering angles of the left and right wheels and applying torque of the same sign. The yaw channel is provided by the reaction torque of the steering actuator around the z-axis and the gyroscopic precession torque generated by the high-speed rotating wheel spin axis. Its authority is less than that of pitch and roll. In engineering, the design goal is active control of pitch and roll and fine-tuning of yaw. The unified quadratic programming airborne phase constraint row is expanded to three rows, the decision variable is increased to steering angle, and the bilinear term is processed by alternating optimization or pre-selection of steering angle gear linearization. The three-axis reachable domain is tabled offline and queried online according to step S1032. The pitch main axis is still superimposed with analytical criteria for rapid pre-screening.

[0141] The following example, Example 3, illustrates the dual-axis, dual-motor degradation configuration method provided in this application:

[0142] Each axle has one motor, and torque is distributed between the wheels via a differential, resulting in two independent torque channels. The pitch torque in the air is the reverse sign of the sum of the torques of the front and rear axles, and the pitch channel is fully preserved. The torque difference between the front and rear axles and load transfer during ground phase are still usable; there is no inter-wheel differential. The angular momentum estimation is of the same order of magnitude as the four-electric mechanism configuration, and the analytical criteria are directly used. This embodiment demonstrates that the invention does not depend on the four-electric mechanism configuration and can be implemented in a dual-motor four-wheel drive vehicle.

[0143] The following example, Example 4, illustrates the phase isolation and stage domain gating adapter (learning embodiment) method provided in this application:

[0144] (1) Architecture: The three phase control subsystems are isolated from each other and coupled only through three boundary condition interfaces: landing terminal constraint, take-off condition feasible domain, and phase switching state continuity condition. Gradient and parameter level communication is prohibited. The take-off related learning model consists of a frozen shared dynamic backbone model and a low-rank residual adapter gated to the take-off phase domain: the adapter is a low-rank bottleneck structure (typically rank 32), adding residuals only to input samples whose stage domain identifier is equal to the preset gate domain, and strictly outputting zero for other inputs; the projected weights on it are initialized to zero, and the model at the training start point is strictly equal to the baseline; the gate domain identifier is stored as a persistent parameter in the model weight file, and the checkpoint is self-describing.

[0145] (2) Training and deployment: During training, all backbone and domain representation parameters are frozen, only adapter parameters are trained, and only the target phase domain corpus is used; during deployment, routes are precisely matched with the preset scene set according to the driving scene identifier. Matched scenes are routed to the gated domain, and unmatched scenes are routed to the unconditional domain. When the scene domain identifier exceeds the number of domains supported by the model, it falls back to the unconditional mode.

[0146] (3) Simulation control experiment (high-fidelity vehicle dynamics simulation environment, 21 random seeds under the same conditions): Control Example 1 (unisolated single-unit aggregation retraining) The local gain of the target skill is offset by the newly added start instability, the net value of the route is about zero, and the non-target retained route deteriorates from 543.8 meters to less than 5 meters. The low-speed manifold relative prediction drift is about 2.7 times that of the cruise manifold; Control Example 2 (global knowledge distillation pullback) is globally overconservative, and the retained route pass rate deteriorates from 19 / 21 to 1 / 21; Control Example 3 (additive domain representation) and Control Example 4 (target domain data downsampling) The non-target domain response slope is still negative, and the contamination has not been eliminated; Control Example 5 (structure of this embodiment) The target route completion rate is improved from 1 / 21 to 5 / 21, the average route progress is improved by about 130 meters, and the difference between the non-target domain response and the baseline is zero step by step, and the non-target scenario closed-loop progress is consistent step by step; Control Example 6 (adapter rank doubling) The response slope is almost unchanged, proving that the benefit comes from the isolation architecture rather than the number of parameters. The above comparison constitutes an isolation intensity spectrum from weak to strong: all conventional improvement paths failed, only the structural isolation of this embodiment was successful, and its target phase gain was accompanied by structural zero downlink of the non-target phase.

[0147] The following example, using Example 5, illustrates the suspension coordination, additional reaction flywheel, and regenerative braking ratio method provided in this application's embodiments:

[0148] In this embodiment of the invention, suspension coordination is implemented as follows: At the takeoff end, front and rear suspension damping and height are preset, and the pitch impulse of the vehicle body is offset by the suspension rebound at the moment of leaving the slope to determine the takeoff state; at the landing end, large damping and maximum travel are preset to widen the vertical landing speed window, directly expanding the feasible region via the reverse transmission chain; suspension commands are used as slow variables and are jointly optimized hierarchically with the torque sequence. An additional reaction flywheel is added to the pitch axis (e.g., a moment of inertia of 0.8 kg·m², a peak speed of 6000 rpm, and an additional angular momentum budget of approximately 502 kg·m² / s, nearly doubling the pitch angular velocity correction authority). The flywheel torque is used as an additional decision component in the unified secondary programming, and the budget constraints are updated accordingly. Regenerative braking ratio is also implemented: during the ground deceleration phase, regenerative braking is prioritized, and the speed of each wheel is kept in the middle range of the speed range, preserving angular momentum margin for attitude adjustment in both directions during takeoff; the adhesion coefficient is estimated online and adhesion constraints are tightened; the feasible region automatically shrinks and the takeoff judgment automatically becomes more conservative during adhesion descent.

[0149] The following example, Example 6, will illustrate the operating mode, human-computer interaction, and failure degradation method provided in the embodiments of this application:

[0150] In this embodiment of the invention, the operating modes are divided into four levels: Novice Mode (the feasible region is contracted according to the maximum uncertainty and an additional safety margin is added, tending to pass conservatively, with full intervention during takeoff to keep the vehicle frame in contact with the ground), Sport Mode (standard feasible region boundary, intervention only when the attitude error exceeds the preset angle), Expert Mode (allows the driver to confirm and override safety prompts, the system does not force deceleration but retains takeoff attitude control and landing buffer), and Autopilot Mode (unmanned operation, the torque sequence is fully executed by the controller, safety actions are automatically selected by the behavior planning device between deceleration abandonment, lane change, and deceleration to pass without responding to confirmation operations, the feasible region contraction margin is set according to the design operating region safety strategy and superimposed with occupant comfort constraints). The contraction coefficient at the moment of mode switching is taken from the more conservative of the two levels. When approaching the predicted jump slope, the human-machine interface displays the takeoff judgment status, the suggested speed window obtained from the feasible region projection, and the remaining adjustment distance. When it is determined to be infeasible, an audible and visual alarm is issued and the reason is displayed. Failure degradation: If the forward-looking perception fails, the jump determination is disabled and all actions are conservatively passed; if the suspension travel sensor fails, the phase determination is degraded to a combination of inertial measurement and wheel speed; if a single motor fails or torque is limited, the battery state of charge is too high (regenerative limitation) or the motor is overheated, the corresponding actuator limit value shrinks and is directly substituted into the feasible region for determination, so that the jump determination automatically becomes conservative, realizing endogenous degradation without the need for a dedicated logic branch.

[0151] The following example, Example 7, illustrates the lightweight ensemble method on a sampled model predictive control and learning dynamics model stack provided in this application:

[0152] In this embodiment of the invention, when the vehicle's autonomous driving stack consists of a learning dynamics model plus sampled model predictive control, the present invention can be used as a lightweight layer: for the calibrated fly slope cross section, the feasible region is reduced to a cross-section critical velocity. Before takeoff, calculate the current speed limit by using the available braking deceleration and remaining distance. (i.e., the dimensionality reduction special case of the analytical criterion in the vertical channel). The maximum braking deceleration is used as a hard constraint for approach speed planning; the airborne skills are carried out according to the stage domain gating adapter in Embodiment 4. This embodiment shows that the method of this application does not depend on a specific model morphology, and both analytical physical models and learning models can be implemented. This embodiment has been closed-loop verified in a high-fidelity vehicle dynamics simulation environment: according to the cross-section gating of this embodiment (adaptive early triggering of physical back-transmission, which expands the braking track from 3.35 meters to 16 to 24 meters), the recovery rate of continuous drop scene is increased from 16.7% to 40%; after adding the over-drop braking hold (braking across the edge of the edge by distance gating, and releasing after climbing over the edge at about 2.5 to 3 meters / second), it reaches 88.9% (8 / 9, close to the theoretical upper limit of about 96%), and all individuals complete the entire route; this result does not depend on wheel-by-wheel torque vector or air attitude actuator, proving that accessibility gating has an independent technical effect. As a comparative example seven, all variations of the cost function-side speed modulation (global speed tracking weights 0.5 / 4 / 8, target speed reduced from 10 to 6, and gradient gating reference speed upper limit of 6.5 confirmed to be effective) were subjected to same-stack simulation tests: the peak speed before the hurdle increased instead of decreasing (from approximately 11.7 to 12.2 m / s) or decreased only slightly, the completion rate decreased from 2 / 8 to 0 / 8, and new upstream instability occurred—proving that the soft-cost path cannot achieve deceleration before the hurdle in this type of architecture. This is the experimental basis for the present invention to use verifiable hard constraint gating instead of soft-cost penalty. The above data are all simulation and forced intervention probe calibrators (judgment and braking intervention are executed by the gating logic), which proves the effectiveness of the gating logic itself. In this field, it is common engineering practice to verify the relative effect of vehicle dynamics control schemes with high-fidelity simulation and to implement them with real vehicle calibration. The above simulation comparison was conducted under the same initial conditions and random seed pairing, and the difference in effect can be attributed to the gating feature itself.

[0153] It must be clearly stated that the two sets of simulation data in this specification belong to different experimental purposes and measurement standards, and should not be directly compared or confused with each other. Example 4, comparing Example 5, shows that the "target route completion rate increased from 1 / 21 to 5 / 21," aiming to evaluate the route completion capability of the retrained learning model on all 21 random seeds. The denominator is the 21 seeds in the entire queue, and the vehicle is autonomously controlled by the learning model throughout the entire process without external intervention. In this example, the "recovery rate increased from 16.7% to 88.9% (8 / 9)," aiming to verify the effectiveness of the accessibility gating logic itself on a subset of drift problems. The denominator is the subset of seeds that drifted and triggered intervention, and it was measured under the forced intervention probe caliber (judgment and braking intervention are executed by the gating logic). The denominators, event definitions, and intervention modes of the two sets of data are different, complementing each other rather than mutually confirming the same indicator: the former confirms the improvement of autonomous learning capability by the phase isolation architecture, while the latter quantifies the rescue capability of accessibility hard gating under forced intervention.

[0154] Modification of this application

[0155] Those skilled in the art, upon considering the specification and practicing the disclosure of this application, will readily conceive of other embodiments of this application, including the use of different electric drive topologies, different reachability solution methods, different rolling time-domain optimization solvers, different combinations of phase determination methods, and different learning model backbones. The specification and embodiments are considered exemplary only.

[0156] As described above, the method of the present invention can be implemented well.

[0157] Compared with the prior art, the present invention has the following outstanding advantages and beneficial effects:

[0158] This invention addresses the physical conditions of uncontrollable airborne translation and strict limitations on in-flight correction authority due to angular momentum budget. It progressively reverses the landing safety constraints into executable torque control of the ground phase before takeoff, expanding the velocity and terrain envelope for safe flight paths, reducing the peak vertical impact upon landing, and shortening the airtime required to achieve the target landing attitude. Takeoff decisions are executed based on analytically verifiable reachability boundaries, proactively implementing safety actions when in-flight actuator capabilities are insufficient. The three phases are uniformly allocated by the same set of electric drives and maintain continuous transmission connection throughout, restoring full drive capability upon landing. The phase isolation architecture ensures that adding airborne attitude capability does not degrade the quality of ground phase control. Simulation comparison experiments show that while the target condition improves, the output of non-target conditions remains strictly consistent with the baseline. Furthermore, actuator failure, limited regeneration capability, or decreased perception confidence are all expressed as a contraction of constraints in reachability determination, automatically transformed from a reverse transmission chain into a more conservative takeoff decision, achieving an inherent graceful degradation. The dimensionality reduction implementation of the aforementioned accessibility gating has been supported by simulation closed-loop experiments: the recovery rate in continuous drop scenarios has increased from 16.7% to 88.9% (pure software, with judgment and intervention executed by gating logic); the concurrent simulation comparison experiment shows that all conventional variants of velocity modulation on the cost function side (global velocity tracking weight, target velocity adjustment, and slope gating reference velocity upper limit) cannot achieve deceleration before the drop (the peak velocity increases instead of decreasing), proving that the verifiable hard gating of the present invention is a necessary means to achieve deceleration before the drop, rather than an obvious alternative to cost parameter tuning.

[0159] Example 3

[0160] like Figures 14-17 As shown, this application embodiment provides a ramp attitude control system for an electric vehicle, employing the ramp attitude control method for electric vehicles described above. The system includes: multiple electric drive wheels, each with independently controllable driving torque and braking torque; a forward-looking perception device configured to acquire terrain information ahead of the electric vehicle's travel path; and a controller configured to include:

[0161] Terrain Pre-aiming and Fly Slope Prediction Module 301: Based on forward-looking perception data, detect the characteristics of the fly slope ahead of the electric vehicle's driving path, and predict the take-off point and the airtime of the electric vehicle after passing the take-off point.

[0162] Landing Terminal Constraint Determination Module 302: Determines the terminal constraints for safe landing based on the terrain information of the landing area;

[0163] Aerial backward reachability analysis module 303: Based on the attitude dynamics of the electric vehicle during takeoff, the torque limits of multiple electric drive wheels, and the takeoff duration, the feasible region of takeoff conditions is determined in reverse with terminal constraints as terminal conditions.

[0164] Pre-jump rolling time domain optimization module 304: During the ground driving phase before the electric vehicle arrives at the take-off point, the torque of multiple electric drive wheels is controlled so that the actual take-off state of the electric vehicle when it arrives at the take-off point falls into the feasible region of take-off conditions.

[0165] Take-off determination and safety action module 305: When the feasible region of take-off conditions is an empty set, or when the actual take-off state cannot fall into the feasible region of take-off conditions within the actuator capability and remaining travel distance of the electric vehicle, a safety action is executed.

[0166] Three-phase unified torque distribution module 306: The ground, airborne and landing phases are distributed by the same group of distributed electric drives according to the same optimized distribution framework;

[0167] Airborne attitude control and landing buffer module 307: During the airborne period of the electric vehicle, the torque of multiple electric drive wheels is controlled to generate a reaction torque so that the attitude of the electric vehicle tends to meet the terminal constraints.

[0168] Phase isolation architecture constraint module 308: The three-phase control is implemented by mutually isolated control subsystems, and is coupled only through terminal constraints, the feasible domain of the start-up condition and the continuous condition of the phase switching state.

[0169] Specifically, in a specific embodiment of the present invention, a system for specifically performing the above method is also provided, comprising:

[0170] like Figure 15 As shown, the fly-slope cross-phase attitude cooperative control system provided by this invention is implemented by modules M1 to M8 (M1 is terrain pre-aiming and fly-slope prediction, M2 is landing terminal constraint determination, M3 is aerial backward reachability analysis, M4 is pre-jump roll time-domain optimization, M5 is jump determination and safety action, M6 is three-phase unified torque distribution, M7 is take-off attitude control and landing buffer, and M8 is phase isolation architecture constraint). The descriptions of each module are as follows:

[0171] Terrain Preview and Fly Slope Prediction Module M1: Constructing Elevation Maps Based on Forward-Looking Perception Detect the characteristics of the jump ahead of the driving path and predict the take-off point and the angle of the slope. And the duration of takeoff. This step includes the following sub-steps:

[0172] M101, Slope Edge Detection: Detects abrupt changes in elevation gradient and convex edges along the planned path; Slope edge detection can be achieved by any one or a combination of geometric detection, learning-based terrain classification, and high-precision map priors.

[0173] M102, Ballistics and Airborne Duration Prediction: Assuming takeoff speed is... ,but The center of mass trajectory is levitation time Depend on The numerical root is used to obtain the result; the takeoff time can also be given by a calibration lookup table or a learning model, and recursively corrected online by inertial measurements after takeoff; for A confidence interval is given based on the slope of the landing surface, and the lower bound of the interval is used for subsequent accessibility assessment to ensure conservatism.

[0174] Landing terminal constraint determination module M2: Constructs landing terminal constraints by fitting the elevation of the landing point's neighborhood with the pitch and roll slope of the landing surface. Its content includes at least one of the following: target pitch angle range, target roll angle range, upper limit of landing angular velocity, vertical landing speed range, and wheel contact sequence. It means that the vehicle body attitude fits the ground, multiple wheels contact the ground almost simultaneously, the landing angular velocity is close to zero, and the vertical landing speed falls within the suspension buffer window.

[0175] Aerial Rear Accessibility Analysis Module M3: Vehicle Attitude Dynamics During Takeoff That is, the reaction torque of the wheels during acceleration and deceleration drives the vehicle body to rotate; the exchangeable angular momentum of the wheels is subject to budget constraints. .by To determine the feasible region of takeoff conditions in reverse under the torque limit and angular momentum budget, for the terminal. This step includes the following sub-steps:

[0176] M301, Pitch Axis Analytical Criterion: Let , , thus obtaining a dual integrator ;remember , , Then, in terms of the duration of levitation The necessary and sufficient condition for the target to reach the landing pitch angle is: and and superimposed This criterion can be determined online in milliseconds and the jump determination boundary can be drawn directly.

[0177] M302, Triaxial General Case: For implementations with independent steering, backward reachability is solved according to the above dynamics. Any one of Hamilton-Jacobi backward reachability, ellipsoid or multicellular approximation, or sampling approximation based on inverse numerical integration can be used.

[0178] M303 Conservatism of the reachable region of learning: When the feasible region is given by the learning model, a conservative inner approximation after shrinking the calibrated residual is taken, and it forms a dual gating with the analytical criterion. Jumping can only be initiated when both criteria are simultaneously determined to be reachable.

[0179] Pre-jump roll time-domain optimization module M4: Constrained by ground phase dynamics, torque limits, and adhesion constraints, it calculates the actual takeoff state upon reaching the takeoff point. Assuming terminal constraints, solve for the motor torque sequence; the objective function is: The speed is regulated by the total torque, and the takeoff pitch angle and pitch velocity are regulated by the torque difference between the front and rear axles. After load transfer adjustment; the solution is obtained using model predictive control or sampled model predictive control, executed in a rolling manner from 50 to 100 Hz; in driver-in-loop mode, the output of this step is the minimum intervention correction amount for the torque requested by the driver.

[0180] Takeoff Decision and Safety Action Module M5: Each control cycle checks whether there is a feasible solution within the speed window and remaining distance that allows the takeoff state to enter the feasible region; if the feasible region is empty or unreachable, a safety action is executed (at least one of deceleration to abandon takeoff, change of driving path, or deceleration to pass through); the safety action trigger threshold and feasible region shrinkage margin are set according to the operating mode. The operating modes include driver-in-the-loop mode and automatic driving mode. In the automatic driving mode, the safety action is automatically selected by the behavior planning device and does not respond to driver confirmation.

[0181] Three-phase unified torque distribution module M6: with ,constraint Together with torque and speed limits, a unified secondary planning is formed, and ground phase is distributed according to the target longitudinal force and adhesion constraints are applied. The takeoff phase is allocated according to the target vehicle's torque, and the landing phase is allocated according to the target wheel speed and buffer torque. Different phases only switch. , and .

[0182] M7 Airborne Attitude Control and Landing Buffer Module: The inner loop of the airborne system tracks the attitude reference trajectory at 200 to 500 Hz, and the control law is... The output is uniformly allocated by the secondary planning; after takeoff, the remaining flight time is updated online based on the measured status and the reachability is recalculated on a rolling basis. When the terminal constraint becomes unreachable, it switches to the attitude with the smallest deviation in the set of reachable attitudes; the transmission connection between the power source and the drive wheels is kept open throughout the takeoff process; after touchdown, the torque of the wheels that touch down first is limited, while the remaining wheels are kept in the air and the torque ramp is limited within a preset time to complete the landing buffer.

[0183] Phase isolation architecture constraint module M8: The control of the three phases of ground, takeoff, and landing is implemented by mutually isolated control subsystems. The subsystems are coupled only through landing terminal constraints, takeoff condition feasible domains, and phase switching state continuous conditions. In the learning embodiment, the takeoff-related learning model consists of a frozen shared backbone model and a low-rank residual adapter gated to the takeoff phase domain. The adapter only outputs non-zero residuals for inputs and outputs with preset stage domain identifiers. The projected weights are zero-initialized, and the gate domain identifiers are persistently stored in the model weight file. During runtime, the route is precisely matched according to the scene identifier, and the unconditional mode is rolled back when the domain identifier goes out of bounds.

[0184] Example 4

[0185] This application provides an electronic device, including: a memory storing computer programs or instructions thereon; and a processor for executing the computer programs or instructions in the memory to implement the above-described method for controlling the attitude of an electric vehicle during ramp crossing.

[0186] Example 5

[0187] like Figure 3 As shown, this application provides a vehicle that includes multiple electric drive wheels, the driving torque and braking torque of each electric drive wheel being independently controlled by a controller, and also includes: electronic equipment as described above, or a ramp attitude control system for electric vehicles as described above.

[0188] Example 6

[0189] This application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of a method for controlling the attitude of an electric vehicle during ramp crossing.

[0190] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0191] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for controlling the attitude of an electric vehicle during a ramp, wherein the electric vehicle comprises multiple electric drive wheels, and the driving torque and braking torque of each electric drive wheel are independently controlled by a controller, characterized in that, The method includes: Based on forward-looking perception data, the characteristics of the ramp ahead of the electric vehicle's driving path are detected, and the take-off point and the airtime of the electric vehicle after passing the take-off point are predicted. Determine the terminal constraints for safe landing based on the terrain information of the landing area; Based on the attitude dynamics of the electric vehicle during takeoff, the torque limits of the multiple electric drive wheels, and the takeoff duration, the terminal constraint is used as the terminal condition for step-by-step reverse propagation to determine the feasible region of takeoff conditions in reverse. During the ground driving phase before the electric vehicle arrives at the take-off point, the torque of the multiple electric drive wheels is controlled so that the actual take-off state of the electric vehicle when it arrives at the take-off point falls within the feasible region of the take-off conditions. During the airborne phase of the electric vehicle, the torque of the plurality of electrically driven wheels is controlled to generate a reaction torque, causing the electric vehicle's posture to tend to satisfy the terminal constraint.

2. The method for controlling the attitude of an electric vehicle during a ramp as described in claim 1, characterized in that, The method further includes: The terminal constraint is used to limit the target attitude range and target speed range when the electric vehicle lands safely. The feasible domain of the take-off conditions is a set of the following take-off states: starting from the take-off state, during the airborne period, the electric vehicle can satisfy the terminal constraint in attitude through the reaction torque generated by the torque of the multiple electric drive wheels. The take-off state includes at least the speed and attitude parameters of the electric vehicle at the take-off point.

3. The method for controlling the attitude of an electric vehicle during a ramp as described in claim 1 or 2, characterized in that, The method further includes: When the feasible region of the take-off condition is an empty set, or when the actuator capability and remaining travel distance of the electric vehicle cannot make the actual take-off state fall into the feasible region of the take-off condition, a safety action is performed. The safety action includes at least one of decelerating and abandoning the take-off, changing the travel path, and decelerating through the flyover feature.

4. The method for controlling the attitude of an electric vehicle during a ramp as described in claim 3, characterized in that, The method further includes: setting the trigger threshold of the safety action and the contraction margin of the feasible region of the take-off condition according to the operating mode, wherein the operating mode includes at least one driver-in-the-loop mode and an automatic driving mode; in at least one of the driver-in-the-loop modes, in response to the driver's confirmation operation of the safety prompt, no forced deceleration is performed, but attitude control and landing buffer during the airborne period are retained; in the automatic driving mode, the safety action is automatically selected by the behavior planning device of the electric vehicle between decelerating and abandoning the take-off, changing the driving path, and decelerating to pass, and does not respond to the driver's confirmation operation.

5. The method for controlling the attitude of an electric vehicle during a ramp as described in any one of claims 1 or 4, characterized in that, The feasible region for the take-off condition is determined by at least one of the following methods: analytical reachability criterion based on saturation switching control, Hamilton-Jacobi backward reachability solution, ellipsoidal or multicellular approximation, sampling approximation based on inverse numerical integration, and conservative inner approximation of the learning model output after shrinking by calibration residuals; wherein, when the conservative inner approximation is used, the take-off condition is determined only when the conservative inner approximation and the analytical reachability criterion are both passed.

6. The method for controlling the attitude of an electric vehicle during a ramp as described in claim 1 or 2, characterized in that, When determining the feasible region of the take-off conditions, a wheel angular momentum budget constraint is also applied. The wheel angular momentum budget is determined by the moment of inertia of each of the electric drive wheels and the margin between their current speed and the speed limit, in order to limit the changeable amount of the vehicle body angular momentum during take-off.

7. The method for controlling the attitude of an electric vehicle during a ramp as described in claim 1 or 2, characterized in that, The wheel torque during the ground driving phase, the airborne phase, and the landing phase is allocated to the same group of multiple electric drive wheels by the same optimization allocation framework. The optimization allocation framework only switches the equality constraints and cost weights at different phases: the ground driving phase is allocated according to the target longitudinal force, the airborne phase is allocated according to the target vehicle body torque, and the landing phase is allocated according to the target wheel speed and buffer torque.

8. The method for controlling the attitude of an electric vehicle during a ramp as described in claim 1, 2, or 7, characterized in that, The airborne phase also includes a neutral wheel speed rearrangement step: applying a torque combination where the sum of the torques of each wheel is zero, minimizing the deviation between the rotational speed of each of the electrically driven wheels and the rotational speed corresponding to the expected landing speed without changing the vehicle's attitude.

9. The method for controlling the attitude of an electric vehicle during a ramp as described in claim 1 or 2, characterized in that, Throughout the entire period of the electric vehicle's levitation, the transmission connection between the power source and the plurality of electric drive wheels is kept uninterrupted, and the attitude control is achieved solely by controlling the bidirectional torque of the plurality of electric drive wheels.

10. The method for controlling the attitude of an electric vehicle during a ramp as described in claim 1 or 2, characterized in that, The torque control during the ground driving phase is solved by roll time domain optimization. The roll time domain optimization takes the actual take-off state falling into the feasible region of the take-off condition as a constraint, including model predictive control or sampled model predictive control. The roll time domain optimization adjusts the speed in the take-off state through the total torque, and adjusts the pitch angle and pitch velocity in the take-off state through the load transfer caused by the torque difference between the front axle and the rear axle.

11. The method for controlling the attitude of an electric vehicle during a ramp as described in claim 1 or 10, characterized in that, The torque control during the ground driving phase is a correction to the torque requested by the driver, wherein the correction is the minimum intervention amount that brings the actual take-off state into the feasible region of the take-off conditions.

12. The method for controlling the attitude of an electric vehicle during a ramp as described in claim 1 or 2, characterized in that, The control of the ground driving phase, the takeoff phase, and the landing phase are executed by mutually isolated control subsystems. The control subsystems are coupled only through the terminal constraints, the feasible region of the takeoff conditions, and the state continuity conditions at the phase switching time, so that the attitude control during the takeoff phase does not change the control characteristics of the ground driving phase.

13. The method for controlling the attitude of an electric vehicle during a ramp as described in claim 1 or 12, characterized in that, At least one of the control subsystems includes a learning model, wherein the trainable parameters of the learning model related to takeoff and the trainable parameters of the learning model related to the ground driving phase are isolated from each other.

14. The method for controlling the attitude of an electric vehicle during a ramp as described in claim 13, characterized in that, The learning model includes a shared backbone model and a residual adapter: the residual adapter outputs a non-zero residual only for input samples with a preset stage domain identifier and zero for other input samples; the shared backbone model and the domain representation parameters remain frozen during the training of the residual adapter. The residual adapter is a low-rank bottleneck structure, and its projected weights are initialized to zero at the start of training. The preset stage domain identifier is stored as a persistent parameter in the model weight file. During runtime, based on the precise matching result between the driving scene identifier and the preset scene set, the matching scene is routed to the preset stage domain identifier, and the unmatched scene is routed to the unconditional domain. When the scene domain identifier exceeds the number of domains supported by the learning model, it is treated as an unconditional domain.

15. The method for controlling the attitude of an electric vehicle during a ramp as described in claim 1, 2, 6, 9, 11, or 14, characterized in that, It also includes at least one of the following features: the terminal constraint includes at least one of the following: target pitch angle range, target roll angle range, upper limit of landing angular velocity, vertical landing velocity range, and wheel contact sequence; at least some of the plurality of electric drive wheels have independently controllable steering angles, and the wheel spin axis direction is changed during takeoff by a combination of torque and steering angle, so that the reaction torque is generated in at least two directions: pitch axis and roll axis; before the electric vehicle arrives at the takeoff point, the damping and / or height of the semi-active or active suspension are preset to offset the actual takeoff state and / or relax the terminal constraint; the rotational speed of each of the electric drive wheels is maintained in the middle range of its rotational speed range to simultaneously retain wheel angular momentum adjustment in both directions during takeoff. The margin; the determination of takeoff and ground contact is based on at least one of the following: vertical acceleration threshold, suspension travel, sum of estimated wheel loads for each wheel, wheel acceleration change rate, and a calibration lookup table or learning classification model indexed by vehicle speed, suspension travel, and vertical acceleration; when the motor of any of the electric drive wheels fails or has limited torque, regenerative braking capability is limited, or perception confidence decreases, the corresponding actuator limit contraction or prediction interval expansion is substituted into the determination of the takeoff condition feasible region, so that the takeoff condition feasible region is correspondingly reduced; and during takeoff, the remaining flight time is updated online based on the measured motion state and reachability is re-determined. When the terminal constraint becomes unreachable, the attitude control target during takeoff is switched to the attitude with the smallest deviation from the terminal constraint in the set of reachable attitudes.

16. A ramp attitude control system for an electric vehicle, employing the ramp attitude control method for an electric vehicle as described in any one of claims 1-15, the system comprising: Multiple electrically driven wheels, the driving torque and braking torque of each of the electrically driven wheels can be controlled independently; A forward-looking perception device is configured to acquire terrain information ahead of the electric vehicle's driving path; And a controller, characterized in that the controller is configured to include: Terrain Prediction and Fly Slope Prediction Module: Based on forward-looking perception data, detect the features of the fly slope ahead of the electric vehicle's driving path, and predict the take-off point and the airtime of the electric vehicle after passing the take-off point. Landing Terminal Constraint Determination Module: Determines the terminal constraints for safe landing based on the terrain information of the landing area; Aerial backward reachability analysis module: Based on the attitude dynamics of the electric vehicle during takeoff, the torque limits of the multiple electric drive wheels, and the takeoff duration, the feasible region of takeoff conditions is determined in reverse with the terminal constraints as the terminal conditions. Pre-jump rolling time domain optimization module: During the ground driving phase before the electric vehicle arrives at the jump point, the torque of the multiple electric drive wheels is controlled so that the actual jump state of the electric vehicle when it arrives at the jump point falls into the feasible region of the jump conditions. Airborne attitude control and landing buffer module: During the airborne period of the electric vehicle, the torque of the multiple electric drive wheels is controlled to generate a reaction torque, so that the attitude of the electric vehicle tends to meet the terminal constraint. Also includes: Take-off determination and safety action module: When the feasible region of the take-off condition is an empty set, or when the actuator capability and remaining travel distance of the electric vehicle cannot make the actual take-off state fall into the feasible region of the take-off condition, a safety action is performed. Three-phase unified torque distribution module: The ground, airborne, and landing phases are allocating torque by the same group of distributed electric drives according to the same optimized distribution framework; Phase isolation architecture constraint module: The three-phase control is implemented by mutually isolated control subsystems, and is coupled only through terminal constraints, the feasible domain of the start-up condition and the continuous condition of the phase switching state.

17. An electronic device, characterized in that, include: A memory on which computer programs or instructions are stored; A processor for executing the computer program or instructions in the memory to implement the ramp attitude control method for an electric vehicle as described in any one of claims 1 to 15.

18. A vehicle, characterized in that, The vehicle includes: the electronic device as claimed in claim 17, or the ramp attitude control system for an electric vehicle as claimed in claim 16.

19. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method for controlling the attitude of an electric vehicle in a ramp according to any one of claims 1 to 15.

Images