A control method and system for cross-domain fusion of a near-earth vehicle

Abstract

The application discloses a kind of control method and system of cross-domain fusion of temporary ground vehicle, it is related to automatic driving and intelligent traffic technical field. Including: pre-establishe the cross-domain behavior rule base of vehicle dynamics parameter constraint in vehicle controller;Trigger domain switching event, controller is activated simultaneously in the same control period in response to domain switching event multiple rules, based on rule base control vehicle executes domain switching behavior;In execution process, controller periodically obtains vehicle pose information, respectively according to the multiple motion actions of activated multiple rules prediction next time, and with corresponding rule is matched, if matching is successful then using current pose information iteratively updates the behavior rule in rule base, realizes the collaborative decision of multiple rules and smooth transition.The application simulates human cross-modal decision-making thinking, using the control architecture of multiple rules parallel activation and collaborative decision-making, realizes the smooth, safe, adaptive intelligent control and switching of temporary ground vehicle between different domains.

Description

Technical Field

[0001] This invention relates to the fields of autonomous driving and intelligent transportation technology, and more specifically to a control method and system for cross-domain integration of near-ground vehicles. Background Technology

[0002] Ground-based vehicles (such as flying cars and vertical takeoff and landing aircraft) can operate flexibly in roads and airspace and are an important component of future intelligent transportation systems. However, existing control methods are mostly designed for a single domain (ground or air) and lack cross-domain integrated control mechanisms, resulting in problems such as response delays, control instability, and poor safety during mode switching. For example, traditional ground-based autonomous driving systems struggle to effectively handle dynamic environmental changes during takeoff and landing, while airborne flight control systems are ill-suited to adapting to ground traffic rules and obstacle avoidance. Furthermore, takeoff and landing transients involve multi-physics coupling, and current methods often rely on preset trajectories, lacking adaptive optimization capabilities.

[0003] To address the aforementioned issues, the industry has begun exploring rule-based autonomous driving control methods. For example, authorized patents CN111290400B, CN111267852B, CN111267858B, and CN111290399B disclose a series of fleet cooperative driving control methods. The core idea is to pre-establish a behavioral rule base for specific driving events (such as splitting, cruising, lane changing, and platooning); when the event trigger conditions are met, the controller controls the vehicle to perform the corresponding driving behavior based on the rule base; during execution, motion actions are predicted by acquiring pose information, and the rules are iteratively updated after matching with the rule base. These methods, by simulating the decision-making process of human drivers, improve the control performance of autonomous vehicles in single scenarios to a certain extent.

[0004] Besides the rule-based autonomous driving methods mentioned above, existing technologies for cross-domain control of near-ground vehicles such as flying cars and vertical takeoff and landing aircraft mostly adopt a "hard switching, segmented control, and preset trajectory" approach. Specifically, these methods typically treat ground driving and flight as two independent modes, directly switching control logic through preset switching points (such as reaching a certain speed or altitude), or pre-planning fixed trajectories for takeoff and landing phases, executing open-loop control according to a predetermined path. For example, some existing solutions directly disconnect the ground control loop and connect the flight control loop after accelerating to the predetermined takeoff speed on the ground; during the landing phase, they rely on a preset glide path for guidance, lacking real-time response capabilities to dynamic environmental changes.

[0005] The above technical solution has the following inherent defects: First, abrupt changes in control parameters can cause switching oscillations. The "hard switching" method directly replaces controller parameters at the moment of mode transition, lacking a transition phase. This makes it prone to causing vehicle attitude oscillations or trajectory deviations due to sudden changes in control quantities. Especially at takeoff, the discontinuity in parameters between ground-based drive control and in-flight rotor control can lead to sudden changes in pitch angle or thrust distribution imbalances, affecting flight safety.

[0006] Second, segmented control lacks unified coordination. The ground segment, takeoff and landing segment, and air segment each employ different control logics, resulting in a lack of information sharing and target coordination between the segments. This makes it impossible to simultaneously address the dual requirements of "safely exiting the current domain" and "safely accessing the target domain" during the handover process. For example, during takeoff, only the climb trajectory tracking is considered, neglecting the collision risk from vehicles approaching from behind on the ground, or ignoring the conflict threat from existing aircraft in the air.

[0007] Third, preset trajectories are difficult to adapt to dynamic environments. Open-loop control methods based on preset trajectories cannot dynamically adjust according to real-time perceived obstacles, environmental vehicles, and weather conditions. When the actual operating environment deviates from the preset conditions (such as encountering sudden crosswinds or the landing point being occupied), the preset trajectory method lacks adaptive capability, which can easily lead to deviations from safety boundaries or even accidents.

[0008] Fourth, there is a lack of multi-rule parallel collaboration and iterative optimization capabilities. Existing methods have not constructed an architecture that can simultaneously activate three types of control rules—ground, air, and takeoff / landing—and make collaborative decisions, nor do they possess the ability to iteratively optimize the rule base based on real-time pose information, thus failing to achieve continuous evolution and adaptive optimization of control strategies.

[0009] In summary, existing cross-domain control technologies for ground-based vehicles generally suffer from problems such as switching oscillations, low safety, and poor adaptability, making it difficult to meet the high requirements of future intelligent transportation systems for cross-domain integrated control. It is necessary to provide a cross-domain integrated control method that can achieve unified rule description, parallel event response, and multi-rule collaborative decision-making, so as to achieve seamless switching and optimized control of ground-based vehicles in ground driving, air flight, and take-off and landing transients.

[0010] Therefore, proposing a control method and system for cross-domain integration of near-ground vehicles to solve the difficulties existing in the prior art is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0011] In view of this, the present invention provides a control method and system for cross-domain integration of temporary ground vehicles, which solves the problems of inconsistent control and unsmooth switching of temporary ground vehicles in the ground, air and take-off and landing transient states.

[0012] To achieve the above objectives, the present invention provides the following technical solution: A control method for cross-domain fusion of near-ground vehicles includes the following steps: S1. Establish a cross-domain behavior rule base in the controller in advance; S2. When the vehicle is in the current domain and the triggering conditions of the domain switching event are met, the domain switching event is triggered. S3. In response to the triggering of a domain switching event, the controller simultaneously activates multiple types of rules in the cross-domain behavior rule base related to the current operating state within the same control cycle. S4. When performing domain switching behavior, the controller obtains the vehicle's pose information and predicts multiple motion actions of the vehicle at the next moment based on the pose information and various rules in the cross-domain behavior rule base. S5. The controller determines whether the predicted multiple motion actions match the corresponding multiple rules. If they do not match, the controller generates the final control command through priority arbitration and weighted fusion. If they match, the controller uses the current pose information to iteratively update the behavior rules in the cross-domain behavior rule library, thereby realizing collaborative decision-making and smooth transition of multiple rules.

[0013] Optionally, a cross-domain behavior rule library is pre-established in the controller in S1, including ground driving behavior rules, air flight behavior rules, and takeoff and landing transient behavior rules constrained by vehicle dynamics parameters.

[0014] Optionally, the cross-domain behavior rule base adopts a condition-action-constraint triplet structure for unified description and storage. The condition set includes the combination logic of the vehicle's own state and environmental events, the action set includes the control instructions output to the actuator module, and the constraint set is composed of vehicle dynamic parameters and environmental safety constraints. The cross-domain behavior rule base is organized into three layers according to the operating mode. The rules in each layer are stored in the form of a state-behavior mapping table, and the condition matching is achieved with high timeliness through hash index.

[0015] Optionally, the iterative update of the cross-domain behavior rule base is based on the error feasible domain boundary. The update trigger condition is: within multiple consecutive control cycles, the weighted comprehensive error between the predicted action of all activated rules and the actual state of the vehicle is lower than the dynamic threshold. The dynamic threshold is adjusted in real time based on the current speed and environmental complexity.

[0016] Optionally, S2 domain handover events include takeoff events (switching from ground domain to air domain) and landing events (switching from air domain to ground domain). Domain switching events are triggered by passive and active reasons. Passive reasons include vehicle being affected by obstacles, environmental changes, and system failures requiring an emergency domain switch. Active reasons include changes in vehicle destination, route planning requirements, and user commands. When a domain switching event is triggered, the controller updates the vehicle's domain status information and adjusts the control parameters to adapt to the new domain. The control parameters include power distribution parameters, control loop gain, and actuator mode.

[0017] Optionally, when performing domain switching behavior in S4, the vehicle pose information obtained by the controller includes the vehicle's own running status information, obstacle information, and environmental vehicle information. The vehicle's operating status information includes speed, acceleration, position, heading, vehicle number, pitch angle, current domain, and domain status; Obstacle information includes the relative direction and relative distance of the obstacle; Environmental vehicle information is obtained through communication with other vehicles, including the speed, acceleration, position, heading, vehicle number, current domain, and domain status of other vehicles within the communication range.

[0018] Optionally, in S5, the controller determines whether the predicted multiple motion actions match the corresponding multi-class rules. If they do not match, the controller generates the final control command through priority arbitration and weighted fusion. When the degree of mismatch exceeds the preset safety boundary, the controller enters the graded safety protection mode and executes preset emergency actions according to the current operating mode; Priority arbitration dynamically sets the weight coefficients of multiple rules in each cross-domain behavior rule base according to the current operation stage. Weighted fusion is based on the safety boundary and combines the weight coefficients to generate a compromise control quantity through convex combination. When multi-rule prediction actions conflict and weighted fusion fails to generate control instructions that satisfy all constraints, a hierarchical safety protection mechanism is activated. The tiered security protection mechanism includes: locking the conflict dimension to the limit value corresponding to the most stringent constraint in the event of a local conflict, and triggering a security fallback strategy to execute preset emergency actions in the event of a global conflict.

[0019] Optionally, the specific details of implementing collaborative decision-making and smooth transition for multiple rule types in S5 are as follows: Construct a joint error feasible domain that integrates ground environment, air environment, and takeoff and landing transient safety constraints; The model predictive control algorithm is adopted. Under the constraint of the joint error feasible region, the objective function is to minimize the comprehensive error between multiple motion actions predicted by multiple rules in the cross-domain behavior rule library and the actual pose of the vehicle. The parameters of relevant behavior rules in the cross-domain behavior rule library are optimized and updated.

[0020] A control system for cross-domain fusion of temporary ground vehicles, and a control method for cross-domain fusion of temporary ground vehicles applying any of the above, comprising: a controller, adopting an architecture of multi-rule parallel activation and collaborative decision-making, utilizing the parallel processing of prediction and matching of multiple types of rules within the same control cycle, thereby controlling the behavior switching and optimization of the vehicle in ground driving, air flight and take-off and landing transients; The vehicle sensor is connected to the first input terminal of the controller and is used to collect the vehicle's own pose information and surrounding obstacle information; The communication module is connected to the second input terminal of the controller and is used to obtain environmental vehicle information through vehicle-to-everything (V2X) or aviation communication. The execution layer, connected to the first output of the controller, is used to execute ground drive, rotor drive, steering, and braking actions according to the control commands generated by the controller. The storage module, connected to the second output of the controller, is used to store a pre-established cross-domain behavior rule library.

[0021] Optionally, the controller has a built-in domain switching configuration unit, behavior prediction unit, and rule matching unit; The configuration unit, connected to the input of the behavior prediction unit, is used to respond to the triggering of domain switching events, simultaneously activate multiple types of rules related to the current running state in the cross-domain behavior rule base within the same control cycle, and dynamically set the priority weight coefficients of various rules according to the current running stage. The behavior prediction unit is connected to the input of the rule matching unit. It is used to predict multiple motion actions of the vehicle in the next moment based on the real-time pose information of the vehicle and the activated multiple rules in parallel, and output multiple sets of predicted trajectories to the rule matching unit. The rule matching unit is used to determine whether the multiple motion actions output by the behavior prediction unit match the corresponding multi-class rules. If they match, the behavior rules in the cross-domain behavior rule library are iteratively updated using the current pose information. If they do not match, the final control command is generated through priority arbitration and weighted fusion, and a hierarchical safety protection mechanism is activated when the conflict cannot be resolved.

[0022] As can be seen from the above technical solution, compared with the prior art, the present invention provides a control method and system for cross-domain fusion of near-ground vehicles, which has the following beneficial effects: (1) This invention draws on human multimodal decision-making thinking, and by establishing a cross-domain behavior rule base and adopting a control architecture of multi-rule parallel activation and collaborative decision-making, it realizes unified control of ground vehicles in ground driving, air flight and take-off and landing transients; unlike the existing technology that adopts the control mechanism of "single event triggering single rule and serial execution" for a single driving event (such as convoy splitting, cruising, lane changing, platooning), this invention fundamentally solves the problems of non-smooth switching and inconsistent control in the existing methods; (2) This invention uses periodic behavior prediction and rule matching, and uses actual pose information to iteratively update the cross-domain behavior rule library, so that the control rules can be dynamically optimized according to the operating environment and vehicle status; in particular, the iterative update of the cross-domain behavior rule library is triggered only when the predicted actions of all activated rule modules match the actual state, which ensures the stability and continuity of rule evolution and improves the system's adaptability. (3) This invention introduces predictive control with error feasible domain constraints, constructs a joint error feasible domain that integrates ground, air and take-off and landing transient safety constraints, and ensures that the rule iteration update process is always carried out within the boundaries of vehicle dynamic parameters and environmental safety constraints; by integrating the optimization objective with the safety boundary, it fundamentally prevents the generation of risky strategies that exceed vehicle capabilities or environmental tolerances in pursuit of performance, and takes into account both control performance and operational safety. (4) The multi-rule parallel activation and collaborative decision-making architecture proposed in this invention not only solves the inherent problem of cross-domain control of near-ground vehicles, but also provides a new and scalable paradigm for the unified control of multimodal mobile robots (such as amphibious vehicles, air and sea cross-domain vehicles, etc.). Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0024] Figure 1 A schematic diagram of a control system structure for cross-domain fusion of near-ground vehicles provided by the present invention; Figure 2 A flowchart of a control method for cross-domain fusion of near-ground vehicles provided by the present invention; Figure 3 This is a schematic diagram of a cross-domain operation scenario for a temporary vehicle provided by the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] See Figure 2 As shown, this invention discloses a control method for cross-domain fusion of near-site vehicles, comprising the following steps: S1. Establish a cross-domain behavior rule base in the controller in advance; S2. When the vehicle is in the current domain and the triggering conditions of the domain switching event are met, the domain switching event is triggered. S3. In response to the triggering of a domain switching event, the controller simultaneously activates multiple types of rules in the cross-domain behavior rule base related to the current operating state within the same control cycle. S4. When performing domain switching behavior, the controller obtains the vehicle's pose information and predicts multiple motion actions of the vehicle at the next moment based on the pose information and various rules in the cross-domain behavior rule base. S5. The controller determines whether the predicted multiple motion actions match the corresponding multiple rules. If they do not match, the controller generates the final control command through priority arbitration and weighted fusion. If they match, the controller uses the current pose information to iteratively update the behavior rules in the cross-domain behavior rule library, thereby realizing collaborative decision-making and smooth transition of multiple rules.

[0027] Furthermore, S1 pre-establishes a cross-domain behavior rule library in the controller, including ground driving behavior rules, air flight behavior rules, and takeoff and landing transient behavior rules constrained by vehicle dynamics parameters.

[0028] Furthermore, the cross-domain behavior rule base adopts a condition-action-constraint triplet structure for unified description and storage. The condition set includes the combination logic of the vehicle's own state and environmental events, the action set includes the control instructions output to the actuator module, and the constraint set is composed of vehicle dynamic parameters and environmental safety constraints. The cross-domain behavior rule base is organized into three layers according to the operating mode. The rules in each layer are stored in the form of a state-behavior mapping table, and the condition matching is achieved with high timeliness through hash index.

[0029] Furthermore, the iterative update of the cross-domain behavior rule base is based on the error feasible domain boundary. The update trigger condition is: within multiple consecutive control cycles, the weighted comprehensive error between the predicted actions of all activated rules and the actual state of the vehicle is lower than the dynamic threshold. The dynamic threshold is adjusted in real time based on the current speed and environmental complexity.

[0030] Furthermore, S2 domain handover events include takeoff events (switching from ground domain to air domain) and landing events (switching from air domain to ground domain). Domain switching events are triggered by passive and active reasons. Passive reasons include vehicle being affected by obstacles, environmental changes, and system failures requiring an emergency domain switch. Active reasons include changes in vehicle destination, route planning requirements, and user commands. When a domain switching event is triggered, the controller updates the vehicle's domain status information and adjusts the control parameters to adapt to the new domain. The control parameters include power distribution parameters, control loop gain, and actuator mode.

[0031] Furthermore, in S4, when performing domain switching behavior, the vehicle pose information obtained by the controller includes the vehicle's own running status information, obstacle information, and environmental vehicle information; The vehicle's operating status information includes speed, acceleration, position, heading, vehicle number, pitch angle, current domain, and domain status; Obstacle information includes the relative direction and relative distance of the obstacle; Environmental vehicle information is obtained through communication with other vehicles, including the speed, acceleration, position, heading, vehicle number, current domain, and domain status of other vehicles within the communication range.

[0032] Furthermore, in S5, the controller determines whether the predicted multiple motion actions match the corresponding multi-class rules. If they do not match, the controller generates the final control command through priority arbitration and weighted fusion. When the degree of mismatch exceeds the preset safety boundary, the controller enters the graded safety protection mode and executes preset emergency actions according to the current operating mode; Priority arbitration dynamically sets the weight coefficients of multiple rules in each cross-domain behavior rule base according to the current operation stage. Weighted fusion is based on the safety boundary and combines the weight coefficients to generate a compromise control quantity through convex combination. When multi-rule prediction actions conflict and weighted fusion fails to generate control instructions that satisfy all constraints, a hierarchical safety protection mechanism is activated. The tiered security protection mechanism includes: locking the conflict dimension to the limit value corresponding to the most stringent constraint in the event of a local conflict, and triggering a security fallback strategy to execute preset emergency actions in the event of a global conflict.

[0033] Furthermore, the specific details of achieving collaborative decision-making and smooth transition among multiple rule types in S5 are as follows: Construct a joint error feasible domain that integrates ground environment, air environment, and takeoff and landing transient safety constraints; The model predictive control algorithm is adopted. Under the constraint of the joint error feasible region, the objective function is to minimize the comprehensive error between multiple motion actions predicted by multiple rules in the cross-domain behavior rule library and the actual pose of the vehicle. The parameters of relevant behavior rules in the cross-domain behavior rule library are optimized and updated.

[0034] Example 1: A cross-domain fusion control method for ground-based vehicles, applied to the controller of ground-based vehicles, is used to control and optimize the behavior switching of the vehicle during ground driving, aerial flight, and takeoff and landing transients. Ground-based vehicles can be flying cars, vertical takeoff and landing aircraft, etc., and have multi-modal operation capabilities.

[0035] The control method is as follows: 1. Design of Cross-Domain Behavior Rule Base A cross-domain behavior rule library is pre-established in the vehicle controller. The cross-domain behavior rule library includes ground driving behavior rules (such as lane keeping, following, and lane changing), air flight behavior rules (such as climb, descent, and hovering), and takeoff and landing transient behavior rules (such as takeoff acceleration and landing deceleration) constrained by vehicle dynamic parameters (such as mass, aerodynamic characteristics, and propulsion efficiency).

[0036] The vehicle's pose information includes its operating status, obstacle information, and environmental vehicle information, acquired through vehicle sensors (such as LiDAR, cameras, and IMU) and communication modules (such as V2X and aeronautical communications). The data formats for each type of information are defined as follows: (1) Self-state events:

[0037] Where ID is the vehicle number, Position is the position, Dir is the heading angle, Speed ​​is the speed, Acc is the acceleration, Pit is the pitch angle, Domain is the current domain (ground / air), and DomainState is the domain state (e.g., ground cruise, air climb).

[0038] (2) Obstacle events:

[0039] Wherein, ObID is the obstacle number, ObAng is the azimuth angle between the obstacle and the controlled vehicle, ObDis is the distance between the obstacle and the controlled vehicle, ObPosition is the spatial coordinates of the obstacle, ObDomain is the current domain of the obstacle, and ObDomainState is the domain state of the obstacle.

[0040] (3) Communication events: Com=[VehID,VehPosition,VehDir,VehSpeed,VehAcc,VehPit,VehDomain,VehDomainState] Wherein, VehID is the environment vehicle number, VehPosition is the position, VehDir is the heading angle, VehSpeed ​​is the speed, VehAcc is the acceleration, VehPit is the pitch angle, VehDomain is the current domain, and VehDomainState is the domain state.

[0041] The data formats for the above three types of events are summarized in Table 1.

[0042] Table 1. Definitions of Environmental Event Types and Formats

[0043] by Figure 3 Taking the ground domain scenario as an example, the controlled vehicle detects two environmental vehicles, Veh1 and Veh2, through sensors, generating obstacle events Obs01 and Obs02 respectively; simultaneously, it establishes communication with Veh1 and Veh2 through the communication module, generating communication events Com01 and Com02 respectively. Through the fusion of the above event information, the controller can obtain complete situational information of the controlled vehicle itself and its surrounding environment in real time, providing a data foundation for subsequent domain switching judgments and behavior control.

[0044] The cross-domain behavior rule base uses a condition-action-constraint triple structure for unified description and storage, and each rule is formalized as follows: ,in: The set of triggering conditions includes the combined logic of the vehicle's own state (such as speed, pitch angle, and current domain) and environmental events (such as obstacle distance and communication events); The control command set is output to the actuator module, including wheel drive torque, rotor speed, steering angle, braking pressure, etc. It is a set of dynamic and environmental constraints, consisting of parameters such as vehicle mass, aerodynamic coefficient, propulsion system response delay, minimum safe distance, and airspace boundary, to ensure that the generated commands do not exceed physical limits.

[0045] The cross-domain behavior rule base is organized in three layers according to operational modes (ground driving, air flight, and takeoff and landing transients). Rules within each layer are stored in the form of a state-behavior mapping table, and high-timeliness condition matching is achieved through hash indexes. Iterative updates of the cross-domain behavior rule base are based on the error feasible domain boundary. The update trigger condition is that within three consecutive control cycles, the weighted comprehensive error between the predicted actions of all activated rules and the actual vehicle state is lower than a dynamic threshold (the threshold is adjusted in real-time by the current speed and environmental complexity, with typical values ​​ranging from position error ≤ 0.2 m, speed error ≤ 0.5 m / s, and attitude angle error ≤ 2°). The update process employs a recursive least squares method with a forgetting factor, prioritizing the adjustment of rule parameters most relevant to the current error while retaining historical optimization results.

[0046] 2. Domain switching event triggering and control

[0047] When a vehicle is in the current domain and encounters a domain switching event, the controller determines whether the domain switching conditions are met (e.g., ground speed reaching the takeoff threshold, or air altitude decreasing to the landing threshold). In this invention, the domain switching conditions are quantitatively defined according to the specific switching event type, mainly including takeoff events and landing events: (1) Triggering conditions for takeoff events (ground domain → air domain) The controller determines that the takeoff conditions are met when all of the following conditions are met simultaneously: Speed ​​condition: Vehicle ground travel speed ,in, The minimum takeoff speed is determined based on the vehicle's aerodynamic characteristics and lift coefficient; a typical value is taken as... m / s; Height conditions: Current height of the vehicle Ensure that the landing gear or supporting structure does not come into contact with the ground. Pick m; Clearance conditions: clearance distance of the forward takeoff aisle Output from the forward obstacle detection module. Pick m, to ensure no obstacles intrude during takeoff; Power system status: Rotor system self-test normal, motor temperature... ( °C), battery level ( ); Airspace status: The V2X communication indicates that there are no other aircraft within ±30 m of the aircraft's takeoff lane in the target airspace, and the airspace control status is set to allow takeoff.

[0048] (2) Triggering conditions for landing events (air domain → ground domain)

[0049] The controller determines that the landing conditions are met when all of the following conditions are met simultaneously: Height conditions: Current height of the vehicle ,in The maximum altitude at which a landing procedure is allowed is typically set to [value missing]. m; Decline rate condition: Current vertical decline rate , A speed of 5 m / s is used to ensure a smooth and controllable descent process. Landing point conditions: The target landing point area is free of obstacles and the ground is flat. ( (m), and the landing point coordinates have been confirmed by RTK-GPS or visual guidance with an accuracy of ≤ 0.1m; Ground traffic conditions: V2X communication confirms that there are no dynamic ground vehicles within 20m of the landing point, or that ground vehicles have received and confirmed the avoidance instruction. System status: The remaining battery power is sufficient for landing and subsequent ground travel. The rotor and wheel drive system switching mechanism self-tested normally.

[0050] If the conditions are met, a domain switching event is triggered, and the controller controls the vehicle to perform the corresponding domain switching behavior based on the cross-domain behavior rule base.

[0051] Domain switching events are triggered by passive and active causes. Passive causes include obstacles, sudden weather changes, and system failures. Active causes include path planning updates, user commands, and energy management.

[0052] When a domain switching event is triggered, the controller updates the vehicle’s current domain and domain state, and adaptively adjusts control parameters (such as switching from wheel drive to rotor drive).

[0053] by Figure 3 Taking the takeoff scenario as an example, when the controlled vehicle (Host Veh) encounters an obstacle, Veh3, while cruising on the ground and cannot avoid it by changing lanes on the ground, the controller determines that the takeoff conditions are met and formally triggers the takeoff event. It should be noted that during takeoff, the controller does not employ the sequential control logic of "first exiting ground rules, then executing takeoff rules, and finally connecting to air rules." Instead, the controller executes the following parallel cooperative control process: Step 1: Parallel Activation of Multiple Rules. In response to the triggering of a takeoff event, the controller simultaneously activates three types of behavioral rules related to the current operating state within the same control cycle: ① Ground driving behavior rules (such as following rules and obstacle avoidance rules); ② Takeoff and landing transient behavior rules (such as takeoff and climb rules); ③ In-flight behavior rules (such as in-flight obstacle avoidance rules and airspace access rules).

[0054] During the parallel activation of multiple rules, the controller incorporates a priority arbiter and conflict resolution module. The priority arbiter dynamically sets the weight coefficients of each rule based on the current operational phase (e.g., initial takeoff, mid-climb, approach to the airspace). In the initial stage of takeoff (before liftoff), ground flight rules have the highest weight (0.6), followed by takeoff and landing transient rules (0.3), and air flight rules have the lowest weight (0.1). During the mid-climb phase (from takeoff to change altitude), the weight of the takeoff and landing transient rules is the highest (0.5), followed by the weight of the air flight rules (0.4) and the ground piloting rules (0.1). As the aircraft approaches the airspace, the weight of the air flight rules increases to the highest (0.6), followed by the takeoff and landing transient rules (0.3), and the ground flight rules decrease to (0.1).

[0055] The conflict resolution module employs a "safety-first, weighted fusion" strategy: when control commands output by different rules conflict in dimensions such as steering angle and thrust distribution (i.e., the difference exceeds a preset threshold, such as steering angle deviation > 5° or thrust deviation > 10%), the module uses the most stringent safety boundary as the benchmark and combines it with weighting coefficients to perform convex combination, generating a compromise control quantity. Specifically, the final control command... ,in To normalize the weights, For the first Class rule output, and The constraint set corresponding to all activation rules must be satisfied. The intersection condition. If the intersection is empty, the safety protection mode is entered according to the preset safety strategy (such as prioritizing braking or hovering).

[0056] Specifically, when multi-rule predicted actions conflict and weighted fusion fails to generate control instructions that satisfy all constraints, the conflict resolution module activates a tiered safety protection mechanism: Level 1 Protection (Local Conflict): If only a single control dimension (such as steering angle) conflicts, while the other dimensions are reconcilable, the conflicting dimension is locked to the limit value corresponding to the most stringent constraint, and the other dimensions are output using a weighted fusion method. For example, when aerial obstacle avoidance rules require a left turn... However, ground rules require right turns due to obstacle restrictions. When a conflict arises, the conflict resolution module determines a steering angle conflict, prioritizes the one with the strictest safety distance constraint (i.e., the aerial obstacle avoidance rule), and locks the steering angle to a left turn. At the same time, thrust and pitch angle are weighted and integrated for output.

[0057] Secondary protection (global conflict): If multiple control dimensions conflict simultaneously, or the weighted fusion result exceeds the vehicle dynamics safety boundary (e.g., total thrust exceeds the maximum allowable value), the module triggers a safety fallback strategy and executes preset emergency actions according to the current operating mode: In the ground domain or takeoff and landing transients, the maximum braking force is prioritized for deceleration and maintaining the heading; in the air domain, the hovering or slow landing procedure is prioritized until the conflict is resolved or the system receives a manual takeover command.

[0058] Conflict Recording and Rule Correction: After each conflict event is triggered, the controller records the pose information at the moment of conflict, the set of activated rules, and the final execution command, serving as input samples for offline optimization of the cross-domain behavior rule base. When the same type of conflict occurs consecutively... Within each control cycle ( When 5) occurs repeatedly, the controller dynamically adjusts the weight allocation of the corresponding rule or triggers iterative updates of the cross-domain behavior rule library to reduce the frequency of conflicts from the root.

[0059] Step Two: Parallel Prediction and Matching Judgment. The controller simultaneously inputs the vehicle's real-time pose information into three parallel prediction modules: The first prediction module is based on the "ground driving behavior rules" and predicts the future trajectory of the vehicle if it continues to execute the ground driving strategy. It is used to assess the safe clearance between the vehicle and obstacles in front and vehicles in the environment behind, and to ensure a safe exit from the ground phase. The second prediction module, based on the "takeoff and landing transient behavior rules", predicts the climb trajectory executed according to the optimal takeoff procedure and generates coordinated control commands for vertical climb and horizontal acceleration. The third prediction module, based on "airborne flight behavior rules," predicts the potential flight trajectory of a vehicle after it enters the air, used to identify vehicles in the airspace environment (such as...) in advance. Figure 3 Potential conflicts between Veh4 and Veh5 in the air phase are addressed to ensure secure access during the air phase.

[0060] Step 3: Collaborative Decision Making and Command Generation. The controller, through its built-in collaborative decision-maker (a decision fusion module used to integrate multi-rule prediction results and generate optimal control commands), comprehensively evaluates the matching degree between the predicted trajectories output by the three prediction modules and their respective rule bases. When the predicted actions of the three prediction modules and the actual state of the vehicle all meet the preset matching threshold, the collaborative decision-maker generates the optimal control command that integrates the three types of rule constraints, driving the actuator to complete the takeoff maneuver.

[0061] Step 4: Rule base iterative update. Only when the predicted actions of all activated rules successfully match the actual state of the vehicle will the controller use the current pose information to iteratively update the corresponding behavior rules in the cross-domain behavior rule base, enabling the cross-domain behavior rule base to be dynamically optimized according to the operating environment and vehicle state.

[0062] The above process ensures that the takeoff process considers both the safe exit from the ground phase and the safe access in the air phase, and realizes the parallel coordination of multiple rules rather than the serial switching, fundamentally avoiding response delays and motion oscillations caused by changes in control parameters.

[0063] Similarly, when a vehicle is cruising in the air and encounters a landing trigger, the controller determines whether the landing conditions are met. Landing triggers include active causes (such as reaching the destination, path planning changes, and user commands) and passive causes (such as system failure, insufficient energy, and airspace restrictions). Landing conditions include: the air altitude has decreased to a preset landing threshold, the airspace below has been confirmed to be safe, the ground landing point is ready, and the vehicle's dynamic state is stable.

[0064] by Figure 3 Taking the landing scenario as an example, when the controlled vehicle (Host Veh) receives a landing command while cruising in the air, the controller determines that the landing conditions are met and then formally triggers the landing event. Similar to the takeoff scenario, the controller also employs a multi-rule parallel and collaborative control strategy: In response to a landing event, the controller activates three types of rules simultaneously within the same control cycle: ① in-flight flight behavior rules (such as glide path tracking rules and in-flight obstacle avoidance rules); ② takeoff and landing transient behavior rules (such as landing descent rules and grounding buffer rules); ③ ground driving behavior rules (such as ground obstacle avoidance rules and parking rules).

[0065] The three parallel prediction modules perform the following tasks respectively: The first prediction module assesses the relative motion state of the vehicle and other aircraft in the airspace based on airborne rules to ensure the safety of the descent path; the second prediction module generates coordinated control commands for descent rate and pitch angle based on takeoff and landing rules to achieve a smooth descent; the third prediction module predicts the interaction with ground obstacles (OBS) and ground environment vehicles (such as ground-based vehicles) after landing based on ground rules. Figure 3 Potential conflicts between Veh6 and Veh7.

[0066] The collaborative decision-maker integrates the prediction results from three parallel prediction modules to generate control commands that ensure a safe exit from the air, a smooth descent, and proactive avoidance of ground risks. The cross-domain behavior rule base is only iteratively updated when the predicted actions of all activated rules match the actual state of the vehicle.

[0067] This collaborative mechanism ensures the continuity of control parameters and the stability of the motion trajectory during the landing transient process, avoiding control jumps and attitude oscillations caused by sequential rule switching.

[0068] 3. Behavior prediction and rule iterative updates

[0069] When performing domain switching, the controller obtains the vehicle's pose information according to a preset working cycle. The preset working cycle can be set according to the controller's computing power, sensor sampling frequency, and control real-time requirements (e.g., 20ms). Based on the vehicle's dynamics model, the controller predicts the motion action at the next moment and matches the predicted action with rules in the cross-domain behavior rule base. If a match is successful, the corresponding rule in the cross-domain behavior rule base is iteratively updated using the current pose information; if no match is found, monitoring continues until a match is found.

[0070] The matching process described above is based on the error between the predicted action and the expected state of the rule. The controller calculates the comprehensive error between the predicted trajectory and the reference trajectory in the cross-domain behavior rule base. If this error is less than a preset dynamic threshold, the match is considered successful. This dynamic threshold is specifically reflected in the error feasible region during the matching judgment stage.

[0071] The core of the iterative update process is a predictive control algorithm based on the error feasible region constraint. The error feasible region is a multi-dimensional state space jointly defined by vehicle dynamic parameters (such as mass, aerodynamic characteristics, and propulsion efficiency) and environmental safety constraints (such as minimum obstacle avoidance distance and airspace boundaries). Specifically, the error feasible region... Defined as the following set of state errors:

[0072] in: Let be the state error vector. To predict the state, Reference trajectory provided for cross-domain behavior rule base; The error boundary vector, determined by vehicle dynamics parameter constraints, includes: Position error: m (ground area) / m (airspace); Speed ​​error: m / s (ground region) m / s (airspace); Attitude angle error: , , ; Environmental safety constraints include: Minimum obstacle avoidance distance: ,in m (ground area) / m (airspace); Airspace boundary constraints: aircraft position Must meet ; To ensure dynamic consistency, the predicted state must satisfy the vehicle's kinematic and dynamic equations, such as: Wheel drive kinematic model: ; Rotorcraft flight dynamics model: .

[0073] Error Feasible Region The boundaries are dynamically adjusted according to the vehicle's operating status and the environment. During takeoff and landing transients, Take the intersection of the constraints of the ground domain and the air domain, that is This ensures that takeoff and landing transient control commands simultaneously meet the safety requirements of both domains. It clarifies the permissible safety boundaries for various state errors between predicted actions and actual pose information during cross-domain control.

[0074] Predictive control algorithms (such as Model Predictive Control, MPC) aim to minimize the error between the predicted action and the actual pose, but their optimization process is strictly limited to the boundaries of the feasible error region. The Model Predictive Control (MPC) algorithm used in this method is configured as follows: Prediction Time Domain Step, control time domain Step, discrete time step ms (aligned with the sensor sampling period). The objective function is:

[0075] in To predict the state, The reference trajectory provided for the cross-domain behavior rule base The state error weight matrix is ​​set according to the importance of the state variables, with position error weight of 10, velocity error weight of 5, and attitude error weight of 8. To control the incremental weight matrix (the weight of the rate of change of each actuator varies from 1 to 3). To control the rate of change of quantity; The constraints are determined by the joint error feasible region. definition:

[0076] in Let be the state error vector. It is determined by the vehicle's dynamic limits (such as maximum roll angle ±15°, maximum pitch angle ±20°) and environmental safety boundaries (such as minimum obstacle avoidance distance 2 m, vertical airspace separation 5 m); The solver employs a quadratic programming solver based on the interior-point method (such as OSQP), with a single-step solution time of ≤ 5 ms, meeting real-time requirements. When the error enters the 5% buffer zone within the feasible region boundary, the solver prioritizes the feasibility recovery strategy to ensure that iterative updates always occur within the safe boundary.

[0077] In other words, the error feasible region constitutes a rigid constraint boundary for the predictive control algorithm's optimization, ensuring that the parameters generated by each rule update do not exceed the vehicle's physical limits and the environment's safety threshold. By integrating the optimization objective with the safety boundary in this way, it fundamentally prevents the generation of risky strategies that exceed the vehicle's capabilities or environmental tolerances in pursuit of performance, thus ensuring the safety and robustness of the rule iteration process.

[0078] The process of cross-domain behavior prediction and rule iteration can be summarized as follows: In the perception and information fusion stage, a unified coordinate system is established to fuse multi-layered perception data from air and ground. Specifically, the perception and information fusion stage adopts a three-layer architecture: spatiotemporal alignment → weighted fusion → state estimation, as follows: Spatiotemporal alignment: Using the controller system time as a reference, heterogeneous data sources such as LiDAR (20 Hz), cameras (30 Hz), IMU (100 Hz), and V2X communication (10 Hz) are synchronized to a unified 50 Hz fusion cycle through timestamp interpolation. Spatially, a Northeast-Eastern Sky (ENU) coordinate system is established with the vehicle's center of mass as the origin. Ground obstacle coordinates, airborne vehicle coordinates, and GPS positioning information are uniformly transformed to this coordinate system, eliminating coordinate reference differences.

[0079] Weighted fusion: An adaptive weighted average is used for similar types of information (such as position and velocity) output from different sensors, with weights dynamically allocated based on the real-time confidence level of the sensors. The confidence level is jointly evaluated by the sensor health status, the degree of environmental interference (such as the impact of rain and fog on vision), and the historical error variance. For example, in ground conditions with good visibility, the camera position weight is set to 0.6, LiDAR 0.3, and V2X 0.1; in rainy, foggy, or aerial conditions, the weights of LiDAR and V2X are increased accordingly.

[0080] State estimation: The fused multi-source data is input into the Extended Kalman Filter (EKF) to establish a 12-dimensional state vector including vehicle position, velocity, acceleration, pitch angle, roll angle, current domain, and domain state. The EKF's prediction step is based on the vehicle dynamics model, while the update step uses the fused observations to correct the state, outputting smooth, high-frequency pose information for use by the behavior prediction module. The filter noise covariance matrix is ​​optimized offline based on measured data, with a convergence time ≤ 0.1 s and steady-state errors: position ≤ 0.05 m, velocity ≤ 0.1 m / s, and attitude angle ≤ 0.5°.

[0081] In the behavior prediction phase, short-term motion trajectories are predicted based on the current state and rules. In the rule matching and decision-making phase, it is determined whether the predicted action matches the existing rules. If they match, the action is executed and the update process begins. If they do not match, monitoring continues and prediction is repeated. In the rule iteration and update phase, within the boundary of the error feasible domain defined by vehicle dynamics parameters and environmental safety constraints, a model predictive control algorithm is used to analyze motion errors. If the errors are within the permissible range, the corresponding rules in the cross-domain behavior rule base are iteratively updated using the current information.

[0082] A control system for cross-domain integration of near-ground vehicles, such as Figure 1 As shown, a control method for cross-domain fusion of temporary ground vehicles applying any of the above includes: a controller that adopts an architecture of multi-rule parallel activation and collaborative decision-making, and utilizes the parallel processing of prediction and matching of multiple types of rules within the same control cycle to control the behavior switching and optimization of the vehicle in ground driving, air flight and take-off and landing transients; The vehicle sensor is connected to the first input terminal of the controller and is used to collect the vehicle's own pose information and surrounding obstacle information; The communication module is connected to the second input terminal of the controller and is used to obtain environmental vehicle information through vehicle-to-everything (V2X) or aviation communication. The execution layer, connected to the first output of the controller, is used to execute ground drive, rotor drive, steering, and braking actions according to the control commands generated by the controller. The storage module, connected to the second output of the controller, is used to store a pre-established cross-domain behavior rule library.

[0083] Furthermore, the controller incorporates a domain switching configuration unit, a behavior prediction unit, and a rule matching unit. The configuration unit, connected to the input of the behavior prediction unit, is used to respond to the triggering of domain switching events, simultaneously activate multiple types of rules related to the current running state in the cross-domain behavior rule base within the same control cycle, and dynamically set the priority weight coefficients of various rules according to the current running stage. The behavior prediction unit is connected to the input of the rule matching unit. It is used to predict multiple motion actions of the vehicle in the next moment based on the real-time pose information of the vehicle and the activated multiple rules in parallel, and output multiple sets of predicted trajectories to the rule matching unit. The rule matching unit is used to determine whether the multiple motion actions output by the behavior prediction unit match the corresponding multi-class rules. If they match, the behavior rules in the cross-domain behavior rule library are iteratively updated using the current pose information. If they do not match, the final control command is generated through priority arbitration and weighted fusion, and a hierarchical safety protection mechanism is activated when the conflict cannot be resolved.

[0084] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0085] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A control method for cross-domain fusion of near-ground vehicles, characterized in that, Includes the following steps: S1. Establish a cross-domain behavior rule base in the controller in advance; S2. When the vehicle is in the current domain and the triggering conditions of the domain switching event are met, the domain switching event is triggered. S3. In response to the triggering of a domain switching event, the controller simultaneously activates multiple types of rules in the cross-domain behavior rule base related to the current operating state within the same control cycle. S4. When performing domain switching behavior, the controller obtains the vehicle's pose information and predicts multiple motion actions of the vehicle at the next moment based on the pose information and various rules in the cross-domain behavior rule base. S5. The controller determines whether the predicted multiple motion actions match the corresponding multiple rules. If they do not match, the controller generates the final control command through priority arbitration and weighted fusion. If they match, the controller uses the current pose information to iteratively update the behavior rules in the cross-domain behavior rule library, thereby realizing collaborative decision-making and smooth transition of multiple rules.

2. The control method for cross-domain fusion of near-site vehicles according to claim 1, characterized in that, The S1 pre-establishes a cross-domain behavior rule library in the controller, including ground driving behavior rules, air flight behavior rules, and takeoff and landing transient behavior rules constrained by vehicle dynamics parameters.

3. The control method for cross-domain fusion of near-site vehicles according to claim 2, characterized in that, The cross-domain behavior rule base adopts a condition-action-constraint triplet structure for unified description and storage. The condition set includes the combination logic of the vehicle's own state and environmental events, the action set includes the control instructions output to the actuator module, and the constraint set is composed of vehicle dynamic parameters and environmental safety constraints. The cross-domain behavior rule base is organized into three layers according to the operating mode. The rules in each layer are stored in the form of a state-behavior mapping table, and the condition matching is achieved with high timeliness through hash index.

4. The control method for cross-domain fusion of near-site vehicles according to claim 2, characterized in that, The iterative update of the cross-domain behavior rule base is based on the error feasible domain boundary. The update trigger condition is: within multiple consecutive control cycles, the weighted comprehensive error between the predicted action of all activated rules and the actual state of the vehicle is lower than the dynamic threshold. The dynamic threshold is adjusted in real time based on the current speed and environmental complexity.

5. The control method for cross-domain fusion of near-site vehicles according to claim 1, characterized in that, S2 domain handover events include takeoff events (handover from ground domain to air domain) and landing events (handover from air domain to ground domain). Domain switching events are triggered by passive and active reasons. Passive reasons include vehicle being affected by obstacles, environmental changes, and system failures requiring an emergency domain switch. Active reasons include changes in vehicle destination, route planning requirements, and user commands. When a domain switching event is triggered, the controller updates the vehicle's domain status information and adjusts the control parameters to adapt to the new domain. The control parameters include power distribution parameters, control loop gain, and actuator mode.

6. The control method for cross-domain fusion of near-site vehicles according to claim 1, characterized in that, In S4, when performing domain switching behavior, the vehicle pose information obtained by the controller includes the vehicle's own running status information, obstacle information, and environmental vehicle information. The vehicle's operating status information includes speed, acceleration, position, heading, vehicle number, pitch angle, current domain, and domain status; Obstacle information includes the relative direction and relative distance of the obstacle; Environmental vehicle information is obtained through communication with other vehicles, including the speed, acceleration, position, heading, vehicle number, current domain, and domain status of other vehicles within the communication range.

7. The control method for cross-domain fusion of near-site vehicles according to claim 1, characterized in that, In S5, the controller determines whether the predicted multiple motion actions match the corresponding multi-class rules. If they do not match, the controller generates the final control command through priority arbitration and weighted fusion. The specific content of the final control command is as follows: When the degree of mismatch exceeds the preset safety boundary, the controller enters the graded safety protection mode and executes preset emergency actions according to the current operating mode; Priority arbitration dynamically sets the weight coefficients of multiple rules in each cross-domain behavior rule base according to the current operation stage. Weighted fusion is based on the safety boundary and combines the weight coefficients to generate a compromise control quantity through convex combination. When multi-rule prediction actions conflict and weighted fusion fails to generate control instructions that satisfy all constraints, a hierarchical safety protection mechanism is activated. The tiered security protection mechanism includes: locking the conflict dimension to the limit value corresponding to the most stringent constraint in the event of a local conflict, and triggering a security fallback strategy to execute preset emergency actions in the event of a global conflict.

8. The control method for cross-domain fusion of near-site vehicles according to claim 1, characterized in that, The specific details of achieving collaborative decision-making and smooth transition among multiple rule types in S5 are as follows: Construct a joint error feasible domain that integrates ground environment, air environment, and takeoff and landing transient safety constraints; The model predictive control algorithm is adopted. Under the constraint of the joint error feasible region, the objective function is to minimize the comprehensive error between multiple motion actions predicted by multiple rules in the cross-domain behavior rule library and the actual pose of the vehicle. The parameters of relevant behavior rules in the cross-domain behavior rule library are optimized and updated.

9. A control system for cross-domain integration of near-ground vehicles, characterized in that, The control method for cross-domain fusion of near-site vehicles according to any one of claims 1-8 includes: The controller adopts an architecture of multi-rule parallel activation and collaborative decision-making. It utilizes the parallel processing of prediction and matching of multiple types of rules within the same control cycle to control and optimize the behavior switching of the vehicle during ground driving, air flight and take-off and landing transients. The vehicle sensor is connected to the first input terminal of the controller and is used to collect the vehicle's own pose information and surrounding obstacle information; The communication module is connected to the second input terminal of the controller and is used to obtain environmental vehicle information through vehicle-to-everything (V2X) or aviation communication. The execution layer, connected to the first output of the controller, is used to execute ground drive, rotor drive, steering, and braking actions according to the control commands generated by the controller. The storage module, connected to the second output of the controller, is used to store a pre-established cross-domain behavior rule library.

10. A control system for cross-domain fusion of near-ground vehicles according to claim 9, characterized in that, The controller has a built-in domain switching configuration unit, behavior prediction unit, and rule matching unit. The configuration unit, connected to the input of the behavior prediction unit, is used to respond to the triggering of domain switching events, simultaneously activate multiple types of rules related to the current running state in the cross-domain behavior rule base within the same control cycle, and dynamically set the priority weight coefficients of various rules according to the current running stage. The behavior prediction unit is connected to the input of the rule matching unit. It is used to predict multiple motion actions of the vehicle in the next moment based on the real-time pose information of the vehicle and the activated multiple rules in parallel, and output multiple sets of predicted trajectories to the rule matching unit. The rule matching unit is used to determine whether the multiple motion actions output by the behavior prediction unit match the corresponding multi-class rules. If they match, the behavior rules in the cross-domain behavior rule library are iteratively updated using the current pose information. If they do not match, the final control command is generated through priority arbitration and weighted fusion, and a hierarchical safety protection mechanism is activated when the conflict cannot be resolved.

Images