Flight control method and system of unmanned aerial vehicle, electronic device and storage medium

Abstract

This application discloses a flight control method, system, electronic device, and storage medium for a drone. The method includes: at the beginning of each control cycle, calculating the comprehensive cost of each of a plurality of preset candidate velocity vectors based on the drone's own state information and the perception data of at least one obstacle within a preset range of the drone; calculating the Ising model input parameters based on the Ising model input parameter calculation rules and the comprehensive cost of each candidate velocity vector, obtaining a set of Ising model input parameters; performing annealing on the Ising model input parameter set to output the ground state spin configuration; and determining the flight control parameters of the drone for the current control cycle based on the ground state spin configuration. This solution can improve the response speed of the drone.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to flight control methods, systems, electronic devices and storage media for unmanned aerial vehicles (UAVs). Background Technology

[0002] With the widespread application of unmanned aerial vehicles (UAVs) and UAV swarms in complex environments (such as forests, power grids, and low-altitude urban areas), extremely high demands are placed on the real-time collision avoidance capabilities of UAVs. Traditional collision avoidance and path planning algorithms typically rely on discrete spatial sampling and iterative calculation of cost functions. When facing high-density, highly dynamic obstacle environments, the computational complexity of collision avoidance and path planning increases exponentially or polynomially with the number of obstacles, resulting in excessive computational latency and failing to meet the reaction requirements of UAVs during high-speed flight. Summary of the Invention

[0003] This application provides a flight control method, system, electronic device, and storage medium for unmanned aerial vehicles (UAVs), aiming to solve the problem that when facing high-density, high-dynamic obstacle environments, the computational complexity of collision avoidance and path planning increases exponentially or polynomially with the number of obstacles, resulting in excessive computational latency and failing to meet the reaction requirements of UAVs during high-speed flight.

[0004] In a first aspect, embodiments of this application provide a flight control method for an unmanned aerial vehicle (UAV), the method being applied to the UAV, the method comprising: At the beginning of each control cycle, based on the UAV's own state information and the perception data of at least one obstacle within the UAV's preset range, the comprehensive cost of each of the multiple preset candidate velocity vectors is calculated. Based on the preset Ising model input parameter calculation rules and the comprehensive cost of each candidate velocity vector, the Ising model input parameters of each candidate velocity vector are calculated to obtain the Ising model input parameter set; Annealing is performed on the input parameter set of the Ising model to output the ground state spin configuration; The flight control parameters for the current control cycle of the UAV are determined based on the ground state spin configuration.

[0005] Secondly, embodiments of this application also provide a flight control system for an unmanned aerial vehicle (UAV), wherein the flight control system is configured in the UAV and includes: The cost calculation layer is used to calculate the comprehensive cost of each candidate velocity vector among multiple preset candidate velocity vectors at the beginning of each control cycle, based on the UAV's own state information and the perception data of at least one obstacle within the UAV's preset range. The Ising mapping layer is used to calculate the Ising model input parameters of each candidate velocity vector based on the preset Ising model input parameter calculation rules and the comprehensive cost of each candidate velocity vector, thereby obtaining the Ising model input parameter set; An annealing coprocessor layer is used to perform annealing solution processing on the input parameter set of the Ising model and output the ground state spin configuration. The decoding and flight control interface layer is used to determine the flight control parameters of the UAV in the current control cycle based on the ground state spin configuration.

[0006] Thirdly, embodiments of this application also provide an electronic device, which includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the above-described method.

[0007] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program, the computer program including program instructions that, when executed by a processor, can implement the above-described method.

[0008] This application provides a flight control method, system, electronic device, and storage medium for a drone. On one hand, when calculating the comprehensive cost of each candidate velocity vector and the corresponding Ising model input parameters, the computational load is linearly related only to the total number of candidate velocity vectors (a preset constant) and the number of obstacles. The overall computational delay no longer increases dramatically with the number of obstacles, enabling the drone to quickly complete collision avoidance decisions within each control cycle in high-density, high-dynamic obstacle environments, meeting the reaction requirements of high-speed flight. On the other hand, in each control cycle, only the drone's own state information and perception data at the current moment need to be collected, and the position of obstacles in the next cycle needs to be predicted to calculate the flight control parameters for the current cycle. The entire decision-making process does not require pre-constructing a global environment map or planning a complete flight path, further improving the obstacle avoidance speed of the drone. Attached Figure Description

[0009] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 A flowchart illustrating the flight control method for an unmanned aerial vehicle (UAV) provided in an embodiment of this application; Figure 2 A schematic diagram of a sub-process of the flight control method for an unmanned aerial vehicle provided in an embodiment of this application; Figure 3Another schematic diagram of a sub-process of the flight control method for an unmanned aerial vehicle provided in an embodiment of this application; Figure 4 Another schematic diagram of a sub-process of the flight control method for an unmanned aerial vehicle provided in an embodiment of this application; Figure 5 A schematic block diagram of a flight control system for an unmanned aerial vehicle (UAV) provided in an embodiment of this application; Figure 6 A schematic block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0011] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0012] It should be noted that any AI models, software tools, or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been obtained by authorized entities (who have known and consented) or fully authorized by all parties through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals.

[0013] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0014] It should also be understood that the terminology used in this application specification is for the purpose of describing particular embodiments only and is not intended to limit the application. As used in this application specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0015] It should also be further understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0016] This application provides flight control methods, systems, electronic devices, and storage media for unmanned aerial vehicles (UAVs).

[0017] The execution entity of the flight control method of the UAV can be the flight control system of the UAV provided in the embodiments of this application (hereinafter referred to as the flight control system), or an electronic device that integrates the flight control system. The flight control system can be implemented in hardware or software, and the electronic device can be the controller in the UAV.

[0018] Figure 1 This is a flowchart illustrating the flight control method for an unmanned aerial vehicle (UAV) provided in an embodiment of this application. Figure 1 As shown, the method includes the following steps S110-S140.

[0019] S110. At the beginning of each control cycle, based on the UAV's own state information and the perception data of at least one obstacle within the UAV's preset range, calculate the comprehensive cost of each of the multiple preset candidate velocity vectors.

[0020] The drone in this embodiment of the application is equipped with a control cycle for flight control, and at the beginning of each control cycle, the comprehensive cost of each candidate velocity vector is calculated by collecting the drone's own state information and the perception data of at least one obstacle within the preset range of the drone.

[0021] In some embodiments, please refer to Figure 2 Step S110 includes S1101. Collect the drone's own status information and the perception data of at least one obstacle within the drone's preset range.

[0022] In some embodiments, the obstacle perception data includes radar perception data and depth image data. Step S1101 includes: acquiring the radar perception data through the radar module mounted on the UAV; synchronously acquiring the depth image data through the binocular vision module mounted on the UAV; and obtaining the UAV's own state information through the navigation module mounted on the UAV.

[0023] In one example, at the start time t of each control cycle, the flight control system triggers synchronous data acquisition from various modules in the multimodal perception layer. The multimodal perception layer includes a radar module, a binocular vision module, and a navigation module. The radar module is either a lidar or millimeter-wave radar. The navigation module includes an inertial measurement unit (IMU) module and a global navigation satellite system (GNSS) module. Specifically, the lidar or millimeter-wave radar scans the local space within a radius R (typically 5-20m) around the UAV at a frequency of at least 50Hz. The radar perception data acquired by the millimeter-wave radar is 3D point cloud data and Doppler velocities at each point, while the radar perception data acquired by the lidar is 3D point cloud data and velocity estimates based on frame differences at each point. The binocular vision module synchronously acquires depth images to supplement fine contour information of nearby obstacles. The IMU and GNSS modules provide the UAV's current position p. uav The current velocity vector of the drone, v uav and attitude quaternion q uav .

[0024] Specifically, in one example, the control cycle can be 20ms. In this case, the flight control system collects its own status information and obstacle perception data every 20ms to control the UAV flight. It should be noted that the specific duration of the control cycle is not limited in this embodiment of the application, and the specific duration can be set according to the actual accuracy requirements.

[0025] S1102. Based on the perception data, predict the position of each obstacle at the start of the next control cycle to obtain a set of predicted obstacle positions.

[0026] In some embodiments, specifically, the set of predicted obstacle locations is obtained through the following steps: By clustering the perceived data, at least one obstacle is identified; for each obstacle, the position of the obstacle is extrapolated to the start time of the next control cycle based on the obstacle state vector maintained by the extended Kalman filter and a uniform or uniformly accelerated motion model, thereby obtaining the set of predicted obstacle positions.

[0027] Specifically, in one example, after clustering, the perceived data is used to extract M independent obstacle targets. For each obstacle, an Extended Kalman Filter (EKF) is used to maintain its state vector (position, velocity, acceleration), and at the current time, the position of each obstacle is extrapolated to a future time t+Δt (i.e., the start time of the next control cycle, or the end time of the current control cycle) based on a uniform or uniformly accelerated motion model, resulting in the set of predicted obstacle positions {p}. obs The prediction time step Δt and the control period T are related to the time step and the control period T. ctrl The event lengths are the same (e.g., 20ms), ensuring that the prediction is the collision risk after the drone executes the current control command (flight control parameters).

[0028] S1103. Calculate the comprehensive cost of each candidate velocity vector among multiple preset candidate velocity vectors based on the predicted obstacle position set and the self-state information.

[0029] In this embodiment, multiple candidate velocity vectors are preset, which are discretized finite sets. The discretization scheme uses uniform sampling in spherical coordinates. V = { v i = v max ·(sinθ i cosφ i sinθ i sinφ i cosθ i ) | i = 1, ..., N};(1) Where V is a set of multiple preset candidate velocity vectors, v i Let θ be the i-th candidate velocity vector, and N be the total number of candidate velocity vectors (typically N = 64~512). i ∈ [0, π] is the polar angle, φ i ∈ [0, 2π) is the azimuth angle, v max This represents the maximum speed of the drone. To ensure uniform coverage, the Fibonacci spherical sampling algorithm can be used to generate {(θ...} i , φ i )}.

[0030] The corresponding binary decision variable is defined as follows: x i ∈ {0, 1}, i = 1, ..., N; (2) When x i = 1 indicates that a candidate velocity vector v is selected. i When x i = 0 indicates that no selection is made.

[0031] In some embodiments, specifically, please refer to Figure 3 Step S1103 includes: S11031. Based on its own state information, predict the UAV's predicted position at the start time of the next control cycle under each of the candidate velocity vectors, and determine the collision risk cost corresponding to each of the candidate velocity vectors according to the UAV's predicted position under each of the candidate velocity vectors and the set of predicted obstacle positions.

[0032] For the candidate velocity vector v i The predicted position of the drone after Δt is: p i =p uav +v i ·Δt; (3) Where, p i p represents the predicted position of the UAV at the start of the next control cycle when the UAV executes the i-th candidate velocity vector. uav v represents the current location of the drone. i Let be the i-th candidate velocity vector, and Δt be the time length of one control cycle.

[0033] After obtaining the predicted UAV position under each candidate velocity vector at the start time of the next control cycle, the collision risk cost corresponding to each candidate velocity vector is determined based on the collision risk cost calculation formula, wherein the collision risk cost calculation formula is as follows: (4-1) in, Let M be the collision risk cost corresponding to the i-th candidate velocity vector, and M be the number of obstacles. The predicted position (vector) of the UAV at the start of the next control cycle when the UAV executes the i-th candidate velocity vector. For the k-th obstacle in the future time... The predicted location (vector) of obstacles. > 0 represents the repulsion weight coefficient, which controls the strength of the repulsive force of the obstacle; > 0 is a very small positive number (usually taken as 1 × 10). -6 ), to prevent the denominator from being zero (i.e., the singularity where the predicted position of the UAV and the predicted position of the obstacle are completely coincident); ||·|| represents the Euclidean distance.

[0034] In some embodiments, to further distinguish between high-risk obstacles at close range and low-risk obstacles at long range, a cutoff radius R may be introduced. safe Only calculate distances less than R safe The contribution of obstacles: At this point, the formula for calculating the collision risk cost is: (4-2) in, For a high-risk obstacle at close range among multiple obstacles, the distance between the drone and the high-risk obstacle at close range meets the following requirements: .

[0035] S11032. Calculate the cosine penalty of the angle between the velocity direction and the target direction of each candidate velocity vector to obtain the target deviation cost of each candidate velocity vector.

[0036] This application embodiment also requires calculating the target deviation cost C for each candidate velocity vector. goal(i) The target deviation cost is defined as the cosine penalty of the angle between the direction of the candidate velocity vector and the target direction, ensuring that the UAV maintains flight in the target direction as much as possible while avoiding collisions. The target direction is the unit direction vector from the current position of the UAV to the mission target point (the final destination coordinates to be reached in this flight mission), and the current position of the UAV is obtained based on its own state information.

[0037] In some embodiments, the target deviation cost is obtained based on the following formula: (5) in, Let d be the target deviation cost for the i-th candidate velocity vector, β > 0 be the attraction weight coefficient, and d goal Let be the unit vector in the target direction. The range of v is [0, 2β], i Let ||·|| represent the Euclidean distance for the i-th candidate velocity vector. i With d goal When completely in the same direction Take the minimum value of 0, and take the value when completely reversed. The maximum value is 2β.

[0038] S11033. Based on each candidate velocity vector and the current velocity vector of the UAV, determine the current acceleration of the UAV within one control cycle for each candidate velocity vector, and determine the dynamic constraint cost corresponding to each candidate velocity vector based on the current acceleration of each UAV and the preset maximum acceleration of the UAV.

[0039] The current velocity vector of the drone is obtained based on its own state information.

[0040] This application embodiment also requires calculating the dynamic constraint cost of each candidate velocity vector, which is used to penalize violent maneuvers that exceed the maximum acceleration of the UAV.

[0041] In some instances, the cost of dynamic constraints is derived based on the following formula: (6) in, The dynamic constraint cost for the i-th candidate velocity vector. For dynamic penalty weights, For the current acceleration of the drone, Maximum acceleration of the drone.

[0042] S11034. Determine the comprehensive cost of each candidate velocity vector based on the collision risk cost, the target deviation cost, and the dynamic constraint cost.

[0043] After obtaining the collision risk cost, target deviation cost, and dynamic constraint cost through the above steps, the sum of these three costs is used to determine the comprehensive cost.

[0044] Specifically, the comprehensive cost of each candidate velocity vector is determined based on the following formula: (7) in, The comprehensive cost for the i-th candidate velocity vector is... Let i be the collision risk cost for the i-th candidate velocity vector. The target deviation cost for the i-th candidate velocity vector. The dynamic constraint cost is the i-th candidate velocity vector.

[0045] S120. Based on the preset Ising model input parameter calculation rules and the comprehensive cost of each candidate velocity vector, calculate the Ising model input parameters of each candidate velocity vector to obtain the Ising model input parameter set.

[0046] In this embodiment, the preset calculation rules for the Ising model input parameters are a set of mathematical operation specifications used to convert the comprehensive cost of each candidate velocity vector into Ising model input parameters that can be recognized by the annealing coprocessor.

[0047] In some embodiments, specifically, please refer to Figure 4 Step S120 includes: S1201. Determine the maximum comprehensive cost among multiple candidate velocity vectors, and determine the current penalty coefficient based on the maximum comprehensive cost.

[0048] S1202. Based on the preset Ising model input parameter calculation rules, the current penalty coefficient, and the comprehensive cost of each candidate velocity vector, calculate the Ising model input parameters of each candidate velocity vector to obtain the Ising model input parameter set.

[0049] Specifically, in one example, the maximum combined cost c among the combined costs of multiple candidate velocity vectors is first determined. max Based on C max Determine the current penalty coefficient γ. The current penalty coefficient γ should be much larger than all the combined costs, i.e., γ >> max(c i This ensures that the uniqueness constraint is strictly satisfied (there is only one x). i (Equal to 1), typically taking the value γ = 10·c max .

[0050] Then, based on the preset Ising model input parameter calculation rules, the current penalty coefficient γ, and the comprehensive cost c of each candidate velocity vector, i Calculate the Ising model input parameters for each candidate velocity vector.

[0051] The input parameters of the Ising model include the local magnetic field input parameter h. i and the coupling strength input parameter J ij The energy function of the Ising model is defined as: (8) Where H(s) is the total energy (scalar) of the system, and the goal of annealing is to find a set of spin configurations s that minimizes this energy value (i.e., the ground state); s i s represents the binary state of the i-th candidate velocity vector, taking only +1 or -1; in this embodiment, s i =+1 means "select the i-th candidate velocity vector", s i =-1 means "do not select the i-th candidate velocity vector".

[0052] h i The local magnetic field input parameter is an independent external magnetic field acting on the i-th spin. In this embodiment, it encodes the comprehensive cost of the candidate velocity vector itself (collision risk, target deviation, and dynamic constraints). i The larger the number (positive number), the more the system energy increases when the spin is +1, meaning the greater the "resistance" for that speed.

[0053] J ij The coupling strength input parameter describes the pairwise interaction between the i-th spin and the j-th spin. In this embodiment, it encodes a "only one can be selected" mutual exclusion hard constraint. When J... ij When s > 0 (positive number), if i and s j If both are +1, then J ij s i s jThis will generate tremendous positive energy, thereby forcing the system to choose only one of the two options.

[0054] The core of this application lies in transforming the "uniqueness constraint" in the collision avoidance problem (i.e., the UAV can only choose one velocity vector at the same time) into spin-to-spin coupling in the Ising model, and the coupling strength is the same for all spin pairs.

[0055] Specifically, the Ising model input parameters include local magnetic field input parameters and coupling strength input parameters. The calculation rules for the Ising model input parameters include local magnetic field calculation formulas and coupling strength calculation formulas. Based on the preset Ising model input parameter calculation rules, the current penalty coefficient, and the comprehensive cost of each candidate velocity vector, the Ising model input parameters of each candidate velocity vector are calculated to obtain the Ising model input parameter set, including: Based on the local magnetic field calculation formula, the current penalty coefficient, and the comprehensive cost of each candidate velocity vector, the local magnetic field input parameters for each candidate velocity vector are determined. The local magnetic field calculation formula is as follows: (9) in, The local magnetic field input parameters are the i-th candidate velocity vector. Let N be the total cost of the i-th candidate velocity vector, and N be the total number of the candidate velocity vectors. The current penalty coefficient; The coupling strength input parameter is determined based on the coupling strength calculation formula and the current penalty coefficient. The coupling strength calculation formula is as follows: (10) in, The coupling strength is input as a parameter, where i and j are the indices of the candidate velocity vectors, and i < j; The set of input parameters for the Ising model is determined based on the local magnetic field input parameters of each candidate velocity vector and the coupling strength input parameters.

[0056] Among them, in the embodiments of this application The input parameter for the coupling strength is globally uniform and only needs to be calculated once in each control cycle.

[0057] This application embodiment specifies the input parameters of the Ising model as local magnetic field input parameters and global coupling strength input parameters, and defines formulas (9) and (10). This scheme greatly simplifies the parameter injection process and data transmission volume of the annealing coprocessor, and provides a fundamental guarantee for achieving microsecond-level annealing solutions.

[0058] S130. Perform annealing on the input parameter set of the Ising model and output the ground state spin configuration.

[0059] This application embodiment performs annealing on the input parameter set of the Ising model to solve for the output ground state spin configuration s. opt .

[0060] The ground state is the lowest energy state in a quantum system. The goal of quantum annealing is to find the ground state of the problem's Hamiltonian, which corresponds to the optimal solution to the optimization problem.

[0061] Specifically, the UAV's main control CPU uses a direct memory access (DMA) channel to transfer the Ising model input parameter set {h} i The J value is transmitted to the input register of the onboard annealing coprocessor (annealing coprocessor layer). After receiving the parameter, the coprocessor immediately starts the annealing process: the initial temperature is set to a high value (putting the system in a high-energy random state), and then the temperature is gradually reduced according to the preset annealing plan. The system crosses the energy barrier through quantum tunneling (real quantum annealing) or thermal fluctuations (FPGA simulated annealing) and finally converges to the ground state.

[0062] The annealing coprocessor can be a quantum annealing chip (such as a miniaturized chip based on the D-Wave LAP architecture) or a simulated annealing IP core implemented on an FPGA. The execution time of the annealing solution is 1μs to 10μs, which is much less than the 20ms control cycle. The annealing coprocessor provided in this application can use either a real quantum annealing chip (for optimal performance) or an FPGA-based simulated annealing chip (for high engineering feasibility). Both interact with the flight control system through a unified data flow interface, facilitating flexible replacement according to mission requirements.

[0063] After annealing is complete, the coprocessor notifies the main CPU via an interrupt signal to read the ground-state spin configuration s from the result register. opt = [s1, s2, ..., s N ], where each spin s i The value can be +1 or -1.

[0064] S140. Determine the flight control parameters of the UAV in the current control cycle based on the ground state spin configuration.

[0065] In this embodiment, it is first determined whether there is only one spin with a positive 1 in the ground state spin configuration; if there is only one spin with a positive 1, the candidate velocity vector with a positive spin is determined as the target velocity vector, and the flight control parameters of the UAV in the current control cycle are determined according to the target velocity vector; if there is not only one spin with a positive 1, the flight control parameters of the UAV in the current control cycle are determined according to a preset safety default strategy, the safety default strategy including a hovering strategy or a deceleration strategy.

[0066] Specifically, determine whether there is only one spin of +1 (corresponding to x) in the ground state spin configuration. i =1). This is because the aforementioned mapping process introduced strong mutual exclusion constraints (global coupling strength). Ideally, the annealing result will have exactly one spin of +1 and the rest of -1.

[0067] If there is only one candidate velocity vector with a spin of +1, then the candidate velocity vector with a spin of +1 is determined as the target velocity vector v. opt That is: v opt = v k , where k satisfies s k = +1, then based on the target velocity vector v opt Determine the flight control parameters for the current control cycle of the UAV. Specifically, v opt The desired attitude angles (pitch angle, roll angle) and desired thrust are converted and sent to the UAV's flight controller (such as PX4 or ArduPilot) via the MAVLink protocol or direct memory mapping, and executed by the flight controller's inner-loop PID (a feedback control mechanism) controller.

[0068] If there is more than one spin of +1 (e.g., 0 or more spins of +1), it indicates that the annealing solution may be due to insufficient annealing time or incomplete convergence of noise interference. In this case, the flight control parameters are determined according to a preset safety default strategy. The safety default strategy includes a hovering strategy (maintaining the current position) or a deceleration strategy (reducing the flight speed) to ensure flight safety.

[0069] After all the above steps have ended, wait for a timer interrupt to trigger the next control cycle, return to step S110, and form a real-time control closed loop.

[0070] That is, after receiving the desired attitude angles (pitch and roll angles) and desired thrust and speed commands, the flight controller immediately begins to execute attitude control and motor speed regulation. The main control CPU enters a waiting state until a 20ms timer interrupt arrives, triggering the start of the next control cycle and returning to step S110. This closed-loop design based on timer interrupts ensures the accuracy of the control frequency and is compatible with the control frequency of traditional UAV flight controllers.

[0071] In this embodiment of the application, formulas (9) and (10) are derived based on the following steps: First, the QUBO model is constructed: Quadratic Unconstrained Binary Optimization (QUBO) is a mathematical expression of a combinatorial optimization problem where variables take values ​​of 0 or 1, and the objective function is a quadratic polynomial of the variables. QUBO and the Ising model can be equivalently transformed through simple variable substitution, and it is the standard input format for quantum annealing hardware.

[0072] The collision avoidance problem is transformed into a QUBO (Quadratic Unconstrained Binary Optimization) problem, with the objective of minimizing the following quadratic function (the standard mathematical representation of a quadratic unconstrained binary optimization (QUBO) problem): (11) in, Let E(x) be the objective function (the goal of annealing is to find a set of x that minimizes E(x)); x = [x1, x2, ..., xN], x i ∈{0,1}; It is an N×N upper triangular matrix with coefficients. These are the diagonal elements (coefficients of linear terms); The elements are off-diagonal (quadratic coefficients, paired interactions), and N is the total number of candidate velocity vectors.

[0073] Since a drone can only choose one velocity vector at a time, a hard constraint of "uniqueness" must be introduced: (12) The constraint formula (12) is transformed into a penalty term and added to the objective function: (13) Where P(x) is the penalty term for the uniqueness constraint of velocity selection, which is incorporated into the total QUBO energy; x is the penalty coefficient; i x ∈{0,1} j ∈{0,1}, where N is the total number of candidate velocity vectors.

[0074] Because x i For x ∈ {0,1}, there exists x i 2 = x i Ignoring the constant term, the complete QUBO objective function is: (14) in, Given a complete QUBO total energy function with uniqueness constraints, the annealing minimization objective is c. i The comprehensive cost of the i-th candidate velocity vector, x i x ∈{0,1} j ∈{0,1}.

[0075] Comparing formula (14) with formula (11), the elements of the corresponding QUBO matrix Q are defined as follows: (Diagonal); (15-1) (Non-diagonal, i) j); (15-2) Here, γ is the penalty coefficient, which must satisfy γ >> max(c_i) to ensure that the uniqueness constraint is strictly satisfied. A typical value is γ = 10·max(c_i). i ).

[0076] Then perform the Ising model transformation: The Ising Model is a mathematical model in statistical physics that describes the ferromagnetism of matter. It consists of discrete spin variables (+1 or -1) and their interactions.

[0077] Quantum annealing hardware (such as D-Wave systems) typically solves directly for the Hamiltonian of the Ising model. The energy function of the Ising model is defined as: (8) By substituting x i =(s i +1) / 2 (i.e., s) i = 2x i -1), which can map QUBO parameters to Ising parameters: (16-1) (16-2) Based on formulas (15-2) and (16-1), formula (10) can be derived. Based on formulas (15-1), (15-2) and (16-2), formula (9) can be derived.

[0078] The annealing scheduling process in this embodiment is as follows: During annealing, the transverse magnetic field strength Γ(t) (quantum annealing) or temperature T(t) (simulated annealing) decreases linearly from its initial value to zero according to the following scheduling function: Γ(t) = Γ0·(1 - t / t anneal ), t ∈ [0, t anneal ] ; (17) Where Γ0 is the initial transverse magnetic field strength, t anneal This refers to the total annealing time.

[0079] The system Hamiltonian is: H total(t) = Γ(t)·H driver + (1 - Γ(t) / Γ0)·H problem (18) in, To drive the Hamiltonian (transverse magnetic field term), H problem The problem Hamiltonian is defined by equation (8). According to the quantum adiabatic theorem, when the annealing time t anneal Given enough time, the system will converge to the ground state with a high probability.

[0080] The following is a computational complexity analysis of the embodiments of this application: The computational complexity of traditional collision avoidance algorithms (such as A* and RRT) increases polynomially or exponentially with the number of obstacles M and the dimension of the search space D, typically O(M·D·log(M·D)). The computational complexity of the QUBO mapping step (step S140) in this application is O(N·M), where N is the number of candidate velocity vectors (a fixed constant), and M is the number of obstacles. The computation time of the annealing solution step (step S130) is determined by the hardware and is independent of M (because the information of M is encoded in the (h, J) matrix, with a fixed size of N×N). Therefore, the overall computational delay is linearly related to the number of obstacles M (O(M)), rather than the superlinear relationship of traditional algorithms, achieving a "sublinear" advantage.

[0081] The drone flight control method provided in this application can be applied to drone swarms traversing high-density obstacle environments. In such environments, when drone swarms pass through forests, power grids, urban building complexes, etc., each drone faces the simultaneous threat of dozens of static and dynamic obstacles. The solution provided in this application can independently calculate the optimal avoidance speed in each control cycle, eliminating the need for internal swarm communication and achieving truly decentralized distributed collision avoidance.

[0082] In this application scenario, each drone runs the annealing solution independently without the need for global communication coordination. The swarm size can be expanded indefinitely without affecting the collision avoidance response time of a single drone, making it suitable for large-scale swarm operations with hundreds of drones.

[0083] The UAV flight control method provided in this application can also be applied to instantaneous emergency avoidance scenarios in mixed airspace. In this case, in low-altitude airspace coexisting with manned aircraft, the UAV needs to perform instantaneous emergency avoidance of a suddenly appearing manned aircraft. The speed of manned aircraft typically reaches 200-300 km / h, leaving the UAV with extremely short reaction time (0.1-1 second). The microsecond-level decision-making capability of this application enables the UAV to output the optimal avoidance command immediately after sensing a manned aircraft, meeting the ICAO (International Civil Aviation Organization) requirements for the response time of UAV perception and avoidance (DAA) systems.

[0084] The UAV flight control method provided in this application can also be applied to extreme defense scenarios. For example, in military defense scenarios, UAVs need to intercept and evade multiple high-speed projectiles (such as anti-UAV nets and kinetic interceptors). These projectiles can reach speeds of 100-500 m / s, which traditional path planning algorithms cannot handle. The solution provided in this application can calculate the optimal avoidance trajectory within microseconds, achieving multi-target avoidance.

[0085] In summary, on the one hand, when calculating the comprehensive cost of each candidate velocity vector and the corresponding Ising model input parameters, the computational load of this scheme is only linearly related to the total number of candidate velocity vectors (a preset constant) and the number of obstacles. The overall computational delay no longer increases sharply with the number of obstacles, enabling the UAV to quickly complete collision avoidance decisions within each control cycle in high-density, high-dynamic obstacle environments, meeting the reaction requirements of high-speed flight. On the other hand, in each control cycle, only the current state information and perception data need to be collected, and the position of obstacles in the next cycle needs to be predicted to calculate the flight control parameters for the current cycle. The entire decision-making process does not require pre-constructing a global environment map or planning a complete flight path, further improving the obstacle avoidance speed of the UAV.

[0086] Figure 5 This is a schematic block diagram of a flight control system 500 for an unmanned aerial vehicle (UAV) provided in an embodiment of this application. Figure 5 As shown, corresponding to the above-described flight control method for a UAV, this application also provides a flight control system 500 for a UAV. The flight control system 500 includes a unit for executing the aforementioned flight control method for the UAV, and the flight control system 500 can be configured within the UAV. Specifically, please refer to... Figure 5 The flight control system 500 of the UAV includes a cost calculation layer 501, an Ising mapping layer 502, an annealing coprocessor layer 503, and a decoding and flight control interface layer 504, wherein: The cost calculation layer 501 is used to calculate the comprehensive cost of each candidate velocity vector among multiple preset candidate velocity vectors at the beginning of each control cycle, based on the UAV's own state information and the perception data of at least one obstacle within the preset range of the UAV. Ising mapping layer 502 is used to calculate the Ising model input parameters of each candidate velocity vector based on the preset Ising model input parameter calculation rules and the comprehensive cost of each candidate velocity vector, and obtain the Ising model input parameter set; Annealing coprocessor layer 503 is used to perform annealing solution processing on the input parameter set of the Ising model and output the ground state spin configuration. The decoding and flight control interface layer 504 is used to determine the flight control parameters of the UAV in the current control cycle based on the ground state spin configuration.

[0087] In some embodiments, when the Ising mapping layer 502 executes the step of calculating the Ising model input parameters of each candidate velocity vector based on the preset Ising model input parameter calculation rules and the comprehensive cost of each candidate velocity vector, and obtaining the Ising model input parameter set, it is specifically used for: Determine the maximum comprehensive cost among multiple candidate velocity vectors, and determine the current penalty coefficient based on the maximum comprehensive cost; Based on the preset Ising model input parameter calculation rules, the current penalty coefficient, and the comprehensive cost of each candidate velocity vector, the Ising model input parameters of each candidate velocity vector are calculated to obtain the Ising model input parameter set.

[0088] In some embodiments, the Ising model input parameters include local magnetic field input parameters and coupling strength input parameters, and the Ising model input parameter calculation rules include local magnetic field calculation formulas and coupling strength calculation formulas; when the Ising mapping layer 502 executes the step of calculating the Ising model input parameters of each candidate velocity vector based on the preset Ising model input parameter calculation rules, the current penalty coefficient, and the comprehensive cost of each candidate velocity vector, to obtain the Ising model input parameter set, it is specifically used for: Based on the local magnetic field calculation formula, the current penalty coefficient, and the comprehensive cost of each candidate velocity vector, the local magnetic field input parameters for each candidate velocity vector are determined. The local magnetic field calculation formula is as follows: ; in, The local magnetic field input parameters are the i-th candidate velocity vector. Let N be the total cost of the i-th candidate velocity vector, and N be the total number of the candidate velocity vectors. The current penalty coefficient; The coupling strength input parameter is determined based on the coupling strength calculation formula and the current penalty coefficient. The coupling strength calculation formula is as follows: ; in, The coupling strength is input as a parameter, where i and j are the indices of the candidate velocity vectors, and i < j; The set of input parameters for the Ising model is determined based on the local magnetic field input parameters of each candidate velocity vector and the coupling strength input parameters.

[0089] In some embodiments, when the decoding and flight control interface layer 504 performs the step of determining the flight control parameters of the UAV's current control cycle based on the ground state spin configuration, it is specifically used for: Determine whether there is only one spin with a positive 1 in the ground state spin configuration; If there is only one spin of positive 1, then the candidate velocity vector with a spin of positive 1 is determined as the target velocity vector, and the flight control parameters of the UAV in the current control cycle are determined based on the target velocity vector. If there is not only one spin of positive 1, the flight control parameters of the UAV in the current control cycle are determined according to the preset safety default strategy, which includes a hovering strategy or a deceleration strategy.

[0090] In some embodiments, when the cost calculation layer 501 performs the step of calculating the comprehensive cost of each of a plurality of preset candidate velocity vectors based on the UAV's own state information and the perception data of at least one obstacle within a preset range of the UAV, it is specifically used for: Collect the drone's own status information and the perception data of at least one obstacle within a preset range of the drone; Based on the perceived data, the position of each obstacle at the start of the next control cycle is predicted, resulting in a set of predicted obstacle positions. Based on the predicted obstacle location set and its own state information, the comprehensive cost of each candidate velocity vector among multiple preset candidate velocity vectors is calculated.

[0091] In some embodiments, when the cost calculation layer 501 performs the step of calculating the comprehensive cost of each candidate velocity vector among multiple preset candidate velocity vectors based on the obstacle prediction position set and its own state information, it is specifically used for: Based on its own state information, the predicted position of the UAV at the start time of the next control cycle under each of the candidate velocity vectors is predicted, and the collision risk cost corresponding to each candidate velocity vector is determined according to the predicted position of the UAV under each of the candidate velocity vectors and the set of predicted positions of obstacles. Calculate the cosine penalty of the angle between the velocity direction and the target direction of each candidate velocity vector to obtain the target deviation cost of each candidate velocity vector. The target direction is the unit direction vector from the current position of the UAV to the mission target point. The current position of the UAV is obtained based on its own state information. Based on each candidate velocity vector and the current velocity vector of the UAV, the current acceleration of the UAV within one control cycle for each candidate velocity vector is determined. Based on the current acceleration of each UAV and the preset maximum acceleration of the UAV, the dynamic constraint cost corresponding to each candidate velocity vector is determined. The current velocity vector of the UAV is obtained based on its own state information. The comprehensive cost of each candidate velocity vector is determined based on the collision risk cost, the target deviation cost, and the dynamic constraint cost.

[0092] In some embodiments, the obstacle perception data includes radar perception data and depth image data; when performing the step of collecting the UAV's own state information and the perception data of at least one obstacle within a preset range of the UAV, the cost calculation layer 501 is specifically used for: The radar sensing data is collected by the radar module carried by the UAV; The depth image data is acquired synchronously by the binocular vision module mounted on the drone; and, The drone's own status information is obtained through the navigation module on board the drone.

[0093] In some embodiments, when the cost calculation layer 501 performs the step of predicting the position of each obstacle at the start of the next control cycle based on the perception data to obtain a set of predicted obstacle positions, it is specifically used for: By performing clustering processing on the perceived data, at least one obstacle is identified; For each obstacle, based on the obstacle state vector maintained by the extended Kalman filter, the position of the obstacle is extrapolated to the start time of the next control cycle based on a uniform or uniformly accelerated motion model, thus obtaining the set of predicted obstacle positions.

[0094] In summary, on the one hand, when calculating the comprehensive cost of each candidate velocity vector and the corresponding Ising model input parameters, the computational load of this scheme is only linearly related to the total number of candidate velocity vectors (a preset constant) and the number of obstacles. The overall computational delay no longer increases sharply with the number of obstacles, enabling the UAV to quickly complete collision avoidance decisions within each control cycle in high-density, high-dynamic obstacle environments, meeting the reaction requirements of high-speed flight. On the other hand, in each control cycle, only the current state information and perception data need to be collected, and the position of obstacles in the next cycle needs to be predicted to calculate the flight control parameters for the current cycle. The entire decision-making process does not require pre-constructing a global environment map or planning a complete flight path, further improving the obstacle avoidance speed of the UAV.

[0095] It should be noted that those skilled in the art can clearly understand that the specific implementation process of the flight control system 500 and each unit of the above-mentioned UAV can be referred to the corresponding description in the aforementioned method embodiments. For the sake of convenience and brevity, it will not be repeated here.

[0096] The flight control system of the aforementioned UAV can be implemented as a computer program, which can be used in, for example... Figure 6 It runs on the electronic device shown.

[0097] Please see Figure 6 , Figure 6 This is a schematic block diagram of an electronic device provided in an embodiment of this application. The electronic device 600 can be a drone or a controller in a drone.

[0098] See Figure 6 The electronic device 600 includes a processor 602, a memory, and a network interface 605 connected via a system bus 601. The memory may include a non-volatile storage medium 603 and internal memory 604.

[0099] The non-volatile storage medium 603 may store an operating system 6031 and a computer program 6032. The computer program 6032 includes program instructions that, when executed, cause the processor 602 to perform a flight control method for an unmanned aerial vehicle.

[0100] The processor 602 provides computing and control capabilities to support the operation of the entire electronic device 600.

[0101] The internal memory 604 provides an environment for the operation of the computer program 6032 in the non-volatile storage medium 603. When the computer program 6032 is executed by the processor 602, the processor 602 can execute a flight control method for an unmanned aerial vehicle.

[0102] This network interface 605 is used for network communication with other devices. Those skilled in the art will understand that... Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the electronic device 600 to which the present application is applied. The specific electronic device 600 may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0103] The processor 602 is used to run a computer program 6032 stored in the memory to perform the following steps: At the beginning of each control cycle, based on the UAV's own state information and the perception data of at least one obstacle within the UAV's preset range, the comprehensive cost of each of the multiple preset candidate velocity vectors is calculated. Based on the preset Ising model input parameter calculation rules and the comprehensive cost of each candidate velocity vector, the Ising model input parameters of each candidate velocity vector are calculated to obtain the Ising model input parameter set; Annealing is performed on the input parameter set of the Ising model to output the ground state spin configuration; The flight control parameters for the current control cycle of the UAV are determined based on the ground state spin configuration.

[0104] It should be understood that, in the embodiments of this application, the processor 602 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0105] It will be understood by those skilled in the art that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program includes program instructions and can be stored in a storage medium, which is a computer-readable storage medium. The program instructions are executed by at least one processor in the computer system to implement the process steps of the embodiments of the above methods.

[0106] Therefore, this application also provides a storage medium. This storage medium can be a computer-readable storage medium. The storage medium stores a computer program, wherein the computer program includes program instructions. When executed by a processor, the program instructions cause the processor to perform the following steps: At the beginning of each control cycle, based on the UAV's own state information and the perception data of at least one obstacle within the UAV's preset range, the comprehensive cost of each of the multiple preset candidate velocity vectors is calculated. Based on the preset Ising model input parameter calculation rules and the comprehensive cost of each candidate velocity vector, the Ising model input parameters of each candidate velocity vector are calculated to obtain the Ising model input parameter set; Annealing is performed on the input parameter set of the Ising model to output the ground state spin configuration; The flight control parameters for the current control cycle of the UAV are determined based on the ground state spin configuration.

[0107] The storage medium can be any computer-readable storage medium capable of storing program code, such as a USB flash drive, portable hard drive, read-only memory (ROM), magnetic disk, or optical disk.

[0108] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0109] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the division of each unit is merely a logical functional division, and there may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0110] The steps in the methods of this application embodiment can be adjusted, merged, or deleted according to actual needs. The units in the apparatus of this application embodiment can be merged, divided, or deleted according to actual needs. Furthermore, the functional units in the various embodiments of this application 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.

[0111] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or 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 an electronic device (which may be a personal computer, a terminal, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0112] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A flight control method for an unmanned aerial vehicle (UAV), characterized in that, The method is applied to a drone, and the method includes: At the beginning of each control cycle, based on the UAV's own state information and the perception data of at least one obstacle within the UAV's preset range, the comprehensive cost of each of the multiple preset candidate velocity vectors is calculated. Based on the preset Ising model input parameter calculation rules and the comprehensive cost of each candidate velocity vector, the Ising model input parameters of each candidate velocity vector are calculated to obtain the Ising model input parameter set; Annealing is performed on the input parameter set of the Ising model to output the ground state spin configuration; The flight control parameters for the current control cycle of the UAV are determined based on the ground state spin configuration.

2. The method according to claim 1, characterized in that, Based on the preset Ising model input parameter calculation rules and the comprehensive cost of each candidate velocity vector, the Ising model input parameters of each candidate velocity vector are calculated to obtain the Ising model input parameter set, including: Determine the maximum comprehensive cost among multiple candidate velocity vectors, and determine the current penalty coefficient based on the maximum comprehensive cost; Based on the preset Ising model input parameter calculation rules, the current penalty coefficient, and the comprehensive cost of each candidate velocity vector, the Ising model input parameters of each candidate velocity vector are calculated to obtain the Ising model input parameter set.

3. The method according to claim 1, characterized in that, The process of determining the flight control parameters for the current control cycle of the UAV based on the ground-state spin configuration includes: Determine whether there is only one spin with a positive 1 in the ground state spin configuration; If there is only one spin of positive 1, then the candidate velocity vector with a spin of positive 1 is determined as the target velocity vector, and the flight control parameters of the UAV in the current control cycle are determined based on the target velocity vector. If there is not only one spin of positive 1, the flight control parameters of the UAV in the current control cycle are determined according to the preset safety default strategy, which includes a hovering strategy or a deceleration strategy.

4. The method according to claim 1, characterized in that, The step of calculating the comprehensive cost of each candidate velocity vector among multiple preset candidate velocity vectors based on the drone's own state information and the perception data of at least one obstacle within the drone's preset range includes: Collect the drone's own status information and the perception data of at least one obstacle within a preset range of the drone; Based on the perceived data, the position of each obstacle at the start of the next control cycle is predicted, resulting in a set of predicted obstacle positions. Based on the predicted obstacle location set and its own state information, the comprehensive cost of each candidate velocity vector among multiple preset candidate velocity vectors is calculated.

5. The method according to claim 4, characterized in that, The calculation of the comprehensive cost of each candidate velocity vector among multiple preset candidate velocity vectors based on the predicted obstacle position set and the self-state information includes: Based on its own state information, the predicted position of the UAV at the start time of the next control cycle under each of the candidate velocity vectors is predicted, and the collision risk cost corresponding to each candidate velocity vector is determined according to the predicted position of the UAV under each of the candidate velocity vectors and the set of predicted positions of obstacles. Calculate the cosine penalty of the angle between the velocity direction and the target direction of each candidate velocity vector to obtain the target deviation cost of each candidate velocity vector. The target direction is the unit direction vector from the current position of the UAV to the mission target point. The current position of the UAV is obtained based on its own state information. Based on each candidate velocity vector and the current velocity vector of the UAV, the current acceleration of the UAV within one control cycle for each candidate velocity vector is determined. Based on the current acceleration of each UAV and the preset maximum acceleration of the UAV, the dynamic constraint cost corresponding to each candidate velocity vector is determined. The current velocity vector of the UAV is obtained based on its own state information. The comprehensive cost of each candidate velocity vector is determined based on the collision risk cost, the target deviation cost, and the dynamic constraint cost.

6. The method according to claim 4, characterized in that, The obstacle perception data includes radar perception data and depth image data; the acquisition of the UAV's own state information and the perception data of at least one obstacle within the UAV's preset range includes: The radar sensing data is collected by the radar module carried by the UAV; The depth image data is acquired synchronously by the binocular vision module mounted on the drone; and, The drone's own status information is obtained through the navigation module on board the drone.

7. The method according to claim 4, characterized in that, The step of predicting the position of each obstacle at the start of the next control cycle based on the perceived data yields a set of predicted obstacle positions, including: By performing clustering processing on the perceived data, at least one obstacle is identified; For each obstacle, based on the obstacle state vector maintained by the extended Kalman filter, the position of the obstacle is extrapolated to the start time of the next control cycle based on a uniform or uniformly accelerated motion model, thus obtaining the set of predicted obstacle positions.

8. A flight control system for an unmanned aerial vehicle (UAV), characterized in that, The flight control system of the UAV is configured in the UAV and includes: The cost calculation layer is used to calculate the comprehensive cost of each candidate velocity vector among multiple preset candidate velocity vectors at the beginning of each control cycle, based on the UAV's own state information and the perception data of at least one obstacle within the UAV's preset range. The Ising mapping layer is used to calculate the Ising model input parameters of each candidate velocity vector based on the preset Ising model input parameter calculation rules and the comprehensive cost of each candidate velocity vector, thereby obtaining the Ising model input parameter set; An annealing coprocessor layer is used to perform annealing solution processing on the input parameter set of the Ising model and output the ground state spin configuration. The decoding and flight control interface layer is used to determine the flight control parameters of the UAV in the current control cycle based on the ground state spin configuration.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the flight control method for the UAV as described in any one of claims 1-7.

10. A storage medium, characterized in that, The storage medium stores a computer program, which includes program instructions that, when executed by a processor, cause the processor to perform the flight control method for the UAV as described in any one of claims 1-7.

Images