Delivery planning control method based on aerial mechanical arm

Abstract

The invention discloses a delivery planning control method based on an air mechanical arm, and relates to the technical field of control planning, and the method comprises the steps: building a mathematical model of the air mechanical arm; the aerial mechanical arm comprises a four-rotor platform and a Delta mechanical arm mechanically connected with the four-rotor platform. Establishing a casting motion model after the Delta mechanical arm in the aerial mechanical arm releases the load; according to the mathematical model and the casting motion model, an air-drop track is planned, and a prediction control instruction for layered disturbance compensation is generated; the release time of the load is evaluated online according to the empty head track and the prediction control instruction; and controlling the aerial mechanical arm to release the load according to the release time, the air-drop track and the predictive control instruction. According to the method, the feasible air-drop trajectory is generated by establishing the casting motion model for load release, the sensitivity of landing precision to the uncertainty of the load release time is effectively reduced, the prediction control instruction is generated through the hierarchical disturbance compensation strategy to control the air mechanical arm, and the robustness can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of control planning, in particular to a delivery planning control method based on an aerial manipulator. BACKGROUND

[0002] With the development of low-altitude economy and aerial transportation, aerial vehicles have been increasingly applied to challenging delivery tasks such as emergency rescue and disaster response. In these tasks, aerial delivery operation becomes an indispensable choice when direct landing or stable placement cannot be achieved due to limited terrain access, unstable ground conditions or time urgency, etc. For example, in forest fire or jungle support tasks, aerial delivery is often required to maintain flight stability and safety. However, manual aerial delivery is highly dependent on operator skills and has poor consistency. In contrast, autonomous aerial delivery systems based on unmanned aerial vehicles can provide more reliable and accurate solutions, thereby improving the efficiency and safety of emergency delivery tasks.

[0003] Traditional autonomous aerial delivery systems have limited robustness and accuracy. Robustness is particularly prominent in aerial delivery tasks: the system needs to switch from pre-release control mode to post-release mode at the instant of load release during high-speed flight, which often leads to system instability. If a fixed parameter controller is used, control divergence may occur, even posing safety risks. On the other hand, aerial delivery accuracy is influenced by multiple factors. First, achieving accurate delivery during high-speed agile flight is highly sensitive to the timing of load release, which poses challenges to the control of load grasping and release actuators. In addition, tracking errors and system delays of unmanned aerial vehicles are inherent characteristics that are difficult to eliminate completely; even under ideal conditions, the accumulation of these errors may cause release timing deviation, further amplifying the drop point error. SUMMARY

[0004] Therefore, embodiments of the present application provide a delivery planning control method based on an aerial manipulator and related equipment to improve the robustness of aerial delivery.

[0005] An aspect of an embodiment of the present application provides a delivery planning control method based on an aerial manipulator, the method comprising the following steps:

[0006] establishing a mathematical model of the aerial manipulator; wherein the aerial manipulator comprises a quadrotor platform and a Delta manipulator mechanically connected thereto;

[0007] establishing a projectile motion model of the Delta manipulator in the aerial manipulator after releasing a load;

[0008] planning an aerial delivery trajectory and generating a predictive control instruction with layered disturbance compensation according to the mathematical model and the projectile motion model, respectively;

[0009] The timing of load release is evaluated online based on the empty trajectory and the predictive control command;

[0010] The control robotic arm releases the load according to the release timing, the airdrop trajectory, and the predictive control command.

[0011] In some embodiments, establishing a mathematical model of the aerial robotic arm includes the following steps:

[0012] The relationship between the quadcopter platform and the Delta robotic arm is modeled as follows:

[0013]

[0014] in, Indicates location, It represents angular velocity, and its antisymmetric matrix is ​​used to express the cross product of vectors;

[0015] The translational mechanics model for the quadrotor platform is as follows:

[0016]

[0017] in, For total thrust, For system quality, For machine system The direction of the axis in the inertial frame. The force generated by the interaction between external disturbances and the Delta robotic arm;

[0018] The rotational kinematics and dynamics of the quadcopter platform are modeled as follows:

[0019]

[0020] in, For the total torque, The inertia matrix, Additional torque applied to the Delta robotic arm's interaction with external disturbances.

[0021] In some embodiments, establishing a projectile motion model of the Delta robotic arm after releasing the load in the aerial robotic arm includes the following steps:

[0022] According to the load at the release time Location With speed Determine the landing point of the load. ;

[0023] Calculate the time it takes for the load to reach the target horizontal plane. ;

[0024]

[0025] in, Indicates the location of the load. Indicates the speed of the load;

[0026] Determine the load in time The trajectory satisfies the following relationship:

[0027] ;

[0028] Therefore, the landing point is determined to be:

[0029] .

[0030] In some embodiments, the step of planning the airdrop trajectory based on the mathematical model and the projectile motion model specifically includes the following steps:

[0031] The trajectory optimization problem is defined as follows:

[0032]

[0033] The trajectory optimization problem includes minimizing control energy, including a time regularization term, and includes corridor constraints, kinematic constraints, dynamic constraints, and landing point constraints. It is defined in The set of inequality constraints on, and These represent the initial state and the terminal state of the quadcopter platform and the end effector, respectively.

[0034] Construct the following penalty function to constrain the landing point error:

[0035]

[0036] in , Indicates time interval, A smoothing function for the landing point penalty. The definition is as follows:

[0037] ;

[0038] in, To control the smoothness of the relaxation function; This is used to smoothly introduce integer variables into the nonlinear trajectory optimization problem through a defined relaxation function;

[0039] In solving the trajectory optimization problem, the same penalty is applied to the landing point within the time interval τ surrounding the optimal release time tr, thereby generating the airdrop trajectory containing a series of feasible release times. .

[0040] In some embodiments, the step of generating predictive control commands for hierarchical disturbance compensation based on the mathematical model and the projectile motion model specifically includes the following steps:

[0041] The state of the quadcopter platform is defined as follows: The control input is ;

[0042] Define the time horizon Classified as There are several equal intervals, with an interval length of [missing information]. ,in The length of the horizon;

[0043] This leads to a constrained nonlinear optimization problem:

[0044] ;

[0045] Among them, the function This represents the dynamic model of the quadcopter platform; reference state. and input The airdrop trajectory is used as a reference for the position, attitude, and velocity of the UAV, and the output is the control commands for the UAV: ​​total thrust and angular velocity. , and These are weighted matrices representing the state, input, and terminal state, respectively. It is the current state estimate when solving the optimal solution to the nonlinear optimization problem;

[0046] A nonlinear disturbance observer is used to compensate for the external disturbance force, and the following calculations are performed:

[0047]

[0048] in, This represents the acceleration measured by the inertial measurement unit;

[0049] Construct the observer matrix, and perform differentiation and discretization to obtain:

[0050]

[0051] in, It is after the cutoff frequency is Butterworth filtering; Represents the reciprocal of the sensor sampling rate; These are parameters adjusted based on actual working conditions.

[0052] Through each angular velocity The thrust is obtained based on the dynamic model of the quadcopter platform. and torque vector :

[0053]

[0054] in, and The dimensions of the quadcopter platform shown are as follows; and These represent the thrust coefficient and the moment coefficient, respectively. The moment of inertia of the propeller;

[0055] Desired angular acceleration The calculation is as follows:

[0056]

[0057]

[0058] in, Represents the filtered feedback angular acceleration, from Differentiation yields; This is the feedback angular torque based on each rotor angular velocity;

[0059] Therefore, the commands for controlling thrust and control torque are obtained as follows:

[0060]

[0061] in, The current rotor angular velocity, The time constant of the motor dynamics; and All filters are processed using a second-order Butterworth filter with the same cutoff frequency.

[0062] In some embodiments, the online assessment of the load release timing based on the short-run trajectory and the predictive control command includes the following steps:

[0063] Define reference airdrop trajectory Reference release time on Corresponding to the optimal landing point;

[0064] Determine the actual optimal release time caused by external disturbances. And satisfy ;

[0065] Define each discrete state of the load in the prediction time domain. Released at the current time, thus generating a set of predicted landing locations corresponding to the current time. :

[0066]

[0067] in, and Indicates the predicted position and velocity in the vertical direction;

[0068] Calculate the corresponding landing error sequence:

[0069] ;

[0070] in, This represents the sequence of landing errors between the predicted landing point and the target location;

[0071] Through continuous updates To minimize the landing error sequence.

[0072] In some embodiments, controlling the control robotic arm to release the load based on the release timing, the airdrop trajectory, and the predictive control command includes the following steps:

[0073] The electromagnet controller is controlled according to the release timing; wherein, the electromagnet controller is used to control whether the end of the Delta robotic arm releases the load;

[0074] The Delta robotic arm controller is controlled according to the airdrop trajectory.

[0075] The movement of the quadcopter platform is controlled according to the predictive control command.

[0076] Another aspect of this application embodiment provides a delivery planning and control device based on an aerial robotic arm, the device comprising:

[0077] A mathematical modeling unit is used to establish a mathematical model of the aerial robotic arm; wherein the aerial robotic arm includes a quadcopter platform and its mechanically connected Delta robotic arm;

[0078] The projectile modeling unit is used to establish a projectile motion model of the Delta robotic arm in the aerial robotic arm after the load is released.

[0079] The instruction generation unit is used to plan the airdrop trajectory and generate predictive control instructions for layered disturbance compensation based on the mathematical model and the projectile motion model, respectively.

[0080] The release time evaluation unit is used to evaluate the timing of load release online based on the empty trajectory and the predictive control command;

[0081] A release control unit is used to control the control robotic arm to release the load according to the release timing, the airdrop trajectory, and the predictive control command.

[0082] Another aspect of this application embodiment provides an electronic device, including a processor and a memory;

[0083] The memory is used to store programs;

[0084] The processor executes the program to implement any of the methods described above.

[0085] Another aspect of this application provides a computer-readable storage medium storing a program that is executed by a processor to implement the method described in any of the above embodiments.

[0086] This application includes at least the following beneficial effects:

[0087] This application establishes a mathematical model of an aerial robotic arm, comprising a quadcopter platform and its mechanically connected Delta robotic arm. It then establishes a projectile motion model of the Delta robotic arm after releasing the load. Based on the mathematical model and the projectile motion model, it plans the airdrop trajectory and generates predictive control commands with layered disturbance compensation. The timing of load release is evaluated online based on the airdrop trajectory and predictive control commands. Finally, it controls the aerial robotic arm to release the load based on the release timing, airdrop trajectory, and predictive control commands. This application, by establishing a projectile motion model for load release, generates a feasible airdrop trajectory, effectively reducing the sensitivity of landing accuracy to the uncertainty of load release timing. Furthermore, by incorporating a layered disturbance compensation strategy to generate predictive control commands, controlling the aerial robotic arm according to these commands improves the system's robustness and the load's landing accuracy. Attached Figure Description

[0088] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0089] Figure 1 A flowchart illustrating the delivery planning and control method based on an aerial robotic arm provided in this application embodiment;

[0090] Figure 2 An example diagram illustrating a scenario of aerial robotic arm delivery provided in an embodiment of this application;

[0091] Figure 3 An example flowchart of a delivery planning and control method based on an aerial robotic arm provided in an embodiment of this application;

[0092] Figure 4 A schematic diagram illustrating the release timing provided in an embodiment of this application;

[0093] Figure 5 A schematic diagram of an aerial robotic arm provided for an embodiment of this application;

[0094] Figure 6 Example diagram of airdrop planning results provided in the embodiments of this application;

[0095] Figure 7 This is a schematic diagram showing the results of a controller ablation experiment provided in an embodiment of this application;

[0096] Figure 8 A schematic diagram of the throwing results and data acquisition device provided in the embodiments of this application;

[0097] Figure 9 This is a structural block diagram of a delivery planning and control device based on an aerial robotic arm, provided in an embodiment of this application. Detailed Implementation

[0098] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0099] Before providing a detailed description of the embodiments of this application, some related technologies involved in the embodiments of this application will be described first, as follows:

[0100] Existing research on aerial dropping does not adequately consider the impact of actuator delay and control error on landing accuracy. On the one hand, applying only point-by-point landing constraints to the parabolic trajectory results in the trajectory remaining highly sensitive to the release timing. On the other hand, to cope with model changes caused by load release, aerial dropping systems should introduce adaptive controllers or disturbance observers. UAV load systems offer a feasible solution for improving the efficiency of aerial dropping missions. Although their mechanical structure is simple and their load capacity is high, their robustness is limited due to the inability to actively control the load. To address these issues, this application employs an aerial robotic arm (AM) with redundant degrees of freedom to perform aerial dropping missions.

[0101] This application proposes an aerial delivery scheme based on an aerial robotic arm (AM), which compensates for the trajectory tracking error of the UAV by introducing additional driving degrees of freedom. During trajectory generation, constraints are imposed on the target landing position within a set time interval, simultaneously obtaining the end effector trajectory and a feasible delivery time window, thus ensuring accurate payload delivery along a reference trajectory. Subsequently, this reference trajectory is provided as input to an NMPC controller with a hierarchical disturbance compensation mechanism. Furthermore, utilizing the model prediction characteristics of the NMPC, a method for online re-evaluation of the payload release timing is proposed. Simulation and real-world experimental results both demonstrate that the proposed method significantly improves landing accuracy under different airdrop trajectories. The key technical solutions of this application include:

[0102] An autonomous aerial delivery scheme based on an aerial robotic arm (AM) is proposed, enabling airdrop missions to be performed with greater flexibility and accuracy.

[0103] A method is proposed to apply continuous constraints to the parabolic landing point to generate a feasible airdrop trajectory, which effectively reduces the sensitivity of landing accuracy to the uncertainty of the payload release timing.

[0104] By incorporating a hierarchical perturbation compensation strategy into the NMPC framework, the robustness and landing accuracy of the system are significantly improved.

[0105] Some relevant technical details are as follows:

[0106] In recent years, the planning and control methods for autonomous airdrop missions have received widespread attention, with research focusing primarily on trajectory planning and the accuracy and precision of the dropping action. Related technologies utilize fixed-wing UAVs to autonomously identify target landing points and perform trajectory planning and control for airdrops in complex scenarios. However, fixed-wing platforms still have limitations in terms of maneuverability and dropping accuracy. Furthermore, other related technologies have experimentally demonstrated the feasibility of using traditional quadcopter platforms to perform dropping actions for UAV firefighting missions. Several studies have employed UAV payload systems for aerial dropping and verified their effectiveness through simulations. Although payload systems offer advantages such as simple mechanical structure and high load capacity, their cable-suspended structure cannot actively drive the load, limiting the system's agility and degrees of freedom.

[0107] In UAV airdrop trajectory planning, existing research has modeled the airdrop and UAV planning tasks as a mixed-integer quadratic programming problem and verified its effectiveness through simulation. Meanwhile, some related technologies apply the same landing point error constraint during trajectory optimization. However, restricting the reference landing point to the vicinity of the target with soft constraints does not guarantee that the reference trajectory landing point will be completely consistent with the actual target, thus generating a first-level landing point error. The second-level error originates from the controller: at the moment of load release, the UAV model parameters undergo abrupt changes, leading to a decrease in control performance. For example, model predictive control (MPC) can improve control performance in airdrop missions, but model parameter uncertainty may cause the AM's NMPC controller to operate in a suboptimal state, ultimately resulting in a significant deviation between the actual and planned landing points. To address these issues, other related technologies have proposed hybrid position-force control methods for air operations, employing online multi-task optimization or combining contact force NMPC to improve system robustness and accuracy under model uncertainty. Still other related technologies propose a framework centered on the end effector, decoupling high-level decision-making from low-level control, thereby enhancing the versatility and robustness of air operations. In addition, a motion and force planning control system that combines contact sensing is proposed in the related technologies, which enables AM to accurately track time-varying contact forces and trajectories, thereby improving dynamic task performance.

[0108] In recent years, robot throwing behavior has become a research hotspot. For example, researchers have analyzed the impact of uncertainties in robotic arm models on throwing accuracy and optimized trajectory generation accordingly. Another example is the proposal of an end-to-end learning framework combining physical simulation and deep learning, enabling robots to grasp and throw arbitrary objects to a target location based on visual input. However, due to limitations in the gripper structure, the system remains highly sensitive to the timing of load release. Furthermore, some researchers have explicitly utilized redundant degrees of freedom to ensure that the motion state of the load remains within an effective throwing configuration at the moment of release. Similarly, a learning-based full-body throwing framework has been proposed for quadrupedal mobile manipulators, significantly improving throwing accuracy and success rate under conditions of release uncertainty.

[0109] To address the problems of existing technologies, this application applies continuous landing point constraints within the space-time trajectory interval and utilizes the spatial degrees of freedom of the aerial drop mission to generate continuous feasible release time windows, thereby reducing the system's sensitivity to the uncertainty of the load release timing.

[0110] Reference Figure 1 This application provides a delivery planning and control method based on an aerial robotic arm, specifically including the following steps S100~S140:

[0111] S100: Establish a mathematical model of the aerial robotic arm; wherein the aerial robotic arm includes a quadcopter platform and its mechanically connected Delta robotic arm;

[0112] S110: Establish a projectile motion model of the Delta robotic arm in the aerial robotic arm after releasing the load;

[0113] S120: Based on the mathematical model and the projectile motion model, plan the airdrop trajectory and generate predictive control commands for layered disturbance compensation, respectively;

[0114] S130: Evaluate the timing of load release online based on the empty trajectory and the predictive control command;

[0115] S140: Control the control robotic arm to release the load according to the release timing, the airdrop trajectory, and the predictive control command.

[0116] The following section will provide a detailed introduction and explanation of the solutions in the embodiments of this application, using specific application examples.

[0117] like Figure 2 As shown, in complex environments where ground access is restricted or hazards are directly placed, autonomous flight systems are becoming increasingly important in transportation and delivery missions. In airdrop missions, flight platforms face the dual challenges of abrupt changes in control modes, inherent system delays, and control errors. To address these issues, this application proposes an autonomous airdrop system based on an aerial manipulator (AM). Additional degrees of freedom (DoF) enable the system to actively compensate for trajectory tracking errors of unmanned aerial vehicles (UAVs). By applying smooth and continuous constraints to the parabolic landing point, the proposed method can generate an airdrop trajectory insensitive to the timing of load release. Furthermore, this application introduces a hierarchical disturbance compensation strategy within a nonlinear model predictive control (NMPC) framework to mitigate the impact of abrupt changes in system parameters and further utilizes the predictive capabilities of NMPC to improve airdrop accuracy. Simulation and field experiments both demonstrate that the proposed system achieves greater agility and higher accuracy in airdrop missions.

[0118] Specifically, the aerial robotic arm used in this embodiment consists of a quadcopter platform and a Delta robotic arm. Since the distance between the end effector and the load is negligible, the state of the load can be approximated as the state of the end effector before release. This embodiment defines three right-handed coordinate systems: an inertial coordinate system... ,That The axis is opposite to gravity; the system ,That Pointing towards the nose of the aircraft Aligned with the direction of total thrust; end effector coordinate system Its origin is located on the end effector, and it only has a translational relationship with the machine system. From arrive The rotation of quaternions Parameterization, the rotation matrix is ​​represented as Subscripts are used to indicate physical meaning, while superscripts are used to indicate the coordinate system in which the variable is located.

[0119] Specifically, the embodiments of this application include the following technical solutions:

[0120] 1. Aerial robotic arm model.

[0121] This application uses the flat output and derivative of the end effector to describe the state and input of the quadcopter platform. Due to the structural characteristics of the Delta robotic arm, its end effector can be approximated as being consistent with the quadcopter platform in terms of rotational dynamics. Changes in the attitude of the end effector introduce a wobbling effect, requiring correction of the translational state of the quadcopter platform. The relationship between the two can be given by the following equation:

[0122]

[0123] in Indicates location, The angular velocity is represented by its antisymmetric matrix, which is used to express the vector cross product. The translational mechanics of a quadrotor platform follows:

[0124]

[0125] in For total thrust, For system quality, For machine system The direction of the axis in the inertial frame. The force generated by the interaction between external disturbances and the robotic arm. The rotational kinematics and dynamics are expressed as follows:

[0126]

[0127] in For the total torque, The inertia matrix, Additional torque applied to the interaction of the robotic arm with external disturbances.

[0128] 2. Parabolic motion model.

[0129] This embodiment describes the projectile motion model of the load after release. The landing point of the load... Especially at the moment of release Location With speed The time it takes for the load to reach the target horizontal plane. It can be calculated using the following formula:

[0130]

[0131] in, Indicates the location of the load. Indicates the speed of the load;

[0132] Load in time The trajectory follows:

[0133]

[0134] Therefore, the final landing point can be written as:

[0135]

[0136] The above model uses simplified assumptions: the load is not affected by external forces in the horizontal direction and maintains the speed at the moment of release; the load is only affected by gravity in the vertical direction.

[0137] Next, the overall process of the autonomous aerial delivery system in the embodiments of this application will be described, such as... Figure 3 As shown, the aerial delivery planning module generates reference trajectories for the UAV and Delta robotic arm through front-end JPS path search and back-end trajectory optimization. The nonlinear model predictive control (NMPC) framework employs a hierarchical disturbance compensation strategy, using NDOB and INDI to estimate external forces and torques respectively. Based on the predictive capabilities of NMPC, this embodiment designs an online re-evaluation module, which updates the payload release time online using a reference release time window and controls the electromagnet accordingly. The delta arm controller is implemented as a PID controller, outputting angular velocity commands for the servo motor.

[0138] The main sources of error for aerial robotic arms include: reference landing point error generated by the planner, tracking error of the controller, and system delay. To address this, this application proposes methods from three levels—trajectory planning (Sec.IV-A), control framework design (Sec.IV-B), and load release timing reassessment (Sec.IV-C)—to tackle error accumulation.

[0139] 3. Airdrop trajectory planning.

[0140] 1) Nonlinear optimization framework: M-segment D-dimensional polynomial splines The order of each segment is , is used to represent the entire flat output trajectory, where Let be the order of the relevant integral chain. In the implementation of this application embodiment, set... This application uses... Indicates the length of time for each segment. This represents the total time. For the [number]th... Segment trajectory, Defined as ,in This is the coefficient matrix for this segment. These are natural basis functions.

[0141] Due to its high efficiency in generating smooth trajectories while minimizing control inputs, the embodiments of this application employ... The framework was developed, and a continuously differentiable penalty function was constructed to impose landing point constraints, thereby ensuring the feasibility and smoothness of the trajectory. coefficients of the original polynomial Mapping to intermediate path points And time allocation, through parameter mapping relationships ,in , It is a smooth mapping function with linear complexity.

[0142] The trajectory optimization problem can be expressed in the following form:

[0143]

[0144] Equation (7a) minimizes the control energy and includes a time regularization term. Equations (7b) and (7c) represent inequalities and equality constraints, including corridor constraints, kinematic constraints, dynamic constraints, and landing point constraints. It is defined in The set of inequality constraints on, and These represent the initial and final states of the quadcopter platform and the end effector, respectively.

[0145] 2) Constraint Definition: This application embodiment uses a unified model for corridor constraints, kinematic constraints, and dynamic constraints. The planner outputs the end effector in... and The complete state in the process is then converted into the complete state of the quadrotor platform using formula (1), and input. Controller. The landing point error is constrained by constructing the following penalty function:

[0146]

[0147] in , Indicates time interval, The smoothing function for the landing point penalty is defined as follows:

[0148]

[0149] in Control the smoothness of the relaxation function. Integer variables are smoothly introduced into the nonlinear programming model using a defined relaxation function. This formula does not directly constrain the release position, but rather allows the optimization process to determine the optimal release position itself.

[0150] Considering the response delay of the actuator when controlling the load holder, applying a landing point constraint only at a single release moment is insufficient during optimization. For example... Figure 4 As shown, to reduce the sensitivity of throwing accuracy to the precise timing of load release, this embodiment applies the same penalty to the landing point within the time interval τ surrounding the optimal release time tr determined during the optimization process. This allows the planner to fully utilize the spatial redundancy of the throwing task and generate a trajectory containing a series of feasible release times. .

[0151] 4. Layered Disturbance Compensation (NMPC) Framework.

[0152] 1) Nonlinear Model Predictive Control : Control commands are generated by solving a finite-time optimal control problem within a rolling time-domain framework. Given a reference throw trajectory, its cost function is constructed based on the deviation between the predicted state and the reference state within the time horizon, thus simultaneously considering multiple reference points within the horizon during the optimization process.

[0153] In this embodiment, the state of the quadcopter platform is represented as follows: The control input is Time horizon Classified as There are several equal intervals, with an interval length of [missing information]. ,in Let be the horizon length. This discretization process yields a constrained nonlinear optimization problem:

[0154]

[0155] function The model of the quadrotor platform is given by equations (2) and (3). Reference state. and input The data comes from the airdrop trajectory, with reference states including the drone's position, attitude, and velocity. The output consists of the drone's control commands: total thrust and angular velocity. , and These are weighted matrices representing the state, input, and terminal state, respectively. This is the current state estimate when solving the optimal control problem. This nonlinear optimization problem is solved by... toolkit and It can be solved jointly, and a real-time iterative approach can be used for the solution.

[0156] 2) Layered disturbance compensation: Considering the variability and uncertainty of payload mass in airdrop missions, this embodiment employs a nonlinear disturbance observer. Compensation is provided for external disturbances to improve system robustness.

[0157] The embodiments of this application can calculate:

[0158]

[0159] in Indicated by inertial measurement unit The measured acceleration. Subsequently, the observer matrix is ​​constructed and differentiated and discretized to obtain:

[0160]

[0161] in After the cutoff frequency is Butterworth filtering. This represents the reciprocal of the sensor sampling rate. To achieve fast convergence in the perturbation estimation, the parameters... Adjustments should be made based on actual working conditions. (This refers to adjustments made using various angular velocities.) The thrust can be obtained based on the model of the quadcopter platform. and torque vector :

[0162]

[0163] in and This refers to the dimensions of the quadcopter platform. and These represent the thrust coefficient and the torque coefficient, respectively. Let be the moment of inertia of the propeller.

[0164] Before release, the load is fixed at an offset position from the center of mass of the quadcopter platform, thereby continuously generating additional torque acting on the system. Since the lever arm depends on the configuration of the robotic arm and is difficult to measure accurately in real time, this embodiment implements incremental nonlinear dynamic inversion in the inner-loop angular velocity controller. The solution effectively compensates for this external torque. The desired angular acceleration... The calculation is as follows:

[0165]

[0166]

[0167] in, The filtered feedback angular acceleration can be represented by... Differentiating yields the result. The feedback angular torque is based on each rotor angular velocity (see formula (13c)). Finally, according to (13a), the control thrust and control torque commands can be obtained:

[0168]

[0169] in, The current rotor angular velocity, This is the time constant of the motor dynamics. To reduce... The impact of noise on sensor measurements and All filters are processed using a second-order Butterworth filter with the same cutoff frequency.

[0170] 6. Reassess the timing of the release.

[0171] Previous studies on airdrops typically release the load at the nominal time specified by the high-level planner, or update the reference release time at a relatively low frequency. However, the accumulation of system latency and control errors can affect the optimality of this reference time. This application proposes a release time re-evaluation strategy utilizing the predictive capabilities of NMPC.

[0172] Although the method in this application embodiment can achieve a shorter feasible time window for load release, precise timing for triggering the release is still required. This application embodiment assumes a reference airdrop trajectory. Reference release time on This corresponds to the optimal landing point. In reality, external disturbances can affect the actual optimal release time. Relative to nominal value Slightly deviated, but satisfactory .

[0173] Therefore, starting from the reference release time From the moment of prediction in the time domain until the current time coincides with the reference release time, the prediction window always includes the current flight trajectory. The optimal release point.

[0174] To fully utilize the predictive properties of the model, the embodiments of this application assume that the load is in each discrete state within the prediction time domain. Release at the location. According to formulas (4) and (6), this will produce a set of predicted landing locations corresponding to the current time. :

[0175]

[0176] in and This represents the predicted position and velocity in the vertical direction. This allows for the calculation of the corresponding landing error sequence, denoted as... :

[0177]

[0178] Since the planned reference landing point may not completely coincide with the expected target location, this embodiment of the application bypasses the reference landing point and directly calculates the error between the predicted landing point and the target location.

[0179] Based on the predicted landing error sequence, incorrect release logic may lead to missing or misjudging the optimal release time. To address this, this application proposes an efficient decision-making mechanism that dynamically searches for the optimal release time in real time during flight, as shown in the algorithm below. This algorithm continuously updates... To minimize landing error. By appropriately setting termination conditions, actuator delay can also be considered and prevented. Behavior.

[0180]

[0181] 7. Experimental verification.

[0182] This section evaluates the performance of the proposed airdrop system through simulation and live experiments. Experimental verification includes three aspects: 1) Verifying the generation of airdrop trajectories with continuous landing point constraints, demonstrating its effect in reducing release uncertainty sensitivity (Sec.VA); 2) Ablation experiments of the hierarchical disturbance compensation framework, illustrating the robustness of the control system (Sec.VB); 3) Comparative experiments on system drop accuracy under various airdrop flight trajectories, verifying the effectiveness of the proposed method in suppressing inherent system errors (such as system delay) (Sec.VC).

[0183] This system was developed using C++11 under Ubuntu 20.04 and ROSNoetic environments. The experimental platform is as follows: Figure 5 As shown, the platform consists of a quadcopter platform equipped with a Delta robotic arm, powered by an NVIDIA Jetson Orin NX16GB as its onboard computer and an NxtPX4v2 flight controller. With a 6S battery installed, the platform's total weight is approximately 1.59 kg. The Delta arm controller uses a PID controller. The Delta arm is driven by three DYNAMIXELXL430-W250-T servo motors, with an electromagnet at its end. During flight, the electromagnet attracts the load; when the current time reaches... Immediately upon activation, the magnetizer is demagnetized to release the load. For state estimation, an extended Kalman filter (EKF) fuses measurements from the NOKOV motion capture system and the onboard IMU to provide robust and accurate system state. Three different load shapes were used in the live-fire experiments, such as... Figure 5 As shown on the right, the masses are 50g, 200g, and 300g respectively.

[0184] 8. Display of airdrop planning results.

[0185] This section evaluates the performance of the planning module in the Rviz visualization environment without physical simulation, where the optimization problem is solved using the L-BFGS method. The parameters in the relaxation function (8) are adjusted during the optimization process. As the value increases, the feasible release time interval in the optimized trajectory is correspondingly extended, such as... Figure 6 As shown in (a) above. This is because during the penalty function's action, the planner reduces the flight speed, causing the drone to hover almost above the target. Therefore, the action changes from a throw dependent on horizontal velocity to a fall, thus reducing the trajectory's sensitivity to release uncertainties. However, overactivation of the relaxation function can affect trajectory smoothness.

[0186] Adjust parameters The optimizer can be made to change the deployment flight state, such as... Figure 6 As shown in (b) above. Lowering the flight altitude shortens the freefall time, thus enabling a more precise landing with minimal changes in flight speed. Therefore, the release action gradually changes from throwing to placing, corresponding to different load release modes.

[0187] 9. Ablation experiment of controller.

[0188] This section evaluates the proposed NMPC control framework through ablation experiments covering three different parabolic velocities. To test robustness, loads of unknown mass (50g and 200g) were dropped, with ten trials conducted under each scenario and condition. The NMPC controller operated at 100Hz, with parameters shown in Table I. Landing point coordinates were recorded during actual flights, and the dispersion of actual landing points relative to the reference point was evaluated using RMSE and MAX (Table II), where vr is the reference release velocity. The results show that the NMPC framework equipped with a hierarchical disturbance observer with NDOB and INDI achieves a more concentrated landing distribution by compensating for the internal coupling error of the aerial robotic arm and external disturbances. In particular, NDOB's compensation for linear forces enhances trajectory tracking performance, resulting in an average increase of approximately 46.6% in the concentration of load landing points under different release velocities.

[0189]

[0190]

[0191] Figure 7 (a) shows the NDOB measurement in the horizontal direction. The load changes from -1.96N to 0N upon release. This enables the autonomous airdrop system to handle loads of unknown and varying mass, and to compensate for the resulting changes in model parameters. Figure 7 (b) shows the position and velocity tracking of the end effector in the x and z axis directions.

[0192] 10. System throwing accuracy experiment.

[0193] This section designed three different flight trajectories and conducted comparative experiments: on the one hand, a nominal release trigger was used, meaning the load was released strictly according to the reference time specified by the planner; on the other hand, a prediction-based release timing re-evaluation method proposed in the embodiments of this application was used. The flight trajectories exhibited different states near the load release, thus their sensitivity to release uncertainty also varied. Three different loads (such as...) were used in the experiments. Figure 5 As shown), the AM controller parameters remain unchanged, and control is achieved through the above-mentioned hierarchical disturbance compensation framework. Ten experiments are conducted under each load condition.

[0194] Figure 8 (a) in the diagram shows the result of the throw. Figure 8 (b) shows the physical data acquisition device, and the experimental results are summarized in Table III. The actual and target landing point errors of both methods were recorded and their performance was quantified using the MEAN / MAX metric. Experimental results show that the proposed re-evaluation method can effectively handle different airdrop missions. Although slight delays and errors may lead to large landing point deviations in aggressive flight maneuvers, the method still maintains high accuracy. For medium-difficulty trajectories, the prediction-based method can identify more suitable spatiotemporal states for payload release within the NMPC prediction time domain.

[0195]

[0196] In summary, this application proposes an autonomous aerial system based on an aerial robotic arm and a corresponding control method for achieving precise payload delivery in complex environments. By planning a trajectory with continuous landing point constraints, the system effectively reduces its sensitivity to release uncertainties. Layered disturbance compensation enhances system robustness, ensuring precise control even when the payload is unknown. Utilizing the predictive capabilities of NMPC, the system dynamically reassesses the release timing during flight to improve landing point accuracy. Future work will focus on safe aerial transport in dynamic environments, including grasping, transporting, and releasing tasks under wind disturbances and payload type uncertainties.

[0197] Reference Figure 9 This application provides a delivery planning and control device based on an aerial robotic arm, comprising:

[0198] A mathematical modeling unit is used to establish a mathematical model of the aerial robotic arm; wherein the aerial robotic arm includes a quadcopter platform and its mechanically connected Delta robotic arm;

[0199] The projectile modeling unit is used to establish a projectile motion model of the Delta robotic arm in the aerial robotic arm after the load is released.

[0200] The instruction generation unit is used to plan the airdrop trajectory and generate predictive control instructions for layered disturbance compensation based on the mathematical model and the projectile motion model, respectively.

[0201] The release time evaluation unit is used to evaluate the timing of load release online based on the empty trajectory and the predictive control command;

[0202] A release control unit is used to control the control robotic arm to release the load according to the release timing, the airdrop trajectory, and the predictive control command.

[0203] It is understood that the content of the above method embodiments is applicable to the present device embodiments. The specific functions implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0204] In some alternative embodiments, the functions / operations mentioned in the block diagrams may not occur in the order shown in the operation diagrams. For example, depending on the functions / operations involved, two consecutively shown blocks may actually be executed substantially simultaneously, or the blocks may sometimes be executed in reverse order. Furthermore, the embodiments presented and described in the flowcharts of this application are provided by way of example to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logical flows presented in the embodiments of this application. Alternative embodiments are contemplated, in which the order of various operations is changed and sub-operations described as part of a larger operation are executed independently.

[0205] Furthermore, although this application is described in the context of functional modules, it should be understood that, unless otherwise stated, one or more of the described functions and / or features may be integrated into a single physical device and / or software module, or one or more functions and / or features may be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding this application. Rather, considering the properties, functions, and internal relationships of the various functional modules in the apparatus disclosed in the embodiments of this application, the actual implementation of the module will be understood within the conventional skills of an engineer. Therefore, those skilled in the art can implement the application set forth in the claims using ordinary techniques without excessive experimentation. It is also understood that the specific concepts disclosed are merely illustrative and not intended to limit the scope of this application, which is determined by the full scope of the appended claims and their equivalents.

[0206] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0207] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0208] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0209] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0210] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0211] Although embodiments of this application have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of this application, the scope of which is defined by the claims and their equivalents.

[0212] The above is a detailed description of the preferred embodiments of this application, but this application is not limited to the embodiments described. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application, and these equivalent modifications or substitutions are all included within the scope defined by the claims of this application.

Claims

1. An aerial manipulator-based delivery planning control method, characterized by, The method comprises the following steps: A mathematical model of the aerial manipulator is established; wherein the aerial manipulator comprises a quadrotor platform and a Delta manipulator mechanically connected thereto; A trajectory model of the aerial manipulator is established; According to the mathematical model and the trajectory model, an air-drop trajectory is planned and a predictive control instruction of hierarchical disturbance compensation is generated; According to the air-drop trajectory and the predictive control instruction, a release time of the load is evaluated online; According to the release time, the air-drop trajectory and the predictive control instruction, the control manipulator releases the load.

2. The aerial manipulator-based delivery planning control method according to claim 1, characterized by, The method comprises the following steps: The relationship between the quadrotor platform and the Delta manipulator is modeled as: wherein denotes position, denotes angular velocity, the skew symmetric matrix of which is used to express the vector cross product; The translational dynamics of the quadrotor platform is modeled as: wherein, is the total thrust, is the system mass, is the aircraft system is the direction of the shaft in the inertial frame, is the force generated by the interaction of the external disturbance with the Delta robot arm; The rotational kinematics and dynamics of the quadrotor platform are respectively modeled as: wherein, is the total moment, is the inertia matrix, is the additional moment applied by the Delta robot interaction and external disturbances.

3. The aerial manipulator-based delivery planning control method according to claim 1, characterized by, The method comprises the following steps: According to the position and velocity of the load at the moment of release the impact point of the load is determined ; calculating a time for the load to reach a target level ; wherein, represents the position of the load, represents the speed of the load; determining that the load meets the following relationship in time of the trajectory of the load. ； The landing point is determined as: 。 4. The aerial manipulator-based delivery planning control method according to claim 1, characterized by, The method comprises the following steps: A trajectory optimization problem is defined as: wherein the trajectory optimization problem includes minimizing control energy and includes a time regularization term, and includes corridor constraints, kinematic constraints, dynamic constraints, and landing point constraints; is a set of inequality constraints defined on , and represent initial and final states of the quadrotor platform and end effector, respectively. A penalty function is constructed to constrain the landing point error as: wherein , denotes a time interval, is a smoothing function of the drop penalty, is defined as follows: ； wherein, to control the smoothness of the relaxation function; for smoothly introducing integer variables into the trajectory optimization problem with non-linearities by defining a relaxation function; In solving the trajectory optimization problem, a time interval τ around the optimal release time tr is determined, and the same penalty is applied to the landing point within the time interval τ, and then the air-drop trajectory containing a series of feasible release times is generated .

5. The aerial manipulator-based delivery planning control method according to claim 1, characterized by, The method comprises the following steps: The state representation of the quadrotor platform is defined as , and the control input is ; Defining a time horizon is divided into equal intervals, the interval length being wherein is the horizon length; A constrained nonlinear optimization problem is obtained as: ； where the function represents the dynamic model of the quadrotor platform; the reference state and the input come from the aerial delivery trajectory, the reference state is the position, attitude and velocity of the UAV respectively, and the output is the control command of the UAV: total thrust and angular velocity; , and are the weighting matrices of the state, input and terminal state respectively; is the current state estimation when the optimal solution of the nonlinear optimization problem is solved; A nonlinear disturbance observer is used to compensate for external disturbance forces, and the following is calculated: wherein represents an acceleration measured by an inertial measurement unit; An observer matrix is constructed, and is differentiated and discretized to obtain: wherein, is passed through a Butterworth filter with a cutoff frequency of represents the inverse of the sensor sampling rate; is a parameter adjusted according to actual working conditions;​ by each angular velocity , a thrust force and a moment vector are obtained according to a dynamic model of the quadrotor platform: wherein, and are the dimensions of the quadrotor platform shown; and denote the thrust and moment coefficients, respectively; is the moment of inertia of the propeller. Desired angular acceleration The calculation is as follows: wherein represents the filtered feedback angular acceleration, which is derived from by differentiation; is the feedback angular moment based on each rotor angular velocity; The control thrust and control torque instructions are obtained as: wherein, is the current rotor angular velocity, is the motor dynamics time constant; and are both filtered by a second order Butterworth filter with the same cut-off frequency.

6. The aerial manipulator-based delivery planning control method according to claim 1, characterized by, The method comprises the following steps: Definition of reference drop trajectory on the reference release time instant corresponding optimal impact point; Determining actual optimal release time due to external disturbances , and satisfying ; defining each discrete state of the load within a prediction horizon at release, in turn generating a set of predicted landing positions corresponding to the current time : wherein and denotes the vertical direction of the predicted position and velocity; The corresponding landing error sequence is calculated: ； wherein, represents the landing error sequence between the predicted landing point and the target position; by continuously updating to minimize the landing error sequence.

7. The aerial manipulator-based delivery planning control method according to any one of claims 1 to 6, characterized by, The method comprises the following steps: According to the release time, the air-drop trajectory and the predictive control instruction, the control manipulator releases the load. The method comprises the following steps: An electromagnet controller is controlled according to the release time; wherein the electromagnet controller is used to control whether the end of the Delta manipulator releases the load; 8. An aerial manipulator-based delivery planning control device characterized by comprising: A Delta manipulator controller is controlled according to the air-drop trajectory; The motion of the quadrotor platform is controlled according to the predictive control instruction. The device comprises: A mathematical modeling unit is configured to establish a mathematical model of an aerial manipulator; wherein the aerial manipulator comprises a quadrotor platform and a Delta manipulator mechanically connected thereto; A trajectory modeling unit is configured to establish a trajectory model of the aerial manipulator; An instruction generation unit is configured to plan an air-drop trajectory and generate a predictive control instruction of hierarchical disturbance compensation according to the mathematical model and the trajectory model; A release time evaluation unit is configured to evaluate a release time of the load according to the air-drop trajectory and the predictive control instruction; A release control unit is configured to control the control manipulator to release the load according to the release time, the air-drop trajectory and the predictive control instruction.

9. An electronic device, comprising: The electronic device comprises a processor and a memory; The memory is configured to store a program; The processor executes the program to implement the method in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The storage medium stores a program, and the program is executed by a processor to implement the method in any one of claims 1 to 7.

Images