Self-adaptive curvature constraint-based autonomous obstacle avoidance smoothness optimization method and system

Abstract

The invention discloses a self-adaptive curvature constraint-based autonomous obstacle avoidance smoothness optimization method and system, and the method comprises the steps: firstly, employing the convex hull characteristics of a B spline to efficiently calculate the third-order and fourth-order derivative information of a trajectory, and enabling the third-order and fourth-order derivative information to be the high-order energy information of the trajectory; and then, curvature information is used as a dynamic regulation factor to be introduced into an autonomous obstacle avoidance algorithm smoothness optimization process, so that constraint conditions of different degrees are applied to different sections. Through the self-adaptive segmented optimization strategy, the unmanned aerial vehicle can generate a flight trajectory which is integrally smooth and meets dynamic constraints in a real-time trajectory planning environment, the flight efficiency, safety and reliability are improved, and the algorithm can effectively reduce violent acceleration change and mechanical structure vibration caused by trajectory abrupt change. The service life of the unmanned aerial vehicle flight platform is prolonged; and meanwhile, the energy consumption in the flight process is reduced, and the inspection duration and the operation range of the unmanned aerial vehicle are prolonged, so that the practicability and the economical efficiency of the autonomous inspection task of the unmanned aerial vehicle are remarkably enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle (UAV) control, and particularly to a method and system for optimizing the smoothness of autonomous obstacle avoidance based on adaptive curvature constraints. Background Art

[0002] In order to achieve autonomous obstacle avoidance flight of UAVs in power inspection scenarios, researchers have been committed to designing and optimizing safe trajectories that adapt to complex environmental constraints. During power inspection, the smoothness of the autonomous obstacle avoidance trajectory is particularly important. It can not only ensure the stability and controllability of UAV flight, but also effectively avoid drastic changes in attitude and abnormal fluctuations in acceleration caused by sudden changes in the trajectory, thereby reducing the risk of collision or loss of control during the inspection task. Therefore, in this embodiment, an attempt is made to further optimize the smoothness and feasibility of the trajectory by combining the geometric and energy characteristics of the trajectory to ensure the safety, stability, and efficiency of UAV inspection operations.

[0003] Therefore, a method for optimizing the smoothness of autonomous obstacle avoidance based on adaptive curvature constraints is needed. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a method for optimizing the smoothness of autonomous obstacle avoidance based on adaptive curvature constraints, which flexibly utilizes the curvature information of different sections of the trajectory and fully considers the high-order energy information of the trajectory on the premise of ensuring real-time performance.

[0005] To achieve the above purpose, the present invention provides the following technical solutions:

[0006] The method for optimizing the smoothness of autonomous obstacle avoidance based on adaptive curvature constraints provided by the present invention includes the following steps:

[0007] S1: Obtain sensor data including IMU data, binocular image data, and depth image data;

[0008] S2: Use the IMU data and binocular image data to obtain real-time UAV position and attitude information;

[0009] S3: Create an occupancy grid map representing environmental information based on the pose information and depth image data;

[0010] S4: Use an autonomous obstacle avoidance algorithm to calculate a reliable and efficient UAV flight trajectory;

[0011] S5: Control the UAV to achieve autonomous flight in a position and speed closed-loop manner according to the flight trajectory.

[0012] Furthermore, in step S2, the IMU data and binocular image data are used to obtain real-time UAV position and attitude information through the visual positioning and mapping method of VINS-Fusion.

[0013] Further, the flight trajectory obtained in step S4 is specifically carried out in the following manner:

[0014] A path that avoids obstacles in free space is obtained through the A* algorithm, a front-end path search method. Then, a flight trajectory that meets requirements such as smoothness, safety, and feasibility is obtained through back-end trajectory optimization methods such as trajectory parameterization, construction of an objective function, and addition of constraint conditions.

[0015] Further, the positioning and mapping, motion planning, and control modules are all implemented within the Robot Operating System under the Ubuntu system.

[0016] Further, the flight trajectory adopts a segmented optimization strategy, which is realized by constructing an adaptive curvature constraint model. The adaptive curvature constraint model uses B-splines to calculate the high-order information of the flight trajectory and introduces the curvature information as a dynamic adjustment factor into the smoothness optimization process of the autonomous obstacle avoidance algorithm, obtaining different degrees of constraint conditions for different sections.

[0017] Further, the high-order information is the third-order and / or fourth-order derivative information of the flight trajectory calculated using the convex hull property of B-splines.

[0018] Further, the high-order information and curvature information of the flight trajectory are calculated according to the following formula:

[0019]

[0020]

[0021]

[0022] Where Q i represents the position of control point i, V i , A i , J i and S i are the first-order, second-order, third-order, and fourth-order derivatives of control point i respectively, Δt is the time interval between control points; κ i is the curvature at control point i; N c is the number of control points; λ1 and λ2 are the weight terms of the third-order and fourth-order derivatives of the trajectory.

[0023] The autonomous obstacle avoidance smoothness optimization system based on adaptive curvature constraint provided by the present invention includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the above method is implemented.

[0024] The beneficial effects of the present invention are as follows:

[0025] The autonomous obstacle avoidance smoothness optimization method based on adaptive curvature constraint provided by the present invention can, first of all, efficiently calculate the third-order and fourth-order derivative information of the trajectory by using the convex hull property of B-spline, which are the high-order energy information of the trajectory. Then, the curvature information is introduced as a dynamic adjustment factor into the smoothness optimization process of the autonomous obstacle avoidance algorithm, so as to impose different degrees of constraint conditions in different sections. Through this adaptive segmented optimization strategy, the unmanned aerial vehicle (UAV) can generate an overall smooth flight trajectory that meets the dynamic constraints in the real-time trajectory planning environment, which not only improves the flight efficiency, but also enhances the safety and reliability of the UAV power inspection system. In addition, the algorithm can effectively reduce the drastic acceleration changes and mechanical structure vibrations caused by trajectory mutations, and extend the service life of the UAV flight platform; at the same time, it also reduces the energy consumption during flight, helps to extend the inspection duration and operation range of the UAV, thus significantly enhancing the practicability and economy of the UAV autonomous inspection task.

[0026] The flight trajectory generated by this method is significantly smoother, and can effectively reduce the attitude fluctuations and trajectory mutation phenomena of the UAV during the inspection task. At the same time, this method can also effectively reduce the energy consumption during the actual flight process, which helps to improve the endurance, operation safety and economy of the UAV in complex power inspection tasks.

[0027] Other advantages, objectives and features of the present invention will be described to some extent in the subsequent specification, and to some extent, will be obvious to those skilled in the art based on the study of the following text, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be achieved and obtained through the following specification. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to make the objectives, technical solutions and beneficial effects of the present invention clearer, the present invention provides the following drawings for illustration.

[0029] Figure 1 It is a flowchart of the autonomous obstacle avoidance smoothness optimization algorithm;

[0030] Figure 2 It is a comparison diagram of the trajectory shapes obtained by two algorithms;

[0031] Figure 3 It is a comparison diagram of the jerk obtained by two algorithms;

[0032] Figure 4 It is a comparison of the method of this embodiment with EGO;

[0033] Figure 5 It is a diagram of the change of trajectory energy with the smooth term coefficient and its mean value;

[0034] Figure 6 The trajectory energy versus the smoothness term coefficient, along with its median and standard deviation;

[0035] Figure 7 The complete flight trajectories output by different methods in the simulation experiment;

[0036] Figure 8 The flight trajectories of different methods in area ①;

[0037] Figure 9 The flight trajectories of different methods in area ②;

[0038] Figure 10 The comparison of the replanned trajectory energies;

[0039] Figure 11 The cross - aircraft equipment;

[0040] Figure 12 The power system of the unmanned aerial vehicle for autonomous obstacle avoidance;

[0041] Figure 13 The experimental scenario of real - flight;

[0042] Figure 14 The comparison of real - flight experimental effects;

[0043] Figure 15 The comparison of real - flight energies. Specific implementation manners

[0044] The following further illustrates the present invention in conjunction with the accompanying drawings and specific embodiments, so that those skilled in the art can better understand the present invention and be able to implement it, but the embodiments cited do not limit the present invention.

[0045] As Figure 1 shown, Figure 1 is the flowchart of the autonomous obstacle - avoidance smoothness optimization algorithm. The autonomous obstacle - avoidance smoothness optimization method based on adaptive curvature constraint provided in this embodiment includes the following steps:

[0046] S1: Obtain sensor data including IMU data, binocular image data, and depth image data;

[0047] S2: Utilize the IMU data and binocular image data to obtain the real - time position and attitude information of the unmanned aerial vehicle through the visual localization and mapping (Simultaneous Localization and Mapping, SLAM) method of VINS - Fusion.

[0048] S3: Create an occupancy grid map representing the environmental information according to the pose information and depth image data.

[0049] S4: Obtain the flight trajectory through the front-end path search method of the A* algorithm. The flight trajectory is a path that avoids obstacles in free space. A flight trajectory that meets requirements such as smoothness, safety, and feasibility is obtained through backend trajectory optimization methods such as trajectory parameterization, constructing the objective function, and adding constraint conditions.

[0050] S5: Control the UAV to achieve autonomous flight through the position and speed closed-loop method according to the flight trajectory.

[0051] The positioning and mapping, motion planning, and control modules of this embodiment are all implemented within the Robot Operating System (ROS) under the Ubuntu system.

[0052] This embodiment focuses on the autonomous obstacle avoidance algorithm module among them, and mainly focuses on depicting the smoothness requirements in the objective function. Improving the smoothness of the trajectory can effectively reduce the drastic attitude changes of the UAV during the power inspection flight, thereby improving the stability and controllability of the inspection operation. A smooth trajectory can significantly reduce the mutation of the control input, make the UAV execute trajectory tracking more smoothly during the inspection task, reduce the impact on the actuator, help extend the service life of the hardware device, and improve the overall reliability and safety of the UAV power inspection system.

[0053] The implementation process of the autonomous obstacle avoidance smoothness optimization method based on adaptive curvature constraints provided in this embodiment is described in detail below.

[0054] The flight trajectory in this embodiment adopts a segmented optimization strategy, which is realized by constructing an adaptive curvature constraint model. The adaptive curvature constraint model uses B-splines to calculate the high-order information of the flight trajectory, and introduces the curvature information as a dynamic adjustment factor into the smoothness optimization process of the autonomous obstacle avoidance algorithm to obtain different degrees of constraint conditions for different sections; the high-order information is the third-order and / or fourth-order derivative information of the flight trajectory calculated using the convex hull property of B-splines.

[0055] The construction of the high-order information and curvature information of the B-splines in this embodiment is carried out in the following manner:

[0056] When performing trajectory planning for a robot in a two-dimensional plane, the trajectory is usually expressed as a polynomial that changes with time:

[0057]

[0058] Among them, \(t\) is a time variable. Here, \(x(t)\) and \(y(t)\) represent two-dimensional coordinates respectively, and \(p(t)\) is the overall trajectory. Polynomial trajectories have the advantages of simple calculation and flexible expression. By controlling the derivatives of each order of the polynomial, the motion characteristics such as the position, speed, and acceleration of the trajectory can be precisely adjusted. In two-dimensional trajectory planning, the curvature is usually determined by the speed and acceleration of the trajectory. The specific expression is:

[0059]

[0060] Among them, and represent the first-order derivative of the position with respect to the position parameter, that is, the speeds in the two coordinate axis directions. and represent the second-order derivative of the position with respect to the position parameter, that is, the accelerations in the two coordinate axis directions. \(\kappa(t)\) is the curvature value of the trajectory at a certain moment. The larger the curvature value, the stronger the bending degree of the curve at that point; the smaller the curvature value, the smoother the curve.

[0061] An unmanned aerial vehicle (UAV) is a robotic unit that moves in three-dimensional space. Its six degrees of freedom of motion enable it to perform complex movements in three-dimensional space, including horizontal flight, vertical takeoff and landing, rotation, etc. Therefore, the trajectory of a UAV is usually represented by a more complex spline curve. In this embodiment, the trajectory of the UAV is represented by the B-spline method. B-spline is a curve representation form widely used in computer graphics.

[0062] An important property of the B-spline is that its \(k\)-th derivative is still a B-spline function. In the research on parameterizing and characterizing trajectories using B-spline curves, it is very convenient to calculate the high-order energy information of the trajectory using the control point information. The B-spline curve has good high-order continuous differentiability, which enables the derivatives such as the speed and acceleration of the trajectory to remain continuous and smooth under the B-spline representation, thus ensuring that the motion states of the trajectory at different time points can be smoothly transitioned, avoiding discontinuous or sharp changes. The high-order information such as the speed and acceleration of the control points can be expressed in the differential flat output space, and the specific form is as follows:

[0063]

[0064] Among them, \(Q\) i represents the position of control point \(i\), \(V\) i , \(A\) i , \(J\) i and \(S\) i are the first-order, second-order, third-order, and fourth-order derivatives of the position of the position control point respectively, and \(\Delta t\) is the time interval between control points;

[0065] At this time, the geometric curvature information of a UAV trajectory represented by a B-spline in three-dimensional space can be conveniently described, and its mathematical expression is as follows:

[0066]

[0067] where κ i is the curvature at control point i. V i and A i are both calculated by formula (3). In this way, the curvature at different control points of the B-spline can be calculated, which is convenient for imposing more stringent smoothness constraints on the trajectory part with large curvature later.

[0068] The autonomous obstacle avoidance smoothness piecewise optimization strategy in this embodiment is carried out in the following manner:

[0069] The trajectory energy in the flight trajectory is a key indicator for measuring the system performance and efficiency. It not only reflects the power demand and energy consumption of the aircraft during the mission execution, but also reflects the comprehensive performance of the system under various dynamic constraints such as speed, acceleration, and attitude adjustment.

[0070] Usually in the formulation of trajectory optimization, a cost function is given to represent the actual task requirements and system constraints. For example, the entire trajectory optimization problem can be formulated through a smooth term, a collision term, and a feasible term:

[0071]

[0072] where J is the overall cost of trajectory optimization, J s is the smooth term penalty, J c is the collision term penalty, J d is the feasible term penalty. Since soft constraints are imposed during trajectory optimization, the requirements for these three penalty terms may be different. Therefore, different coefficients can be used to adjust the proportional relationship of these three penalty terms:

[0073]

[0074] where λ s represents the smooth term proportionality coefficient; λ c represents the collision term proportionality coefficient; λ d represents the feasible term proportionality coefficient;

[0075] This embodiment mainly discusses the smooth term penalty among them. The smoothness penalty can be formulated as an integral form of the trajectory derivatives (such as velocity and acceleration), and usually it is expressed as the time integral of the trajectory derivatives:

[0076]

[0077] where p (i) (t) is the i-th derivative of the trajectory, and λ i is the corresponding order weight. The above function has a closed-form solution for the trajectory segment represented by the B-spline curve. However, calculating the trajectory smoothness using discretized time integration will undoubtedly result in high computational overhead, and the required computational resources increase polynomially with the improvement of the discretization resolution.

[0078] In this embodiment, without relying on time integration, the smoothness of the trajectory is penalized. The B-spline is used to represent the trajectory of the UAV. Thanks to the convex hull property of the B-spline, an attempt is made to minimize the control points corresponding to the higher-order derivatives of the B-spline trajectory, thereby reducing the change in the higher-order derivatives of the entire curve and thus improving the smoothness of the trajectory.

[0079] By minimizing the control points of the second and third derivatives of the B-spline trajectory, the purpose of reducing the energy of the entire trajectory is achieved. The formula for the smoothness term penalty function is as follows:

[0080]

[0081] where N c is the number of control points. This method minimizes the higher-order derivatives, ultimately resulting in a smooth entire trajectory. By using the higher-order information of the trajectory to maintain a smooth control output, the body jitter is reduced and the mechanical damage is decreased.

[0082] Therefore, this embodiment will utilize the third and fourth derivative information of the B-spline trajectory to improve the smoothness of the entire trajectory. The mathematical formula is expressed as:

[0083]

[0084] where S i represents the fourth derivative of the corresponding control point (i.e., the jerk-jerk acceleration), which is calculated by formula (3).

[0085] Formula (9) represents that this experiment introduces a higher-order constraint index in the objective function or constraints.

[0086] The fourth derivative of the trajectory is introduced as a penalty for the smoothness term during trajectory optimization. Because the fourth derivative of the trajectory can constrain the mutation of the trajectory jerk (the third derivative of the trajectory). A lower fourth derivative of the trajectory means a gentle change rate of the jerk, thereby reducing the impact and vibration generated by the mechanical structure and extending the service life of the executing mechanism.

[0087] The method of adaptive curvature constraint disclosed in this embodiment, within the framework of differential geometry, curvature is a quantity used to quantitatively describe the degree of "bending" of a curve or surface at a certain point. This embodiment hopes to impose a more stringent smoothness term penalty at points with a larger trajectory curvature. At points with a smaller curvature, less smoothness penalty is imposed, and at this time, more consideration is given to requirements such as trajectory collision.

[0088] Therefore, this embodiment will further characterize the three-dimensional trajectory followed by the drone, rather than simply considering the extreme cases of the curvature of the entire curve. And in combination with the curvature of the trajectory, more energy constraints on the fourth derivative of the trajectory are imposed in large turning parts, and more energy constraints on the third derivative of the trajectory are utilized in small turning or non-turning parts. It is characterized by the following formula:

[0089]

[0090] Among them, λ1 and λ2 are the weight terms of the third and fourth derivatives of the trajectory.

[0091] Too sharp turning of the trajectory or frequent violent acceleration and deceleration changes will increase the pressure and energy consumption of the actuator. The controller needs to output a large control torque or thrust under conditions of rapidly changing acceleration and high-order derivatives, which not only increases power consumption but may also shorten the service life of the device. And in practical engineering applications, the aircraft cannot change direction or accelerate arbitrarily quickly. The change in the thrust of the aircraft rotor needs to have a smooth transition.

[0092] Therefore, incorporating curvature and trajectory high-order derivative indicators into the smoothness constraint or cost function of trajectory optimization ultimately realizes the trajectory smoothness optimization of adaptive curvature.

[0093] The effectiveness of the high-order trajectory energy information in this embodiment: In the field of trajectory smoothness optimization, in order to obtain a smooth and efficient motion trajectory, it is usually necessary to constrain or optimize the high-order derivatives of the motion equation. Integrating the minimization of the second norm of the third derivative of the trajectory (Minimum Jerk) and the minimization of the second norm of the fourth derivative of the trajectory (Minimum Snap) are two commonly used polynomial trajectory optimization methods. There are differences in their objectives and complexities. The basic idea of Minimum Jerk is to minimize the third derivative of the trajectory. According to the differential flatness characteristics of the drone, this method can reduce the angular velocity. It saves energy by adjusting the thrust of the drone system. The basic idea of Minimum Snap is to minimize the fourth derivative of the trajectory. Compared with Minimum Jerk, Minimum Snap goes further, and it can reduce the angular acceleration. And it saves energy by adjusting the thrust and torque of the drone system.

[0094] Such as Figure 2As shown Figure 2 Figure 2-1 is a comparison chart of the trajectory shapes obtained by two algorithms. Among them Figure 2 (a) Minimum Jerk trajectory shape diagram; (b) Minimum Snap trajectory shape diagram Figure 2 Figure 2-2 is a comparison chart of the trajectory shapes solved by the Minimum Jerk and Minimum Snap methods in a two-dimensional space. From a visual perspective, the trajectory morphologies obtained by the two methods are relatively similar, and it is difficult to directly distinguish which trajectory is smoother. However, by further analyzing the position, velocity, acceleration, and jerk curves of the trajectories generated by the two methods and comparing the fluctuations of the derivatives of each order, obvious differences in the trajectory smoothness of the two methods can be observed

[0095] As Figure 3 shown Figure 3 Figure 2-3 is a comparison chart of the jerks obtained by two algorithms Figure 3 (a) Minimum Jerk jerk diagram; (b) Minimum Snap jerk diagram; Jerk is the third-order derivative information of the trajectory Figure 3 Figure 2-3 is the jerk diagram obtained by the two methods respectively. Different color curves represent different coordinate axes. It can be seen that there are obvious convex points on both the x-axis and y-axis of the jerk curve obtained by Minimum Jerk, and it is not smooth. This means that although the trajectory shape optimized by Minimum Jerk is relatively smooth, the jerk curve may still have a relatively abrupt transition in some sections. This may cause obvious attitude vibrations when the UAV executes flight tasks. The jerk curve obtained by Minimum Snap is continuous and smooth on both the x-axis and y-axis. Minimum Snap further suppresses the change rate of jerk by restricting the fourth-order derivative of the trajectory. The trajectory optimized by this method will greatly reduce the mutation and fluctuation of the jerk curve, and the change of the jerk curve in different time intervals is also more gentle. At this time, the acceleration change of the UAV system is gradual, without sudden severe fluctuations, thereby reducing the dynamic impact and system oscillation caused by uneven transitions

[0096] The following is the simulation experiment provided in this embodiment. The software and hardware equipment and experimental parameters can be referred to Table 1 below, which are the software and hardware equipment for the simulation experiment

[0097] Table 1 Experimental environment

[0098]

[0099] The key parameters in the simulation environment are shown in Table 2

[0100] Table 2 Flight parameters and environmental parameters

[0101]

[0102] As Figure 4 shown Figure 4 in the figure, the method of this embodiment is compared with the trajectory energy of EGO under formula (5), introducing curvature and the fourth derivative of the trajectory. In the research of trajectory optimization and flight control, trajectory energy is a key index used to quantify the energy consumed by an unmanned aerial vehicle (UAV) or a robot when performing a specific task. It refers to the total energy consumption required for the UAV to execute a certain path or task, which mainly includes the energy consumed by the thrust, torque, and related control inputs of the aircraft during the execution of the trajectory. Trajectory energy can generally be obtained by integrating the control inputs (acceleration, jerk, etc.).

[0103] The method provided in this embodiment can explore smoothness through trajectory energy, and numerically reflect the average result after the trajectory is replanned multiple times from the starting point to the ending point of the UAV. It can be seen that in terms of the performance of acceleration energy (integral of acceleration squared), the value of this method is 42.39, slightly better than 47.71 of EGO. In terms of the performance of jerk energy (integral of jerk squared), the difference between the two is more obvious. The integral of the second norm of jerk of EGO reaches 1874.29. The method in this paper maintains good energy characteristics under strong trajectory deformation force, and the integral of the second norm of jerk is reduced to 1117.90. Numerical experiments show that this method can optimize the energy output of the trajectory. Among them, the smooth term penalty of EGO uses the second and third derivative information of the trajectory respectively, and its mathematical form is formula (8). The smooth term penalty of this method uses the third and fourth derivative information of the trajectory, and its mathematical form is formula (9). Therefore, by introducing curvature and the fourth derivative of the trajectory, the trajectory energy consumption can be effectively reduced.

[0104] This embodiment analyzes different smoothness requirements: in the trajectory optimization problem, it is a very common practice to use the collision term, smoothness term, and feasibility term as penalty functions. These penalty terms can effectively handle the constraint problems related to the trajectory and help the optimization algorithm consider multiple objectives simultaneously. Formula (6) is the mathematical model representing the trajectory optimization problem in actual engineering. The difference between it and formula (5) lies in whether these penalty terms are comprehensively utilized by adding a proportionality coefficient, enabling the trajectory optimization method to more comprehensively consider various constraints and balance the contradiction between performance and constraints. The correct selection of the weighting parameter is crucial for robust replanning and should be adjusted in the simulation until satisfactory performance is obtained. The weighting parameter is related to the flight speed, obstacle density, etc.

[0105] Among them, different smooth term coefficients represent different smoothness requirements. However, a larger smooth term parameter will result in a shorter or smoother local trajectory, but with a smaller clearance from the obstacle. Although introducing the fourth derivative of the trajectory through curvature can reduce energy consumption, this method still has advantages under different smoothness requirements, that is, different smooth term coefficients.

[0106] As Figure 5 shown, Figure 5 Fig. is the variation of the trajectory energy with the smooth term coefficient and its mean value graph, Figure 5 where (a) is the result of the EGO method; (b) is the result of the method of this embodiment.

[0107] As Figure 6 shown, Figure 6 Fig. is the variation of the trajectory energy with the smooth term coefficient and its median and standard deviation, Figure 6 where (a) is the result of the EGO method; (b) is the result of the method of this embodiment. Modeled using formula (6), while ensuring that the planner can solve, continuously adjust the smooth term proportional coefficient to observe the change of the trajectory energy. Figure 5 and 6 are the experimental comparison results of the two methods. The abscissa of the line chart is the proportional coefficient of the smooth term, and the ordinate is the trajectory energy (the square integral of the third derivative of the trajectory).

[0108] First, from the overall trend, it can be seen that as the smooth term coefficient increases, the overall trajectory energy consumption decreases rapidly. During the period when the coefficient is from 0.1 to 1.0, the energy consumption decreases significantly; during the period when the coefficient is from 1.0 to 3.0, the trajectory energy decreases slowly. Figure 3 - 5 It reflects the mean value of the trajectory energy of the two methods under different coefficients. The red dashed line is the mean value of the trajectory energy. The mean value of the trajectory energy obtained by the EGO method is 1065.14. While the mean value of the trajectory energy of the method of this embodiment is 1052.52. On the one hand, the trajectory energies obtained by the two methods are both significantly less than Figure 4 the trajectory energy index in, which proves the necessity of parameter adjustment for robust trajectory planning. On the other hand, the method of this embodiment can still obtain overall better trajectory energy consumption under different smooth term coefficients.

[0109] Figure 6Reflect the fluctuations of the trajectory energy of the two methods under different coefficient conditions. The red dashed line is the median of the trajectory energy, and the green dashed line is the standard deviation of the trajectory energy. It can be seen from the line chart that in the range of the smoothing term coefficient from 0.6 to 2.5 for the EGO method, there are obvious fluctuations in the trajectory energy, and there are many peaks and valleys in the line chart. The trajectory energy fluctuation of the method is weaker in this range, there are only a few valleys in the line chart, and the overall shows a smooth downward trend. In the experimental results, the median and standard deviation of the trajectory energy of the original method are 974.37 and 335.76 respectively;

[0110] While the median and standard deviation of the trajectory energy of the method proposed in this embodiment are 947.56 and 327.05. These values indicate that the method of this embodiment, under different smoothing term coefficients (i.e., when facing different smoothness requirements), the generated trajectory energy not only has a lower median, meaning that the optimized trajectory energy consumption is less overall. Moreover, its standard deviation has also decreased, indicating that the volatility of the trajectory energy obtained in multiple experiments is smaller and has higher stability.

[0111] This embodiment provides visualization and comparative analysis of the flight results of the experiment: Next, the complete flight results in the simulation experiment will be visually displayed; The simulation experiment of this embodiment selects the Berlin noise map as the test environment, aiming to fully verify the extreme performance of the autonomous obstacle avoidance smoothness optimization algorithm in a complex and unfamiliar three-dimensional environment. This verification under extreme conditions provides a solid and reliable algorithm foundation for the drone to cope with complex flight environments in the power inspection task, ensuring the robustness and applicability of the algorithm in the power inspection scenario. And the smoothing term coefficient here uses the default parameter 1.0 of the EGO paper. The experimental results include the complete drone flight trajectory, the energy line chart of multiple replanned trajectories, and the effect comparison table of the trajectories.

[0112] As Figure 7 shown, Figure 7 is the complete flight trajectory output by different methods in the simulation experiment, Figure 7 showing the complete flight trajectory obtained in the simulation environment, which is a three-dimensional top view. Among them, blue represents the obstacles generated by the Berlin noise map. The green curve is the flight trajectory generated by EGO. The red curve is the trajectory generated by the method. The arrow indicates the flight direction of the drone from the starting point to the end point. Generally speaking, the trajectory obtained by the method of this embodiment is smoother. Because the overall convex part of the red curve is less, and it is closer to a natural straight line from the starting point to the end point, and it rarely bends significantly to the side. In contrast, the green curve has more prominent bending parts. The most obvious parts are the areas ① and ② enclosed by the dashed boxes.

[0113] As Figure 8 shown, Figure 8 is the flight trajectory of different methods in area ①,Figure 8 The (a) EGO method flight trajectory; (b) the flight trajectory of the method of this embodiment. As Figure 9 shown, Figure 9 are the flight trajectories of different methods in area ②, Figure 9 the (a) EGO method flight trajectory; (b) the flight trajectory of the method of this embodiment.

[0114] Figure 8 and Figure 9 detail the differences in the local flight conditions. It should be noted that the simulation experiment is carried out in three-dimensional space, and some obstacles in the visualization pictures block the trajectories. However, this is only the perspective effect of the pictures, and there is no collision between the flight trajectories and the obstacles in the simulation experiment. Figure 8 The left picture is the trajectory generated by the EGO method in Figure 7 area ①, and it can be seen that the trajectory has undergone a large deformation. The part protruding to the side of the trajectory is obvious, so the curvature is also large. Figure 8 The right picture is the trajectory generated by the method of this embodiment in Figure 7 area ②, and it can be seen that the deformation of the trajectory is smaller at this time. There is no obvious part protruding to the side of the trajectory, so it is smoother. A similar situation is as Figure 9 shown. The trajectory generated by the EGO method shows a relatively obvious bend in three-dimensional space. From the attitude information of the UAV, it can be observed that the attitude change of the UAV under the EGO method is relatively violent and not smooth enough, showing a large fluctuation. In contrast, when using the method of this embodiment under the condition of passing the same obstacles, the attitude change of the UAV is significantly reduced, and the change of the trajectory is highly consistent with the change of the attitude, presenting a more stable flight state. This shows that the method of this embodiment can provide a smoother and more stable flight trajectory while ensuring the autonomous obstacle avoidance of the UAV.

[0115] As Figure 10 shown, Figure 10 Energy comparison of the replanned trajectory, (a) the result of the EGO method; (b) the result of the method of this embodiment, Figure 10It is the trajectory energy diagram of a complete flight of the UAV. The abscissa is the number of trajectory replanning times, and the ordinate is the trajectory energy (the integral of the square of the third derivative of the trajectory). The red dashed line is the average value of the trajectory energy, and the green dashed line is the standard deviation of the trajectory energy. The straight-line distance from the starting point to the ending point is 28m, and the planner performs 20 to 22 replanning times during the entire flight process. Select the results of the first 20 replanning times to compare the changes in the trajectory energy of the two methods. From the overall structure, the changing trends of the trajectory energy of the two methods are the same because the areas where the UAV moves and the obstacles it avoids are the same. During the first two trajectory planning processes, since the UAV transitions from a stationary state to a moving state, obvious peaks in the trajectory energy appear. In the subsequent flight stage, as the UAV enters a stable flight state, the trajectory energy gradually shows a downward trend. However, when avoiding some obstacles, the trajectory energy experiences short-term fluctuations. When approaching the end point, the trajectory energy in both figures shows an obvious decrease. This indicates that when the task is approaching the end, the energy consumption of the UAV tends to be stable. However, there are differences in the numerical indicators. In terms of the average value of energy consumption, the average value of the trajectory energy of the method is 922.41, which is better than 952.45 of EGO, reducing the energy output by approximately 3%. In terms of the change in energy consumption, the standard deviation of the trajectory energy of the method is 813.38, which is also better than 871.63 of EGO. Moreover, the peak value of the trajectory energy of EGO is significantly greater than 3500, while the peak value of the trajectory energy of the method in this embodiment is less than 3500. For the red dashed line, that is, the average trajectory energy line, the trajectory energy obtained by the method during the stable flight process is also significantly lower than this energy average line.

[0116] Table 3 Comparison of Trajectory Effects

[0117]

[0118] Table 3 is a comparison table of the trajectory states obtained by different methods in the simulation flight. The differences in the trajectory length, speed, and acceleration between the two are small. In terms of the time consumed for the calculation of the smoothing term, although the method in this embodiment has an increase in complexity, the increased calculation time is only about 0.0005 milliseconds. This is completely acceptable for real-time trajectory planning tasks. Finally, in terms of energy consumption, the method shows obvious advantages. The optimized trajectory can not only effectively reduce the flight energy consumption of the UAV in the power inspection scenario but also maintain a flight performance similar to that of the EGO method. Overall, the method proposed in this embodiment can better control the energy consumption of the power inspection UAV while meeting the trajectory optimization accuracy, demonstrating more excellent comprehensive performance and contributing to improving the safety and economy of actual inspection operations.

[0119] As Figure 11 shown, Figure 11For the drone equipment, this embodiment also provides a real experiment, which uses the drone power system, including the remote control; aircraft, etc. In order to verify the effectiveness of the method of this embodiment, this embodiment builds a drone experimental platform with autonomous flight capabilities and conducts experiments in an indoor environment. Since this platform is a self-built system, it is quite different from mature commercial drones and is difficult to control. Therefore, it is recommended to first use Figure 11 The drone equipment shown in the figure is used for basic adaptive training such as ground bouncing to improve the mastery of flight control. After being fully familiar with the control characteristics of the drone, autonomous flight experiments can be gradually carried out to ensure the controllability and safety of the experiment. Figure 11 (a) is the LiteRadio 3 remote controller from FPV DreamWorks, and (b) is the manufacturer's customized version of the drone without image transmission.

[0120] like Figure 12 As shown, Figure 12 It is a UAV power system with autonomous obstacle avoidance. Figure 12 (a) front view; (b) side view, Figure 12 The flight platform shown in the figure is equipped with an Intel NUV11TNKi5 microcomputer, combined with the D435 camera sensor of the Intel RealSense series for visual perception. The flight control hardware uses Holybro Pixhawk 6c. The flight control software uses PX4 to ensure precise flight control. In addition, the platform is equipped with a TATTU 2300mAh 4S battery assembly to provide sufficient power support for flight. The workstation uses the author's laptop, and its computing power is consistent with Table 1.

[0121] The visual positioning solution provided by VINS-Fusion was used to provide accurate odometer information for the drone. Through the D435 depth camera, the system can obtain depth data of the surrounding environment. Based on this information, the autonomous obstacle avoidance algorithm calculates the flight trajectory and outputs it to the flight control system. Finally, the PX4 flight control system drives the drone to complete the entire flight experiment, ensuring the smooth progress of the indoor power inspection autonomous obstacle avoidance scenario. Figure 13 As shown, Figure 13 In order to create a real flight experiment scene, since the obstacles in the power inspection scene are usually towers, trees, transmission lines and other objects with obvious three-dimensional features and easy to cause collision, a set of Figure 13The experimental scenario shown in the figure is to simulate the actual flight environment of power inspection. The circular side bracket in the yellow dotted box is regarded as a three-dimensional obstacle in the power inspection task, so as to more realistically evaluate the reliability and performance of the autonomous obstacle avoidance smoothness optimization algorithm in the power inspection scenario. The drone is initially placed on the ground about 4m away from the obstacle. The drone will first hover at a height of 1m from the ground according to the command. After setting the end point, the drone will bypass the obstacle from the right side and finally arrive at the predetermined destination smoothly.

[0122] This experimental verification and comparative analysis, such as Figure 14 As shown, Figure 14 For the comparison of the actual flight experiment results, Figure 14 It is a rendering of a real flight. The green curve is the flight trajectory of EGO. The red one is the flight trajectory of the method of this embodiment. The colored grid is the occupancy grid map of the surrounding environment. Different colors represent different heights from the ground. The picture proves that both methods can achieve flight around obstacles. However, through comparative analysis, it can be seen that in the power inspection scene built indoors, the flight trajectory generated by the method of this embodiment is obviously smoother. Specifically, when the drone passes the same obstacle, the blue curve presents a smaller convex part than the green curve, and the curvature of the trajectory is significantly reduced, which is more conducive to the stable flight of the drone. On the contrary, the green curve has an obvious twisted part, showing a large curvature change, which is not conducive to the stability of the power inspection task.

[0123] like Figure 15 As shown, Figure 15 This is a comparison of real flight energy. It should be noted that there are certain differences in values ​​between the real flight experiment and the simulation results. There are two main reasons for this difference:

[0124] First, considering the safety of the drone, the speed and acceleration limits in actual flight are set to lower values, 1m / s and 1m / s respectively. 2 This is much lower than the setting in the simulation environment in Table 2.

[0125] Secondly, the simulation experiments were conducted on local computers, while the real flight experiments were conducted on microcomputers. Differences in hardware facilities may also have led to these differences.

[0126] In terms of the final energy value, the result obtained by the method of this embodiment is 1.348, slightly better than 1.408 of the original method, and the energy output is reduced by approximately 4%. This result indicates that in the power inspection scenario, the method proposed in this embodiment can effectively reduce the energy consumption during the actual flight of the UAV. Specifically, lower energy consumption means that the UAV can extend the endurance time when performing long-distance inspection tasks, increase the inspection range covered by a single flight operation, and reduce the number of mid-course returns for charging. In addition, lower energy consumption also means a reduced load on the UAV's battery and power system, which is beneficial to extending the service life of the equipment, reducing the operation and maintenance costs of inspection operations, and thus significantly improving the economy and practicality of power inspection tasks.

[0127] This method can be used for autonomous obstacle avoidance during power inspection. The importance of introducing high-order energy information is emphasized through comparative experiments between Minimum Jerk and Minimum Snap. In the simulation experiment, a set of comparative experiments first illustrate that this method can effectively reduce the trajectory energy consumption, and then it is verified that under different smoothness requirements, this method performs better in terms of trajectory energy. Finally, the complete process of the simulation experiment is visually displayed, and a detailed comparative analysis of the first 20 replanned trajectories is carried out. In the real flight experiment, first, the experimental equipment and experimental scenario are introduced, and the flight task is also described. Then, the visual effect of the real flight is shown, and a comparative analysis of indicators such as trajectory properties and trajectory energy is carried out, proving the authenticity and effectiveness of this method. The results of both the simulation experiment and the real experiment show that the flight trajectory generated by this method is significantly smoother, and can effectively reduce the attitude fluctuations and trajectory mutation phenomena of the UAV during the inspection task. At the same time, this method can also effectively reduce the energy consumption during the actual flight process, helping to improve the endurance, operation safety and economy of the UAV in complex power inspection tasks.

[0128] The above-described embodiments are only preferred embodiments given to fully illustrate the present invention, and the protection scope of the present invention is not limited thereto. Equivalent substitutions or transformations made by those skilled in the art on the basis of the present invention are all within the protection scope of the present invention. The protection scope of the present invention shall be subject to the claims.

Claims

1. An autonomous obstacle avoidance smoothness optimization method based on adaptive curvature constraints, characterized in that: It includes the following steps: S1: Obtain sensor data including IMU data, binocular image data, and depth image data; S2: Use the IMU data and binocular image data to obtain real-time UAV position and attitude information; S3: Create an occupancy grid map representing the environmental information based on the pose information and depth image data; S4: Use an autonomous obstacle avoidance algorithm to calculate a reliable and efficient UAV flight trajectory; S5: Control the UAV to achieve autonomous flight in a position and speed closed-loop manner according to the flight trajectory.

2. The autonomous obstacle avoidance smoothness optimization method based on adaptive curvature constraint according to claim 1, wherein: In step S2, the IMU data and binocular image data are used to obtain real-time UAV position and attitude information through the visual positioning and mapping method of VINS-Fusion.

3. The autonomous obstacle avoidance smoothness optimization method based on adaptive curvature constraint according to claim 1, characterized in that: The flight trajectory obtained in step S4 is specifically carried out in the following manner: A path that avoids obstacles in free space is obtained through the A* algorithm, a front-end path search method. Then, a flight trajectory that meets requirements such as smoothness, safety, and feasibility is obtained through backend trajectory optimization methods such as trajectory parameterization, construction of an objective function, and addition of constraint conditions.

4. The autonomous obstacle avoidance smoothness optimization method based on adaptive curvature constraint according to claim 1, wherein: The positioning and mapping, motion planning, and control modules are all implemented within the Robot Operating System under the Ubuntu system.

5. The autonomous obstacle avoidance smoothness optimization method based on adaptive curvature constraint according to claim 1, characterized in that: The flight trajectory adopts a segmented optimization strategy, which is achieved by constructing an adaptive curvature constraint model. The adaptive curvature constraint model uses B-splines to calculate the high-order information of the flight trajectory and introduces the curvature information as a dynamic adjustment factor into the smoothness optimization process of the autonomous obstacle avoidance algorithm to obtain different degrees of constraint conditions for different sections.

6. The method for optimizing the smoothness of autonomous obstacle avoidance based on adaptive curvature constraint according to claim 5, characterized in that: The high-order information is the third-order and / or fourth-order derivative information of the flight trajectory calculated using the convex hull property of B-splines.

7. The autonomous obstacle avoidance smoothness optimization method based on adaptive curvature constraint according to claim 6, wherein: The high-order information and curvature information of the flight trajectory are calculated according to the following formula: Among them, Q i represents the position of control point i, and V i , A i , J i and S i are the first, second, third, and fourth derivatives of control point i respectively, and Δt is the time interval between control points; κ i is the curvature at control point i; N c is the number of control points; λ1 and λ2 are the weight terms of the third and fourth derivatives of the trajectory.

8. An autonomous obstacle avoidance smoothness optimization system based on adaptive curvature constraints, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method described in any one of claims 1 to 7 above.

Images