A method for constructing a control system for an air-ground dual-modal robot with high-speed motion capability

Abstract

The present invention discloses a method for constructing an air-ground dual-modal robot control system with high-speed motion capability. By combining the strong motion performance of a multi-rotor UAV with the high energy efficiency of a ground wheeled robot, a stronger operating capability is achieved. In particular, the high-speed ground motion and rapid mode switching of a four-rotor UAV based on differential flat output control of a passive universal wheel are achieved, thereby extending the UAV's flight time while ensuring the UAV's high-speed motion capability.

Description

Technical Field

[0001] The present invention relates to the field of robotics technology, and in particular to a method for constructing a control system of an air-ground dual-mode robot with high-speed motion capability. Background Art

[0002] In recent years, multirotor aircraft have attracted great attention in various fields such as package delivery, aerial photography and exploration of unknown scenes due to their high maneuverability and hovering capabilities. [1] However, the most worrying limitation of multirotor applications is its poor power efficiency. [2] For example, the mobility of multirotors greatly benefits large-scale exploration and long-distance transportation, but it also poses a significant challenge to the aircraft's endurance. Furthermore, the power efficiency of multirotors drops sharply when faced with the large payloads required for most missions, further exacerbating the dilemma.

[0003] In contrast, other widely used unmanned aerial vehicles, such as unmanned ground vehicles (UGVs) [3] , generally enjoy satisfactory power efficiency - typical UGVs can operate for 1-3 hours, while multirotor aircraft have a typical endurance of 5-20 minutes [2] This is mainly because most of the energy of a multirotor is wasted on counteracting gravity, while a UGV needs to overcome mostly friction. However, kinetic energy limitations and lack of maneuverability also limit the application of UGVs. For example, when a rock blocks the road, a UGV may be forced to take a roundabout approach, while a multirotor can simply fly over it.

[0004] Therefore, it is an intuitive idea to combine multirotors with UGVs, while maintaining the powerful maneuverability of multirotors, taking advantage of the complementary advantages of UGVs' high power efficiency. In addition, this air-ground vehicle can be applied to confined scenarios where air movement is restricted, such as pipes, sewers, and tunnels, thereby expanding the application of multirotors to a wider range of fields. At the same time, an independent air-ground vehicle can complete challenging tasks such as exploration and rescue in large-scale environments, which usually require a collaborative robot system consisting of a multirotor and a UGV. Previous researchers have developed many configurations of air-ground robots, mainly based on drive wheels. [5]－[9] and passive wheel

[10] －

[15] Leg-based flying machines [16，17] It also shows great potential.

[0005] 1. Robot system design

[0006] From the perspective of the field of air-ground dual-mode UAVs, current designs can be divided into two categories: active wheels and passive wheels. Among these two types of designs, the design of active wheels mainly includes: a. Transforming between aerial robot configurations such as multi-rotors and ground robot configurations such as differential carts through robot deformation; b. Directly connecting flying devices such as multi-rotors with ground motion devices such as Ackerman carts. The design of passive wheels mainly includes: a. Adding a fixed shaft to the axis of the multi-rotor UAV and installing passive wheels on both sides of the shaft to form a drum-type UAV b. Adding one or more passive wheels to the bottom of robots such as multi-rotor UAVs to form an air-ground robot.

[0007] A. Aircraft based on driving wheels

[0008] A basic idea to achieve air-ground locomotion is to add drive wheels to the UAV. Tan et al. [5] A six-rotor aircraft is connected to a four-wheel drive system to make it easy to control both in the air and on the ground. Tanaka et al. [6] Adding two drive wheels on either side of the quadcopter allows the dynamics of ground locomotion to become a differential vehicle. [7]－[9] A similar drive wheel was used. Mintchev et al.

[19] A deformable robot is proposed that can fold its arms and travel on tracks on the ground.

[0009] This design gives the robot strong off-road capabilities, but results in slow mode switching. As a trade-off for better ground control, this design with additional actuators is relatively heavy, adding a significant burden to the robot during aerial movement, which may conflict with the original intention of saving energy.

[0010] B. Aircraft based on passive wheels

[0011] Researchers

[10] －

[14] Lighter passive wheels, cylindrical cages, or spherical shells are installed on drones. These vehicles are mainly driven by the horizontal component of thrust, which means they do not require additional actuators and have simpler mechanisms. Qin et al.

[15] A small passive wheel is installed on the bottom of the biocopter to minimize the additional equipment weight. However, these designs share a common disadvantage, namely the huge friction force to be counteracted while performing yaw angle control, which leads to poor control performance at low thrust. In addition, none of these works has developed a system that can follow high-speed (||v||>1.5m / s) trajectories during ground locomotion, which severely limits their application, in stark contrast to the active trajectory tracking capabilities of multirotors (||v||>5m / s). On the other hand, typical designs based on driving wheels or passive wheels usually have integral or non-integral dynamic constraints, such as differential robot models [9，11]－[14，18] Ackerman Model [5] The non-integral constraints of the yaw angle and velocity control lead to coupling. These constraints can severely limit its application in photography, exploration, and other scenarios where active yaw control is necessary for better sensor perception range.

[0012] 2. Control

[0013] From a motion control perspective, most previous work using unified dynamical systems

[10] －

[15] None of them proposed a unified controller for both motion modes. Instead, they tended to design two controllers separately, which usually resulted in slow mode switching. In addition, as mentioned in

[20] , the slow switching between stabilizing subsystems may lead to instability, which poses a challenge to the planning and control of the aircraft. However, unlike typical aerospace vehicles, aerospace vehicles enjoy the advantage of differential flatness, which provides convenience for control and planning.

[21] , while ground-based aircraft are affected by support forces and friction forces. Therefore, a unified control scheme based on differential flatness that takes support forces and friction forces into account for high-speed trajectory tracking is urgently needed.

[0014] References:

[0015] [1] Quan Quan. Introduction to multicopter design and control. Springer, 2017.

[0016] [2]Lun Quan,Luxin Han,Boyu Zhou,Shaojie Shen,and Fei Gao.Survey ofuav motion planning.IET Cyber-systems and Robotics,2(1):14–21,2020.[3]JoYung Wong.Theory of ground vehicles.John Wiley&Sons,2022.

[0017] [4]Jeffrey Delmerico,Elias Mueggler,Julia Nitsch,and DavideScaramuzza.Active autonomous aerial exploration for ground robot pathplanning.IEEE Robotics and Automation Letters,2(2):664–671,2017.

[0018] [5]Qifan Tan,Xinyu Zhang,Huaping Liu,Shuyuan Jiao,Mo Zhou,and JunLi.Multimodal dynamics analysis and control for amphibious flydrivevehicle.IEEE / ASME Transactions on Mechatronics,26(2):621–632,2021.

[0019] [6]Katsuaki Tanaka,Di Zhang,Sho Inoue,Ritaro Kasai,Hiroya Yokoyama,Koki Shindo,Ko Matsuhiro,Shigeaki Marumoto,Hiroyuki Ishii,and AtsuoTakanishi.A design of a small mobile robot with a hybrid locomotion mechanismof wheels and multi-rotors.In 2017IEEE International Conference onMechatronics and Automation(ICMA),pages 1503–1508.IEEE,2017.

[0020] [7]Hyungho Chris Choi,Inhwan Wee,Micah Corah,Sahand Sabet,TaeyeonKim,Thomas Touma,David Hyunchul Shim,and Ali-akbar Agha-mohammadi.Baxter:Bi-modal aerial-terrestrial hybrid vehicle for long-endurance versatilemobility.In International Symposium on Experimental Robotics,pages 60–72.Springer,2020.

[0021] [8]Arash Kalantari,Thomas Touma,Leon Kim,Rianna Jitosho,KyleStrickland,Brett T Lopez,and Ali-Akbar Agha-Mohammadi.Drivocopter:A concepthybrid aerial / ground vehicle for long-endurance mobility.In 2020IEEEAerospace Conference,pages 1–10.IEEE,2020.

[0022] [9]Nitzan Ben David and David Zarrouk.Design and analysis of fcstar,ahybrid flying and climbing sprawl tuned robot.IEEE Robotics and AutomationLetters,2021.

[0023]

[10] Arash Kalantari and Matthew Spenko.Design and experimentalvalidation of hytaq,a hybrid terrestrial and aerial quadrotor.In 2013IEEEInternational Conference on Robotics and Automation(ICRA),pages 4445–4450.IEEE,2013.

[0024]

[11] Ruibin Zhang,Yuze Wu,Lixian Zhang,Chao Xu,and Fei Gao.Autonomousand adaptive navigation for terrestrial-aerial bimodal vehicles.IEEERobotics and Automation Letters,7(2):3008–3015,2022.

[0025]

[12] J Colmenares-Vazquez,P Castillo,Nicolas Marchand,and D Huerta-′Garc′la.Nonlinear control for ground-air trajectory tracking by a hybridvehicle:theory and experiments.IFAC-PapersOnLine,52(8):19–24,2019.

[0026]

[13] Yoshiro Hada,Manabu Nakao,Moyuru Yamada,Hiroki Kobayashi,NaoyukiSawasaki,Katsunori Yokoji,Satoshi Kanai,Fumiki Tanaka,Hiroaki Date,SarthakPathak,et al.Development of a bridge inspection support system using two-wheeled multicopter and 3d modeling technology.Journal of Disaster Research,12(3):593–606,2017.

[0027]

[14] Jianan Yang,Yimin Zhu,Lixian Zhang,Yifei Dong,and YihangDing.Sytab:A class of smooth-transition hybrid terrestrial / aerialbicopters.IEEE Robotics and Automation Letters,7(4):9199–9206,2022.

[0028]

[15] Youming Qin,Yihang Li,Xu Wei,and Fu Zhang.Hybrid aerialgroundlocomotion with a single passive wheel.In 2020IEEE / RSJ InternationalConference on Intelligent Robots and Systems(IROS),pages 1371–1376.IEEE,2020.

[0029]

[16] Kailin Li,Baoling Han,Yuting Zhao,and Chen Zhu.Motion planningand simulation of combined land-air amphibious robot.In IOP ConferenceSeries:Materials Science and Engineering,volume 428,page 012057.IOPPublishing,2018.

[0030]

[17] Yash Mulgaonkar,Brandon Araki,Je-sung Koh,Luis GuerreroBonilla,Daniel M Aukes,Anurag Makineni,Michael T Tolley,Daniela Rus,Robert J Wood,andVijay Kumar.The flying monkey:a mesoscale robot that can run,fly,and grasp.In2016IEEE International Conference on Robotics and Automation(ICRA),pages4672–4679.IEEE,2016.

[0031]

[18] Scott Morton and Nikolaos Papanikolopoulos.A small hybridgroundair vehicle concept.In 2017IEEE / RSJ International Conference onIntelligent Robots and Systems(IROS),pages 5149–5154.IEEE,2017.

[0032]

[19] Stefano Mintchev and Dario Floreano.A multi-modal hovering andterrestrial robot with adaptive morphology.In Proceedings of the 2ndInternational Symposium on Aerial Robotics,number CONF,2018.

[0033]

[20] Daniel Liberzon.Switching in systems and control,volume190.Springer,2003.

[0034]

[21] Daniel Mellinger and Vijay Kumar.Minimum snap trajectorygeneration and control for quadrotors.In 2011IEEE International Conference onRobotics and Automation(ICRA),pages 2520–2525.IEEE,2011.

[0035]

[22] Zhepei Wang,Chao Xu,and Fei Gao.Robust trajectory planning forspatial-temporal multi-drone coordination in large scenes.arXiv preprintarXiv:2109.08403,2021.

[0036]

[23] B.Zhou,F.Gao,L.Wang,C.Liu,and S.Shen.Robust and efficientquadrotor trajectory generation for fast autonomous flight.IEEE Robotics andAutomation Letters,4(4):3529–3536,2019.

[0037]

[24] Xin Zhou,Zhepei Wang,Hongkai Ye,Chao Xu,and Fei Gao.Egoplanner:Anesdf-free gradient-based local planner for quadrotors.IEEE Robotics andAutomation Letters,6(2):478–485,2021.

[0038]

[25] Constantin Paleologu, Jacob Benesty, and Silviu Ciochina. A robust variable forgetting factor recursive least-squares algorithm for system identification. IEEE Signal Processing Letters, 15: 597–600, 2008.

[0039]

[26] Zhepei Wang, Xin Zhou, Chao Xu, and Fei Gao. Geometrically constrained trajectory optimization for multicopters. IEEE Transactions on Robotics, 2022. Summary of the Invention

[0040] In view of the shortcomings of the existing technology, the present invention aims to provide a method for constructing an air-ground dual-modal robot control system with high-speed motion capability.

[0041] In order to achieve the above object, the present invention adopts the following technical solutions:

[0042] A method for constructing a control system for an air-ground dual-mode robot capable of high-speed motion, wherein the robot comprises a multi-rotor drone and wheels that can passively rotate freely along a mounting axis; the specific process of the construction method is as follows:

[0043] S1. Constructing a dynamic model:

[0044] The body coordinate system is (x b ,y b , z b ) and the F-L-U world coordinate system is (x w ,y w , z w ); When the robot is hovering in the air, it only needs to increase the support force F S By setting it to zero, you can get the dynamics of the robot;

[0045] First, assume that the wheel radius, deflection, and air resistance are negligible and that the robot moves on a flat surface. Consider the robot's state x = {r, R}, where r is the robot's center of mass in the world coordinate system and R is the rotation from the world coordinate system to the body coordinate system. The input is u = {f, τ}, where f is the total thrust and τ is the torque generated by the thrust. This gives the dynamic model based on the Newton-Euler equations:

[0046]

[0047]

[0048] In formula (1), m is the total mass of the robot, g is the acceleration due to gravity, F S is the supporting force, R φ is the rotation matrix, which is composed of the horizontal velocity Angle between rotation, friction;

[0049] In formula (2), M is the inertia matrix, ω is the angular rate in the fuselage coordinate system, and l is the length between the center of mass and the wheel center. According to the friction law, F f =F S μ, where μ is the coefficient of rolling friction;

[0050] S2, differential flat output considering friction:

[0051] The options for planar output are:

[0052]

[0053] Among them, x [s] is a stack of finite derivatives ψ is the yaw angle, and there is an additional term The specific content of flatness transformation is given as follows:

[0054] (x,u)=Ψ(ξ) (4)

[0055] First, multiply equation (1) by the body axis and

[0056]

[0057] in:

[0058]

[0059]

[0060]

[0061] make:

[0062]

[0063] There is x b ⊥k and y b ⊥k, so zb / / k; When the system is in steady state, we can get k=(gF S / m)e3, which means z b is in the same direction as k; therefore:

[0064]

[0065] in z b Multiplying formula 1 on the left yields:

[0066]

[0067] Next, use the Hough transform to decompose the yaw quaternion q ψ and the tilt quaternion q z :

[0068] q φ =((cos(ψ / 2),0,0,sin(ψ / 2)) T (9)

[0069] Because q z represents the tilt transformation, so in q z There is no z component in ; let q z =(w q , x q ,y q ,0) T , and q z By solving the equation q z e3=z b get:

[0070]

[0071] The rotation matrix is defined as:

[0072]

[0073] Where R is the transformation from quaternion to rotation matrix, according to Can get Right now

[0074]

[0075] Through the above formula, we can get:

[0076]

[0077] where sψ represents sin(ψ), cψ represents cos(ψ), and:

[0078]

[0079] in

[0080]

[0081] In addition, you can also get

[0082] S3. Minimum support force trajectory generation:

[0083] As a dimension of the differential flatness output, F S The trajectory and have a certain degree of independence; therefore, an optimization problem is formulated - minimum collective thrust to minimize energy consumption, which allows the support force F S The planning of the position and yaw angle is separated from the planning of the

[0084] 3.1) Optimization problem modeling: The objective function is the collective thrust norm with three linear inequality constraints and one nonlinear equation constraint; F S Set to a predetermined constant F Spre On the other hand, the support force must be actively committed, otherwise the robot will not stay on the ground; Note that during ground locomotion, the robot's tilt angle θ (q z ) is restricted due to its structural limitations, otherwise it would touch the ground; therefore, considering the dynamic model of the robot, an optimization problem is designed to solve the minimum support force trajectory problem:

[0085]

[0086]

[0087] stEd≤D (15)

[0088] in:

[0089]

[0090]

[0091]

[0092] D=Fd0,d0=[θ max ,F Spre ,0] T ,

[0093] d=[θ(q z ),F S ] T ,

[0094]

[0095] 3.2) A feasible solution: Due to another flatness output is generated by the planner, and an F with variables can be designed. S Function: Horizontal acceleration and the tilt angle θ(q z ):

[0096]

[0097] F S =F Spre : when ||a h When || is low, due to the inverted pendulum structure of the robot, there should be a large enough torque to stabilize the posture; in this case, F S Set to a constant value F Spre , in order to avoid the vibration caused by frequent thrust changes; it can be inferred that the minimum torque required for the robot to maintain the tilt angle θ(q z ) stable equilibrium:

[0098]

[0099] For the maximum inclination angle θ max F aircraft smaller than π / 4 S As θ increases, it decreases, so:

[0100]

[0101] when ||a h || is greater than the maximum tilt angle θ of the aircraft max and F Spre When the acceleration is lower than what it can provide, the robot must reduce the support force to provide more thrust;

[0102] F S =0: The maximum acceleration that the robot can perform during ground motion is a lim , because its F S Should always be positive; if the robot is required to reach ||a h ||>a lim The ideal state is to S Set to zero and limit ||a h || is a lim In this case, the robot cannot catch up with the trajectory, which can be avoided by limiting the acceleration of the ground motion in the planning stage; at the same time, F can be easily obtained using the chain rule S The derivative of

[0103] S4. Unified controller design:

[0104] The only difference between air sports and ground sports is F S Is zero, so a unified controller can be applied to both motions; the controller is a cascade position-velocity controller; first, take the desired state from the trajectory and will It is added to the proportional error of the velocity controller; the required support force is then calculated according to formula (16); then, by applying the flatness transformation formula (4), the collective thrust f and the command {R, ω, τ} are obtained, which are then converted by the flight controller into the thrust f of each motor; finally, the proportional position error is added back to the velocity controller;

[0105] S5. Online identification of thrust coefficient

[0106] The thrust signal required by the flight controller is usually a normalized number Γ∈[0,1], so a thrust coefficient k is required f To transform f into Γ:

[0107]

[0108] k f can be easily measured by pre-calibration, but it is actually a variable that depends on the battery voltage, air density, propeller integrity and other external factors; therefore, a forgetting factor recursive least squares algorithm is used to identify k online. f : The basic form is:

[0109]

[0110] where a k and b k is the observed value, x k is the target to be updated, λ is the forgetting factor, which is usually set between [0.95, 1] and Between; Based on the model:

[0111]

[0112] in is the estimated horizontal acceleration in the world coordinate system, F h is in x W－ y W Normalized thrust projected on the plane, choose k f is x k , for b k , F h for a k .

[0113] Furthermore, a quad-rotor drone is used as the multi-rotor drone part; the quad-rotor drone adopts a frame structure made of carbon fiber plates; the frame structure is provided with four motors, an electronic speed controller, four propellers, a flight controller, PX4 firmware and a battery; and a propeller protector is installed at the bottom of each motor.

[0114] The beneficial effects of the present invention are: by combining the strong motion performance of a multi-rotor UAV with the high energy efficiency of a ground wheeled robot to achieve stronger operating capabilities, especially combining the high-speed ground motion and fast mode switching of a four-rotor UAV with a passive universal wheel based on differential flat output control, the purpose of extending the UAV's flight time while ensuring the UAV's high-speed motion capability is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0115] Figure 1 This is a schematic diagram of the robot structure of Example 1 of the present invention;

[0116] Figure 2 is the dynamic model of the robot in Example 1 of the present invention;

[0117] Figure 3 Schematic diagram of the framework of the controller in Example 2 of the present invention;

[0118] Figure 4 This is a diagram of a hybrid trajectory tracking experiment in Example 3 of the present invention;

[0119] Figure 5 This is a graph of hybrid trajectory tracking experimental data in Example 3 of the present invention;

[0120] Figure 6 This is a graph of ground trajectory tracking experimental data in Example 3 of the present invention;

[0121] Figure 7 This is a diagram illustrating a yaw experiment in Example 3 of the present invention;

[0122] Figure 8 This is a diagram of yaw test data in Example 3 of the present invention;

[0123] Figure 9 This is a schematic diagram of the comparison results in Example 3 of the present invention. DETAILED DESCRIPTION

[0124] The present invention will be further described below in conjunction with the accompanying drawings. It should be noted that this embodiment is based on the technical solution and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to this embodiment.

[0125] Example 1

[0126] This embodiment provides an air-ground dual-mode robot with high-speed motion capability, such as Figure 1 As shown, it includes a general multi-rotor drone and wheels that can passively rotate freely along the mounting axis;

[0127] In this embodiment, a quad-rotor drone is used as the multi-rotor drone part to simplify modeling and control. Specifically, the quad-rotor drone adopts a frame structure 5 made of carbon fiber plates with a wheelbase of 250 mm. The frame structure 5 is provided with four motors 1 (T-engine F60KV2550 brushless motor), an electronic speed controller, four propellers 2 (Gemfan51477 propellers), a flight controller 3 (Holybro Pixhawk4 mini flight controller), PX4 firmware and a battery 4 (ACE4S2300mAh lithium battery). Due to the powerful propulsion system and lightweight structure, the robot of this embodiment can carry a payload of up to 1.5 kg. In order to prevent the propeller from hitting the ground when starting and stopping, in this embodiment, a propeller protector 6 is installed at the bottom of each motor 1.

[0128] As for the wheels, for better mobility, this embodiment installs an omnidirectional roller 7 on the bottom of the quadcopter drone, which can be easily found in the suitcase. The wheel weighs 105 grams and adds little burden to the robot during aerial movement, but provides a fairly simple way for the robot to move freely on the ground. The onboard computer 8 is an NVIDIA Xavier NX, installed on a miniature customized carrier board. The onboard computer 8 communicates with the FCU (flight control unit) via MAVROS, transmitting IMU (inertial measurement unit) and control command data.

[0129] Example 2

[0130] This embodiment provides a method for constructing a control system of the robot described in Example 1. The specific process is as follows:

[0131] 1. Dynamic model

[0132] First introduce two coordinate systems: body coordinate system (x b ,y b , z b ) and the F-L-U world coordinate system (x w ,y w , z w ). When the robot is hovering in the air, it only needs to increase the support force F S By setting it to zero, we can obtain the dynamics of the robot, so here we focus on the dynamics of ground motion, such as Figure 2 shown.

[0133] First, assume that the wheel radius, deflection, and air resistance are negligible and that the robot is moving on a flat surface. Consider the robot's state x = {r, R}, where r is the position of the robot's center of mass in the world coordinate system and R is the rotation from the world coordinate system to the body coordinate system. The input is u = {f, τ}, where f is the total thrust and τ is the torque generated by the thrust. This gives us a dynamic model based on the Newton-Euler equations:

[0134]

[0135]

[0136] In formula (1), is the vector representation of r, m is the total mass of the robot, g is the acceleration due to gravity, F S It's the supporting force. is the rotation matrix, which is composed of the horizontal velocity Angle between Rotation, F f is the friction force;

[0137] In formula (2), M is the inertia matrix, ω is the angular rate in the fuselage coordinate system, and l is the length between the center of mass and the wheel center. According to the friction law, F f =F S μ, where μ is the coefficient of rolling friction.

[0138] 2. Differential flat output considering friction

[0139] In this section, we prove that the robot dynamics for input u are differentially flat, taking into account friction, according to the inference in

[22] . The choice of flat output for this example is:

[0140]

[0141] Among them, x [s] is a stack of finite derivatives x is used to represent a variable, and ψ is the yaw angle. Compared with the typical selection of multi-rotor plane output, the selection of this embodiment has an additional term The specific content of flatness transformation is given as follows:

[0142] (x,u)=Ψ(ξ) (4)

[0143] First, multiply equation (1) by the body axis and e2=(0,1,0)

[0144]

[0145] in:

[0146]

[0147]

[0148]

[0149] make:

[0150]

[0151] There is x b ⊥k and y b ⊥k, so z b / / k. When the system is in steady state, we can get k=(gF S / m)e3, which means z b The direction is the same as k. Therefore:

[0152]

[0153] in z b Multiplying formula 1 on the left yields:

[0154]

[0155] Next, use the Hough transform to decompose the yaw quaternion q φ Tilt quaternion q z :

[0156] q φ =((cos(ψ / 2),0,0,sin(ψ / 2)) T (9)

[0157] Because q z represents the tilt transformation, so in q z There is no z component in ; let q z =(w q , x q ,y q ,0) T , and q z By solving the equation q z e3=z b get:

[0158]

[0159] The rotation matrix is defined as:

[0160]

[0161] Where R is the transformation from quaternion to rotation matrix, according to Can get Right now

[0162]

[0163] Through the above formula, we can get:

[0164]

[0165] where sψ represents sin(ψ), cψ represents cos(ψ), and:

[0166]

[0167] in

[0168]

[0169] In addition, you can also get meaning.

[0170] 3. Minimum support force trajectory generation:

[0171] As a dimension of the differential flatness output, F S The trajectory and Therefore, an optimization problem can be formulated - minimum collective thrust to minimize energy consumption, which allows the support force F to be S The planning of the position and yaw angle is separated from the planning of the This provides the convenience that the system can adopt most planners designed for four rotors to be combined with a unified controller.

[0172] 3.1) Optimization problem modeling: The objective function is the collective thrust norm with three linear inequality constraints and one nonlinear equation constraint. Initially, there should be a torque large enough to stabilize the attitude. So F S Set to a predetermined constant F Spre On the other hand, the support force must be actively committed, otherwise the robot will not stay on the ground. Note that during ground motion, the robot's tilt angle θ (q z ) is restricted due to its structural limitations, otherwise it would touch the ground. Therefore, considering the dynamic model of the robot, an optimization problem can be designed to solve the minimum support force trajectory problem.

[0173]

[0174]

[0175] stEd≤D (15)

[0176] in:

[0177]

[0178]

[0179]

[0180] D=Fd0,d0=[θ max ,F Spre ,0] T ,

[0181] d=[θ(q z ),F S ] T ,

[0182]

[0183] 3.2) A feasible solution: No numerical calculations are used to solve F in real time S , nor does it use F S Instead, this embodiment uses a well-performing feasible solution. Due to the other flatness output is generated by the planner, and an F with variables can be designed. S Function: Horizontal acceleration and the tilt angle θ(q z ).

[0184]

[0185] a upper =tan(θ max )g

[0186] a upper =tan(θ max )(gF Spre / m).

[0187] F S =F Spre : when ||a h When || is low, due to the inverted pendulum structure of the robot, there should be a large enough torque to stabilize the posture. In this case, F S Set to a constant value F Spre , in order to avoid the vibration caused by frequent thrust changes. It can be inferred that the minimum torque required for the robot to maintain the tilt angle θ(q z ) stable equilibrium:

[0188]

[0189] For the maximum inclination angle θ max F aircraft smaller than π / 4 S As θ increases, it decreases, so:

[0190]

[0191] when ||a h || is greater than the maximum tilt angle θ of the aircraft max and F Spre When the acceleration is lower than what it can provide, the robot must reduce its support force to provide more thrust.

[0192] F S =0: The maximum acceleration that the robot can perform during ground motion is a lim , because its F S Should always be positive. If the robot is required to reach ||a h ||>a lim The ideal state is to S Set to zero and limit ||a h || is a lim In this case, the robot cannot catch up with the trajectory, which can be avoided by simply limiting the acceleration of the ground motion during the planning stage. At the same time, F can be easily obtained using the chain rule. S The derivative of .

[0193] 4. Unified controller design

[0194] The only difference between air sports and ground sports is F S Is zero, so a unified controller can be applied to both motions. The controller framework is as follows Figure 3 As shown, it is mainly a cascade position-speed controller. First, take the desired state from the trajectory and will It is added to the proportional error of the velocity controller. The required support force is then calculated according to Equation (16). By applying the flatness transformation Equation (4), the collective thrust f and the command {R, ω, τ} are obtained, which are then converted by the flight controller into thrust f for each motor. Finally, the proportional position error is added back to the velocity controller.

[0195] 5. Online identification of thrust coefficient

[0196] The combined thrust f is calculated above. However, in practical applications, the thrust signal required by the flight controller is usually a normalized number Γ∈[0,1], so a thrust coefficient k is required. f To transform f into Γ:

[0197]

[0198] k f It can be easily measured by pre-calibration, but it is actually a variable that depends on the battery voltage, air density, propeller integrity and other external factors. Therefore, a recursive least squares algorithm with the forgetting factor is used.

[24] To identify k online f The basic form of the algorithm is:

[0199]

[0200] where a k and b k is the observed value, x k is the target to be updated, λ is the forgetting factor, which is usually set between [0.95, 1] and Based on the model:

[0201]

[0202] in is the estimated horizontal acceleration in the world coordinate system, F h is in x W－ y W Normalized thrust projected on the plane, choose k f is x k , for b k , F h for a k The experiments show that the algorithm has good convergence.

[0203] Example 3

[0204] This embodiment is intended to verify the performance of the control system constructed in Example 2.

[0205] A. Hybrid trajectory tracking experiment: In this experiment, the robot was asked to perform an air-ground hybrid trajectory, in which the robot performed two air-to-ground and ground-to-ground motion mode switches. The maximum speed was 5.0 m / s, the average speed was 3.1 m / s, and the maximum acceleration was 4.2 m / s. 2 , the average acceleration is 2.9m / s 2 .

[0206] B. The results are as follows Figure 4 and Figure 5 As shown in the figure, the robot can smoothly switch between motion modes without any transition time. The RMSE of the three-dimensional motion and the z-axis are 0.129m and 0.043m respectively, demonstrating the excellent hybrid trajectory tracking capability of the control system.

[0207] C. Ground track tracking experiment: In this experiment, the robot is asked to perform an eight-shaped trajectory on the ground. The maximum speed is 4.5m / s, the average speed is 3.3m / s, and the maximum acceleration is 4.3m / s. 2 , the average acceleration is 3.0m / s 2 The results are as follows Figure 6 As shown, the RMSE is 0.080m.

[0208] D. Yaw-decoupling experimental verification: Due to the omnidirectional wheel design, the robot's yaw control is separated from the pitch and roll control. In this experiment, let the robot perform the same r [s1] The eight-shaped trajectory, while executing different ψ [s2] The trajectory, such as Figure 7 As shown,

[0209] The calculation results are as follows Figure 8 As shown. The RMSE of ψ=0 is 0.092m, The RMSE is 0.128m, The RMSE is 0.125m. The results show that the control system has good yaw tracking capability.

[0210] E. Endurance experiment: Let the robot perform two '8' trajectories, the only difference between which is the height (r(3) = 0, 1m), and record the total endurance time. The air-ground movement lasts 482 seconds, while the ground movement lasts 1626 seconds. Through simple data analysis, we know that the air movement power P a 254W, ground motion power P g In addition, the robot's standby power P s , including onboard computers, flight controllers and other components, are all 15W. So the corrected energy efficiency is:

[0211]

[0212] The proposed system is compared with other representative works with different configurations (i.e., [2, 5, 6, 9]). Taking into account the motion capabilities of the mobile robot, the following five performances are mainly benchmarked:

[0213] 1) Movement capability: This is assessed by the fastest hybrid trajectory the robot can follow, the faster the speed, the better the performance.

[0214] 2) Switching speed: This is evaluated by the average time of modal switching. The shorter the switching time, the better the performance.

[0215] 3) Kinematic constraints: This is evaluated by the kinematic constraints of the robot's ground motion. The fewer the constraints and the simpler the form, the better the performance.

[0216] 4) Structural simplicity: This is evaluated by the structural simplicity coefficient ξ. The larger the simplicity coefficient, the better the performance.

[0217]

[0218] The extra mass is the mass of the extra components moving on the ground, and the total mass is the total mass of the robot excluding the payload.

[0219] 5) Energy efficiency: According to formula (20), the energy efficiency is evaluated by the energy that the robot can save when moving on the ground. The greater the energy efficiency, the better the performance.

[0220] Table 1

[0221]

[0222] * Mintchev et al. [2] only gave a speed of 1.5 m / s under manual control, so it is believed that the trajectory tracking speed is less than 1.5 m / s

[0223] The calculation results are shown in Table 1 and Figure 9 The results show that the system has the best motion capability, switching speed and dynamic freedom, while performing well in terms of structural simplicity and energy efficiency.

[0224] Those skilled in the art can make various corresponding changes and modifications based on the above technical solutions and concepts, and all of these changes and modifications should be included in the scope of protection of the claims of the present invention.

Claims

1. A method for constructing a control system for an air-ground dual-mode robot capable of high-speed motion, the robot comprising a multi-rotor drone and wheels capable of passively rotating freely along a mounting axis; characterized in that: The specific process of the construction method is: S1. Constructing a dynamic model: The body coordinate system is (x b ,y b , z b ) and the world coordinate system is (x w ,y w , z w ); When the robot is hovering in the air, it only needs to increase the support force F S By setting it to zero, we can obtain the dynamic model of the robot; First, assume that the wheel radius, deflection, and air resistance are negligible and the robot moves on a flat surface; consider the robot's state in is the center of mass position of the robot in the world coordinate system, is the rotation matrix from the world coordinate system to the body coordinate system; the input is Where f is the total thrust, is the torque generated by the thrust, and the dynamic model based on the Newton-Euler equation is obtained: In formula (1), m is the total mass of the robot, g is the acceleration due to gravity, F S It's the supporting force. is the rotation matrix, which is composed of the horizontal velocity The angle φ between them rotates; In formula (2), is the inertia matrix, is the angular velocity in the body coordinate system, l is the length between the center of mass and the wheel center; according to the friction law, F f =F S μ, where μ is the coefficient of rolling friction; S2, differential flat output considering friction: The options for flat output are: Among them, x [s] is a stack of finite derivatives ψ is the yaw angle, and there is an additional term The specific content of flatness transformation is given as follows: (x,u)=Ψ(ξ) (4) First, multiply equation (1) by the body axis Among them: make: have and Therefore When the system is in steady state, we have This shows and are in the same direction; therefore: in Multiplying formula (1) on the left yields: Next, use the Hough transform to decompose the yaw quaternion q φ and the tilt quaternion q z : q φ =((cos(ψ / 2),0,0,sin(ψ / 2)) T (9) Because q z represents the tilt transformation, so in q z There is no z component in ; let q z =(w q , x q ,y q ,0) T ,q z By solving the equation get: The rotation matrix is defined as: in is the transformation from quaternion to rotation matrix, according to get Right now Through the above formula, we get: where sψ represents sin(ψ), cψ represents cos(ψ), and: in In addition, we also get S3. Minimum support force trajectory generation: As a dimension of the differential flatness output, F S The trajectory and have a certain degree of independence; therefore, an optimization problem is formulated - minimum collective thrust to minimize energy consumption, which allows the support force F S The planning of the position and yaw angle is separated from the planning of the 3.1) Optimization problem modeling: The objective function is the collective thrust norm with three linear inequality constraints and one nonlinear equation constraint; F S Set to a predetermined constant F Spre On the other hand, the support force must be actively committed, otherwise the robot will not stay on the ground; Note that during ground locomotion, the robot's tilt angle θ (q z ) is restricted due to its structural limitations, otherwise it would touch the ground; therefore, considering the dynamic model of the robot, an optimization problem is designed to solve the minimum support force trajectory problem; 3.2) A feasible solution: Due to another flatness output is generated by the planner, design an F with variables S Function: Horizontal acceleration and the tilt angle θ(q z ): F S =F Spre :when When it is low, due to the inverted pendulum structure of the robot, there should be a large enough torque to stabilize the posture; in this case, F S Set to a constant value F Spre To avoid vibration caused by frequent thrust changes; according to the tilt angle θ(q z ) to infer the minimum torque required for the robot to maintain stable equilibrium: For the maximum inclination angle θ max F aircraft smaller than π / 4 S As θ increases, it decreases, so: when Greater than the maximum tilt angle θ of the aircraft max and F Spre When the acceleration is lower than what it can provide, the robot must reduce the support force to provide more thrust; F S =0: The maximum acceleration that the robot can perform during ground motion is a lim , because its F S Should always be positive; if the robot is required to reach The ideal state is to S Set to zero and limit In this case, the robot cannot catch up with the trajectory, which can be avoided by limiting the acceleration of ground motion in the planning stage; at the same time, the chain rule can be used to find F S The derivative of S4. Unified controller design: The only difference between air sports and ground sports is F S Is zero, so a unified controller can be applied to both motions; the controller is a cascade position-velocity controller; first, take the desired state from the trajectory and will Added to the proportional error of the speed controller; then the required support force is calculated; then by applying the flatness transformation formula (4), the collective thrust f and the command are obtained This is then converted by the flight controller into thrust f for each motor; finally, the proportional position error is added back to the speed controller; S5. Online identification of thrust coefficient The thrust signal required by the flight controller is usually a normalized number Γ∈[0,1], so a thrust coefficient k is required f To transform f into Γ: k f can be measured by pre-calibration, but it is actually a variable that depends on the battery voltage, air density, propeller integrity and other external factors; therefore, a forgetting factor recursive least squares algorithm is used to identify k online. f : The basic form is: where a k and b k is the observed value, x k is the target to be updated, λ is the forgetting factor, which is usually set between [0.95, 1] and Between; Based on the model: in is the estimated horizontal acceleration in the world coordinate system, is in x W－ y W Normalized thrust projected on the plane, choose k f is x k , for b k , for a k .

2. The construction method according to claim 1, characterized in that A quad-rotor drone is used as the multi-rotor drone part; the quad-rotor drone adopts a frame structure made of carbon fiber plates; the frame structure is equipped with four motors, an electronic speed controller, four propellers, a flight controller, PX4 firmware and a battery; and a propeller protector is installed at the bottom of each motor.

Images