A stable obstacle-climbing control method for quadruped robots

Abstract

The present invention relates to a method for controlling a stable obstacle crossing of a quadruped robot. The structure of the quadruped robot is a front elbow and rear knee joint, and the method comprises the following steps: 1) establishing a kinematic model of the quadruped robot to obtain the force direction of the foot end in a world fixed coordinate system; 2) establishing a CPG model; 3) selecting an intermittent Trot gait for obstacle crossing based on the obstacle crossing requirements of the robot; 4) the single-leg phase of the quadruped robot is divided into three stages: a swing front phase, a swing back phase, and a support phase; combining the three types of obstacle contact force states of kicking obstacles, stepping on obstacles, and stepping on empty space, it is summarized that the robot's obstacle contact states are divided into four types: kicking obstacles in the swing front phase, kicking obstacles in the swing back phase, stepping on obstacles in the swing back phase, and stepping on empty space; 5) performing robot reflex motion planning for each obstacle contact state. The present invention fully considers the gait symmetry problem when designing the obstacle contact reflex motion, and compensates for these changes, effectively reducing the robot's yaw problem, thereby avoiding the need for subsequent repeated deviation correction.

Description

Technical Field

[0001] The invention belongs to the technical field of quadruped robot motion control technology. Background Art

[0002] With the development of science and technology, robots are increasingly used in complex and harsh environments such as tunnel construction and rescue. These scenarios put forward higher requirements for robots' ability to adapt to terrain. Although wheeled or tracked robots perform well on simple terrain, legged robots have become a research hotspot due to their excellent adaptability. In particular, quadruped robots have received widespread attention due to their excellent stability and energy efficiency.

[0003] At present, quadruped robots have made some progress in hardware, but there are still major challenges when facing outdoor environments. The main reason is that the robot's motion control in complex and rugged environments is full of difficulties. Traditional modeling-based control methods require accurate modeling of robots and environments, but these are difficult to apply to complex and changeable outdoor environments. Although intelligent control methods have emerged in recent years, which reduce the need for modeling the external environment, such methods usually require a lot of data training and parameter adjustment, making it difficult to achieve online control of robots. In view of this, researchers turned their attention to nature, seeking inspiration from animal movement mechanisms. Finally, through the study of the central nervous system of animals, a motion control method based on the central pattern generator (CPG) was proposed.

[0004] CPG-based motion control is a control method that imitates the self-excited behavior of biological nerve centers to achieve rhythmic motion. It has the advantages of simple structure, strong coupling, good adaptability, and no need for precise modeling. However, the current research on CPG-based motion control methods still focuses on rhythmic motion signal generation and gait conversion on structured terrain. Research on terrain adaptation control in complex terrain environments and the resulting motion stability problems is still relatively insufficient. At present, the complex terrain adaptation methods for quadruped robots have ignored the problem of gait symmetry. If they pass through asymmetric complex terrain continuously, the changes in step length and leg length during the obstacle crossing process will cause the robot's body to deviate. Summary of the invention

[0005] The technical problem to be solved by the present invention is how to improve the obstacle crossing stability of a quadruped robot.

[0006] In order to solve the above problems, the present invention adopts the following technical solution:

[0007] A method for controlling a quadruped robot to smoothly overcome obstacles, wherein the structure of the quadruped robot is a front elbow and rear knee joint, and the method comprises the following steps: step 1: using the DH method to establish a kinematic model of the four legs of the quadruped robot, and obtaining the force direction of the foot end in a world fixed coordinate system; step 2: establishing a CPG motion control model based on a Hopf oscillator, and controlling the rhythmic motion of the quadruped robot through a standard oscillation signal output by the CPG motion control model to realize walking on a flat road; step 3: based on the robot's obstacle-overcoming requirements, selecting an intermittent Trot gait for obstacle-overcoming; step 4: the single-leg phase of the quadruped robot is divided into three phases: a swing front phase, a swing back phase, and a support phase, wherein the swing front phase and the swing back phase are separated by the midpoint position of the single-leg swing phase; combining the three types of obstacle-touching force states of kicking obstacles, stepping on obstacles, and stepping into the air, it is summarized that the robot's obstacle-touching states are divided into four types: kicking obstacles in the swing front phase, kicking obstacles in the swing back phase, stepping on obstacles in the swing back phase, and stepping into the air; step 5: performing robot reflex motion planning for each obstacle-touching state.

[0008] Compared with the prior art, the present invention adopting the above structure has the following beneficial effects:

[0009] The present invention realizes the smooth movement of the robot in a complex terrain environment through rhythmic motion control, motion reflex and other means, and fully considers the gait symmetry problem in the obstacle crossing process. When designing the obstacle reflex movement, the gait symmetry problem is fully considered, and these changes are compensated, which effectively reduces the yaw problem of the robot, thereby avoiding the need for subsequent repeated corrections.

[0010] As a preferred embodiment, a further technical solution of the above structure is:

[0011] Preferably, the judgment formula of the single-leg phase stage of the quadruped robot is as follows:

[0012]

[0013] Where sngi is the phase judgment sign function of the ith leg; where sng i When the value is 1, it is the pre-swing phase, when it is -1, it is the post-swing phase, and when it is 0, it is the support phase;

[0014] The force state of the foot touching the obstacle is judged as follows:

[0015] The formulas for the robot's foot-end kicking and stepping on obstacles are as follows:

[0016]

[0017] Where Fyi is the vertical force on the foot of the ith leg in the world fixed coordinate system;

[0018] Fxi—the forward force on the foot end of the ith leg in the world fixed coordinate system;

[0019] μ—static friction coefficient of the ground;

[0020] If the value of the function f(Fxi, Fyi) is 1, it is in the obstacle-kicking state; if the value of f(Fxi, Fyi) is -1, it is in the obstacle-stepping state; if the value of f(Fxi, Fyi) is 0, and the motion phase is the transition from the robot's swinging back phase to the supporting phase, and the three-dimensional force sensor at the foot end does not detect any force, then the robot is in the foot-end stepping empty state.

[0021] Preferably, the swing pre-phase kick barrier reflex movement process is:

[0022] S1: Backward obstacle avoidance: When the foot detects a collision in the pre-swing phase, the knee joint is kept stationary and the hip joint swings backward to escape from the obstacle;

[0023] S2: Leg lifting contraction: by swinging the hip joint backward and the knee joint upward, a scissor contraction is performed to lift the foot end to cross the obstacle;

[0024] S3: Unchanged obstacle crossing: The hip joint swings forward and the knee joint does not move, thus completing the obstacle crossing action with one leg;

[0025] S4: Landing back: By swinging the hip joints forward and the knee joints downward, perform a scissor-like extension movement so that the foot ends land smoothly in a basically vertical direction.

[0026] Preferably, the pre-swing phase obstacle kicking reflex movement is decomposed, wherein the movement of the knee joint includes the knee joint remaining unchanged, contraction and lifting, the knee joint remaining unchanged over the obstacle and extension and falling back, and the knee joint control signal is as follows:

[0027]

[0028] Where t f —The moment of hitting an obstacle during the movement cycle;

[0029] y f —The angle value of the knee joint when it hits the obstacle;

[0030] x f —The angle value of the anterior and posterior hip joint when it hits the obstacle;

[0031] t—current time;

[0032] A1—change in knee joint angle;

[0033] B i , B2, B3—changes in the anterior-posterior angle of the hip joint;

[0034] T i , T2, T3—Obstacle crossing stage time;

[0035] T0—total obstacle crossing time;

[0036] The kicking reflex movement in the pre-swing phase is decomposed, where the movement of the front and rear joints of the hip includes retreating to avoid obstacles, contracting and retreating, swinging forward and extending and swinging forward. The control signals of the front and rear joints of the hip are as follows:

[0037]

[0038] Where x f —The angle value of the anterior and posterior hip joint when it hits the obstacle;

[0039] The formulas for time T0, T1, T2, and T3 are as follows:

[0040] T0=T sw +T kong =T / 2

[0041]

[0042]

[0043] A1 is the knee angle in the stage of lifting the foot to cross the obstacle, B1, B2, and B3 are the hip anterior and posterior angles in the stage of retreating after hitting the obstacle, the stage of lifting the leg and contracting, and the stage of falling back to the ground. The angle A1 needs to meet the following formula:

[0044]

[0045] Where L i —length of the leg at stage i;

[0046] h z —Height of obstacle, h z =100mm;

[0047] H0—the distance between the hip joint base and the ground;

[0048] y f —Knee joint angle when hitting the obstacle;

[0049] The rotation angle B1 is a set angle and is considered as a known quantity;

[0050] The rotation angle B2 is derived from the following formula:

[0051]

[0052] B2=θ h1 +θ h2 -x f -B1

[0053] Where θ hi —Anterior and posterior auxiliary angle of the hip joint at stage i;

[0054] x f —Anterior-posterior angle of the hip joint when hitting the obstacle;

[0055] In the unchanged obstacle crossing stage, the knee joint remains unchanged, so it is easy to get θ h3 =θ h2 , so the corner B3 is derived as follows:

[0056] B3=θ h1 +x f +B1+B2-θ h3

[0057] Preferably, the obstacle-kicking reflex movement process in the post-swing phase is as follows: S1: obstacle detection: when the foot end detects the obstacle-kicking state in the post-swing phase, the obstacle-kicking leg performs the post-swing phase obstacle-kicking reflex movement; S2: obstacle withdrawal: when the post-swing phase obstacle-kicking reflex movement is triggered, the knee joint does not move, and the front and rear joints of the hip move backward to get away from the obstacle; S3: emergency landing: due to time constraints, it is not enough to complete the obstacle crossing action, and the foot end of the robot lands urgently and enters the support phase in advance. At this time, the leg length of the obstacle-kicking leg changes; S4: leg lifting and contraction: through the backward swing of the front and rear joints of the hip and the upward lifting of the knee joint, the foot end is raised by a scissor-like contraction action to facilitate the subsequent crossing of the obstacle; S5: unchanged obstacle crossing: the hip joint swings forward, and the knee joint keeps the angle unchanged to complete the single-leg obstacle crossing action; S6: falling back and landing: the knee joint falls back to the ground, and the front and rear joints of the hip continue to swing forward, and the scissor-like extension is used to ensure that the foot end basically falls steadily in the vertical direction. At this time, the leg length of the obstacle-kicking leg changes again.

[0058] Preferably, the post-swing phase kicking reflex movement is decomposed, wherein the reflex movement of the knee joint in the first movement cycle includes the knee joint remaining unchanged, landing and lowering, and the support phase. The specific reflex movement formula is as follows:

[0059]

[0060] In the second movement cycle, the reflex movement of the knee joint is decomposed into contraction and lifting, unchanged obstacle crossing, extension and fall, support phase and leg length reset. In the leg length reset phase, in order to prevent the continuous accumulation of leg length changes when continuously touching obstacles, the specific reflex movement formula is as follows:

[0061]

[0062] In the support phase of the reflex motion, the swing amplitude adjustment caused by the change of leg length is introduced. When the support phase leg length of the robot increases from L0 to L', the support phase swing amplitude is adjusted from A to L. h Adjust to A h ' to ensure stride consistency;

[0063] The swing phase kicking reflex movement is decomposed, in which the hip front and back joint movement in its first cycle, the reflex action is decomposed into obstacle avoidance retreat, retreat landing and support phase. The specific reflex movement formula is as follows:

[0064]

[0065] In the second movement cycle of the anterior and posterior hip joints, the reflex action is decomposed into contraction and withdrawal, forward swing, extension forward swing, support phase and hip joint immobility. In the forward swing and extension forward swing phases, the change in step length caused by the landing phase of the previous cycle is considered. At the same time, during the support phase, the effect of leg length changes on the swing amplitude is considered, and the leg length needs to be reset at the end of the support phase. The specific reflex movement formula is as follows:

[0066]

[0067] G1=x f -D1-D2-D3+D4+D5

[0068] T1=(T sw -t f +T kong ) / 3

[0069] T2=2(T sw -t f +T kong ) / 3

[0070] Among them, C1, C2, and C3 are the knee joint rotation angles in the backward obstacle avoidance stage, emergency landing stage, and leg lifting and contraction stage respectively; D1, D2, D3, D4, and D5 are the hip anterior and posterior joint rotation angles in the backward obstacle avoidance stage, emergency landing stage, leg lifting and contraction stage, unchanged obstacle crossing stage, and falling back and landing stage respectively; the angles D1 and D2 are the set hip joint rotation angles, which are regarded as known quantities, and D1 and D2 are set to 0.1rad. The angle C1 is derived by the following formula:

[0071] C1=θ k3 +y f

[0072] θ h3 =θ h0 -(x f -D1-D2)

[0073]

[0074] Where θki—knee joint auxiliary angle at stage i;

[0075] The rotation angle D3 is derived from the following formula:

[0076] D3=θ h4 -θh3

[0077]

[0078] Where S3 is the step length of the previous motion cycle;

[0079] The rotation angle C2 is derived from the following formula:

[0080] C2=θ k3 -θ k4

[0081]

[0082] The rotation angle D4 is derived from the following formula:

[0083] D4=θ h4 -θ h5

[0084]

[0085] The rotation angle D5 is derived from the following formula:

[0086] D5=θ h5 -θ h6

[0087]

[0088] The rotation angle C3 is derived from the following formula:

[0089] C3=θ k6 -θ k5 =θ k6 -θ k4

[0090]

[0091] Preferably, the post-swing phase obstacle stepping reflex motion modeling is specifically as follows: an impedance control method is introduced, and the second-order differential form expression of the impedance control is as follows:

[0092]

[0093] F e =Mg / n

[0094] Where K d — expected stiffness matrix;

[0095] B d — expected damping matrix;

[0096] M d — expected inertia matrix;

[0097] △F—deviation between actual force and expected force;

[0098] F r —actual force;

[0099] F e —Expectancy;

[0100] n—number of supporting legs;

[0101] M—the overall mass of the robot;

[0102] Rewrite the above formula into a transfer function form, which belongs to a typical second-order system. It is tuned according to the second-order system, and the formula is as follows:

[0103]

[0104] Stiffness parameter K d It mainly affects the steady-state output value of the system, while the damping parameter B d and inertia parameter M d It only affects the dynamic process of the system;

[0105] Planning parameters for knee joint angular velocity in CPG motion control model Take the derivative It is regarded as the relevant planning parameter of knee joint angular acceleration; use The impedance stiffness is formed and the above formula is optimized into the following formula:

[0106]

[0107] Where k is the stiffness coefficient;

[0108] b—Stiffness coefficient 2;

[0109] —When the i-th leg triggers the impedance The value of

[0110] Δθ ki —Impedance adjustment value;

[0111] The input in the above formula is the vertical force component received by the foot end in the fixed world coordinate system, and the output is the angle change caused by impedance. At the end of the swing phase, the angle change caused by impedance will be maintained during the support phase. At the same time, because the raised terrain is only a local terrain, the impedance value is cleared at the end of the swing phase, so that the length of the leg that touches the obstacle is restored to the initial length; the control signal of the knee joint is as follows:

[0112]

[0113] The front and rear hip joints complete the swing of the stance phase before the reset time. The control signals of the front and rear hip joints in the stance phase after correction are as follows:

[0114]

[0115] Where L sw — leg length at the end of the swing phase;

[0116] Taking into account the situation where the expected step length is not achieved due to early landing, the control signal of the hip anterior and posterior joint swing phase in the next movement cycle is as follows:

[0117]

[0118] For ease of calculation, the step length and time are simplified into a linear relationship in the above formula, and the proportion of the time of the trigger impedance to the entire swing phase time is used to approximately estimate the step length that has been walked, so as to compensate for the step length in the next swing phase.

[0119] Preferably, the single-leg stepping reflex movement process is:

[0120] S1: Ground contact waiting: When the foot end does not detect force at the end of the swing phase, the stepping reflex movement is not triggered immediately, but the ground contact waiting time is increased. If the foot end still does not detect force during the waiting period, it is determined that the robot has stepped on the ground with one leg;

[0121] S2: Lifting the leg to escape: When the robot triggers the stepping reflex, in order to avoid hitting the obstacle again, the robot first keeps the front and rear hip joints still and lifts the knee joints to get the foot out of the pit.

[0122] S3: Forward swing and landing: After the foot leaves the pit, the hip joint swings forward and the knee joint drops, which increases the leg length and allows the foot to land on the ground, thus crossing the pit terrain.

[0123] Preferably, the single-leg stepping reflex movement modeling is specifically: the single-leg stepping reflex movement is decomposed, and the included knee joint reflex movement includes ground contact waiting, knee joint lifting, landing and lowering, support phase and leg length reset. The reflex movement model of the knee joint is as follows:

[0124]

[0125] Where E1, E2—changes in knee joint angle;

[0126] The single-leg stepping reflex motion is decomposed. The hip anterior and posterior joint reflex motion includes waiting for touchdown, immobilization of the hip anterior and posterior joints, swinging forward to overcome obstacles, swinging back after support, and resetting the leg length. During the support phase, the robot's leg length changes again. Therefore, considering the swing amplitude change caused by the leg length change, the hip anterior and posterior joint reflex motion model is as follows:

[0127]

[0128] Where F1 is the change in the joint angle of the anterior and posterior hip joint;

[0129] The values ​​of T1, T2, and T3 are as follows:

[0130]

[0131] In the next swing phase, the step length is adjusted from S / 2 to S / 2-S4. The formula for the anterior and posterior hip joint in the next swing phase is as follows:

[0132]

[0133] E1 and E2 are the knee joint rotation angles in the leg lifting and separation phase and the front swing landing phase respectively, and F1 is the hip anterior-posterior joint rotation angle in the front swing landing phase.

[0134] Preferably, an intermittent Trot gait with a load factor β of 0.7 is selected during obstacle crossing. BRIEF DESCRIPTION OF THE DRAWINGS

[0135] Figure 1 is a flow chart of the steps of the present invention;

[0136] Figure 2 This is a schematic diagram of the front leg structure of the present invention;

[0137] Figure 3 This is a schematic diagram of the hind leg structure of the present invention;

[0138] Figure 4 This is the intermittent Trot gait diagram of the robot of the present invention;

[0139] Figure 5 This is a flow chart of judging the foot end obstacle state of the present invention;

[0140] Figure 6 It is a schematic diagram of the barrier-kicking reflex motion in the pre-swing phase of the present invention;

[0141] Figure 7 It is a diagram of the barrier kicking motion in the pre-swing phase of the present invention;

[0142] Figure 8 This is a schematic diagram of the first cycle of the kicking barrier reflex movement after the swing of the present invention (knee joint);

[0143] Fig. 9 This is a schematic diagram of the second cycle of the kicking barrier reflex movement after the swing of the present invention (knee joint);

[0144] Fig.10 It is a schematic diagram of the support phase swing compensation of the present invention;

[0145] Fig.11This is a schematic diagram of the first cycle of the kicking barrier reflex after the swing phase of the present invention (hip joint);

[0146] Fig.12 This is a schematic diagram of the second cycle of the kicking barrier reflex after the swing of the present invention (hip joint);

[0147] Fig.13 is the value under Trot gait of the present invention;

[0148] Fig.14 This is a schematic diagram of the step-in-the-air reflex motion of the present invention;

[0149] Fig.15 It is a schematic diagram of swing phase step length compensation of the present invention;

[0150] Fig.16 It is a diagram of the reflex motion of stepping in the air motion of the present invention;

[0151] Fig.17 This is a simulation of complex terrain obstacle crossing in the present invention, wherein Figure a is a top view of the starting stage, Figure b is the left front leg kicking the obstacle after swinging, Figure c is the left hind leg kicking the obstacle before swinging, Figure d is the right hind leg stepping into the air at the end of the swing phase, Figure e is the right front leg stepping on the obstacle after swinging, and Figure f is a top view of the completion of obstacle crossing;

[0152] Fig.18 It is the control signal for the complex terrain obstacle crossing of the present invention;

[0153] Fig.19 is the Euler angle curve of the center of mass of the quadruped robot of the present invention;

[0154] Fig. 20 It is the centroid displacement curve in the lateral displacement direction of the centroid displacement in complex terrain obstacle crossing of the present invention;

[0155] Fig.21 It is the centroid displacement curve in the vertical direction of the centroid displacement of the complex terrain obstacle crossing method of the present invention;

[0156] Fig. 22 The centroid displacement curve of the centroid displacement in the forward direction when overcoming obstacles in complex terrain. DETAILED DESCRIPTION

[0157] The present invention will be further described below in conjunction with embodiments, the purpose of which is only to provide a better understanding of the content of the present invention. Therefore, the examples given do not limit the protection scope of the present invention.

[0158] See also Figure 1 The quadruped robot stable obstacle crossing control method provided by the present invention comprises the following steps:

[0159] Step 1: Use the DH method to establish the kinematic model of the four legs of the quadruped robot and obtain the force direction of the foot end in the world fixed coordinate system.

[0160] Specifically, the shutdown configuration structure of the quadruped robot of the present invention is a front elbow and rear knee joint, and the kinematic model of the four legs of the quadruped robot is established by using the DH method. The schematic diagram of the front and rear legs of the quadruped robot is as follows: Figure 2 , Figure 3 As shown, the coordinate system O b -X b Y b Z b , established at the center of mass of the robot, represents the robot's trunk coordinate system, whose coordinate axis X b The positive direction is the robot's forward direction, coordinate axis Y b The positive direction is the opposite direction of gravity, and the coordinate axis Z b The positive direction is determined by the right-hand rule.

[0161] Coordinate system O w -X w Y w Z w , represents the fixed world coordinate system, coordinate axis X w The positive direction is the robot's forward direction, coordinate axis Y w The positive direction is the opposite direction of gravity, and the coordinate axis Z w The positive direction is determined according to the right-hand rule, and the world coordinate system is highly consistent with the initial center of mass of the robot.

[0162] The coordinate system establishment method of the robot's four legs is basically the same. First, a single-leg base coordinate system O0-X0Y0Z0 is established above each leg. The origin of the base coordinate system of each leg is set at the center of the hip lateral swing joint. The positive direction of the coordinate axis Z0 is the forward direction of the robot, and the positive direction of the coordinate axis X0 is the opposite direction of gravity. The four single-leg base coordinate systems are based on the torso coordinate system O b -X b Y b Z b Symmetrical arrangement.

[0163] Coordinate systems O1-X1Y1Z1, O2-X2Y2Z2, and O3-X3Y3Z3 are established at the three joints of each leg, with the origin located at the center of each joint. Z1, Z2, and Z3 are rotation axes of three degrees of freedom, the positive directions of X1, X2, and X3 are the extension directions of the hip side swing rod, thigh rod, and calf rod, and Y1, Y2, and Y3 are obtained by the right-hand rule; θ1, θ2, and θ3 are the rotation angles of each joint of the machine, θ1 swings to the right as the positive direction, θ2 swings forward as the positive direction, and θ3 swings upward as the positive direction.

[0164] Finally, a foot-end coordinate system O4-X4Y4Z4 is set at the foot of each leg, where the positive direction of the Z4 axis is the extension direction of the robot's calf, the positive direction of the X4 axis is perpendicular to the calf and upward, and the positive direction of the Y4 axis is determined by the right-hand rule.

[0165] according to Figure 2 , Figure 3 From the schematic diagram of the single leg structure, the DH parameters of the front and rear legs can be obtained, see Table 1 and Table 2.

[0166] Table 1 Parameters of the front leg DH joint

[0167]

[0168] Table 2 Parameters of DH joint of hind leg

[0169]

[0170] Simplify the notation for trigonometric functions, i.e. s i = sinθ i , c i = cosθ i Therefore, the change matrix of the foot end of the left front leg in the world fixed coordinate system can be deduced as follows:

[0171]

[0172] Where x0, y0, z o —The distance between the single-leg base coordinate system and the center of mass coordinate system;

[0173] x w 、z w —The distance of the center of mass coordinate system relative to the world coordinate system;

[0174] R F —The rotation matrix of the foot end coordinate system of the front leg relative to the world coordinate system.

[0175] The change matrix of the foot end of the right front leg in the world fixed coordinate system is as follows:

[0176]

[0177] The change matrix of the foot end of the right hind leg in the world fixed coordinate system is as follows:

[0178]

[0179]

[0180] Where R H —The rotation matrix of the foot-end coordinate system of the hind leg relative to the world coordinate system.

[0181] The change matrix of the foot end of the left hind leg in the world fixed coordinate system is as follows:

[0182]

[0183] By multiplying the three-dimensional force sensor information at the end of each foot of the robot's four legs by the rotation matrix of the corresponding leg, the force direction of the foot end in the world fixed coordinate system can be obtained.

[0184] Step 2: Establish a CPG motion control model based on the Hopf oscillator, and control the rhythmic motion of the quadruped robot through the standard oscillation signal output by the CPG motion control model to achieve walking on flat roads.

[0185] The mathematical model of the CPG motion control model is as follows:

[0186]

[0187] Where x, y—state variables of the oscillator;

[0188] ω — frequency after the oscillator converges;

[0189] α—oscillator convergence coefficient;

[0190] u1, u2—oscillator external feedback terms.

[0191] ω sw —support phase frequency;

[0192] ω st —swing phase frequency;

[0193] a—the speed of switching between the swing phase and the support phase;

[0194] β—load factor;

[0195] — the relative phase between oscillators i and j;

[0196] — the phase of the i-th oscillator;

[0197] — rotation matrix between i and j oscillators;

[0198] θ hi —Control signal of the front and rear hip joints of the i-th leg;

[0199] θ ki —Control signal of the knee joint of the i-th leg;

[0200] A h —Amplitude of hip anteroposterior motion;

[0201] A k —Amplitude of motion of the knee joint.

[0202] Through θ hi and θ kiIt can generate hip and knee joint control signals that produce different gaits, thereby realizing the rhythmic movement of the quadruped robot on flat ground.

[0203] Step three: Based on the robot's obstacle crossing requirements, select the intermittent Trot gait for obstacle crossing. In order to improve the robot's adaptability and obstacle crossing ability in complex terrain, the improved intermittent Trot gait is introduced. This gait is improved on the basis of the conventional Trot gait. By adding a transition period in which all four legs touch the ground at the same time at the phase change point, the stability of the alternating movement of the legs is enhanced. In addition, the intermittent Trot gait also reserves more reflex movement time for the robot's possible obstacle contact, further enhancing its ability to adapt to complex terrain. This gait combines the advantages of Walk and Trot, allowing the robot to maintain a relatively fast speed and good dynamic stability while achieving more flexible and robust terrain adaptability. The intermittent Trot gait load coefficient β set in this scheme is 0.7. By adding a transition period in which all four legs touch the ground at the same time at the phase change point, the stability of the alternating movement of the legs is enhanced. For specific gait timing, see Figure 4 As shown, in the intermittent Trot gait, the swing phase time (T sw ) accounts for 30% of the entire gait cycle (T), and the time of the stance phase (T st ) accounts for 70% of the entire gait cycle, and each phase transition time (T kong ) is 20% of the total cycle, during which all four legs are in contact with the ground.

[0204] Step 4: Analysis and judgment of single-leg obstacle status,

[0205] The quadruped robot of the present invention is mainly used to perceive the terrain in a tunnel environment and explore the internal structure of the tunnel. After on-site investigation, it is determined that the ground it faces is compact ground. Therefore, during the support phase, the robot will not have the problem of support collapse. Therefore, the present invention focuses on the adaptation strategy of the robot encountering obstacles during the single-leg swing phase.

[0206] In order to refine the obstacle adaptation strategy, the obstacle time is first classified into: pre-swing phase, post-swing phase and support phase. The pre-swing phase and post-swing phase are separated by the midpoint of the single-leg swing phase, that is, when the robot lifts its foot to the highest point, the judgment formula of the single-leg phase of the quadruped robot is as follows:

[0207]

[0208] In the formula, sng i —The phase judgment sign function of the i-th leg;

[0209] In the above formula, sng i When the value is 1, it is the pre-swing phase, when it is -1, it is the post-swing phase, and when it is 0, it is the support phase.

[0210] The obstacle contact force states are divided into three categories: kicking obstacles, stepping on obstacles, and stepping on empty obstacles. The judgment of stepping on empty obstacles is relatively easy. When the single leg swing phase ends and transitions to the support phase, if no force is detected at the foot end, it is judged that the stepping on empty obstacles is not performed. The judgment of kicking obstacles and stepping on obstacles is relatively complicated, because in the swing phase, the single leg always produces displacement in the forward direction and the vertical direction. In addition, since the contact surface between the foot end and the obstacle is rough, when a collision occurs, the foot end will be simultaneously subjected to the collision force and the friction force caused by the contact surface. Therefore, when the robot foot kicks or steps on an obstacle, it will be simultaneously subjected to the force in the forward direction and the force in the vertical direction. This phenomenon provides a way to distinguish the direction of the kicking and stepping on obstacle states by comparing the ratio of the forward direction force to the vertical direction force. Based on this, the formula for judging the robot foot end's kicking and stepping on obstacle states is as follows:

[0211]

[0212] Where F yi —The vertical force on the foot of the ith leg in the world fixed coordinate system;

[0213] F xi —The forward force on the foot of the ith leg in the world fixed coordinate system;

[0214] μ—static friction coefficient of the ground.

[0215] In the above formula, if the function f(F xi , F yi ) is 1, it means that the force in the forward direction of the robot's foot exceeds the maximum static friction force, indicating that the robot has encountered an obstacle kicking state; if f(F xi , F yi ) is -1, it means that the forward force on the robot foot in the world fixed coordinate system is less than the static friction force and the force is detected in the vertical direction, and it is judged as the obstacle stepping state; if f(F xi , F yi ) is 0, indicating that the robot foot is not under force. If the three-dimensional force sensor at the foot does not detect any force during the transition from the robot swing phase to the support phase, this indicates that the robot is in an empty foot state.

[0216] Based on the above analysis, the robot's obstacle-hitting moments are divided into the pre-swing phase, post-swing phase, and support phase (i.e., the end of the swing phase). The foot-end force states can be divided into kicking obstacles, stepping on obstacles, and stepping into the air. Since the foot is always raised during the pre-swing phase, there will be no stepping on obstacles. Therefore, the robot's obstacle-hitting states can be subdivided into kicking obstacles in the pre-swing phase, kicking obstacles in the post-swing phase, stepping on obstacles in the post-swing phase, and stepping into the air. For the judgment process of the robot's single-leg foot-end obstacle-hitting state, see Figure 5 .

[0217] Step 5: Perform robot reflex motion planning for each obstacle state.

[0218] A. Pre-swing phase barrier kicking reflex movement: The reflex movement in the barrier-touching state is designed to be divided into four stages: ① retreat to avoid the barrier, ② lift the leg and contract, ③ cross the barrier unchanged, and ④ fall back to the ground. Figure 6 ,The foot end trajectory in the figure is an approximate motion trajectory for reference, not the real foot end trajectory.

[0219] Specifically, the movement process of the kicking reflex in the pre-swing phase is:

[0220] S1: Backward obstacle avoidance: When the foot detects a collision in the pre-swing phase, keep the knee joint still and swing back through the front and rear joints of the hip to get away from the obstacle.

[0221] S2: Leg lift contraction: By swinging the hip joints backward and the knee joints upward, a scissor-like contraction movement is performed to lift the foot end over the obstacle. The purpose of this movement is to prevent the foot end from touching the obstacle forward again.

[0222] S3: Unchanged obstacle crossing: The hip joint swings forward and the knee joint remains stationary, thus completing the obstacle crossing action with one leg.

[0223] S4: Landing phase: By swinging the hip joints forward and the knee joints downward, a scissor-like extension movement is performed to allow the foot to land smoothly in a vertical direction.

[0224] The pre-swing phase barrier-kicking reflex movement will return to the initial expected position after crossing the obstacle, so there is no need to consider the influence of changes in step length and leg length.

[0225] The above-mentioned pre-swing phase kicking reflex movement is decomposed, in which the movement of the knee joint includes the knee joint remaining unchanged, contraction and lifting, the knee joint remaining unchanged over the obstacle and extension and falling back. The knee joint control signal is as follows:

[0226]

[0227] Where t f —The moment of hitting an obstacle during the movement cycle;

[0228] y f —The angle value of the knee joint when it hits the obstacle;

[0229] x f —The angle value of the anterior and posterior hip joint when it hits the obstacle;

[0230] t—current time;

[0231] A1—change in knee joint angle;

[0232] B i , B2, B3—changes in the anterior-posterior angle of the hip joint;

[0233] T i , T2, T3—Obstacle crossing stage time;

[0234] T0—Total obstacle crossing time.

[0235] The above swing-front phase kicking reflex movement is decomposed, wherein the movement of the front and rear joints of the hip includes retreating to avoid obstacles, contracting and retreating, swinging forward and extending and swinging forward, and the control signals of the front and rear joints of the hip are as follows:

[0236]

[0237] Where x f —The angle value of the anterior and posterior hip joint when it hits the obstacle.

[0238] The formulas for time T0, T1, T2, and T3 are as follows:

[0239] T0=T sw +T kong =T / 2 (5-3)

[0240]

[0241] Illustration of the swing front phase kicking movement Figure 7 As shown in the figure, during the movement of the robot, the swing phase and the support phase work together to achieve a swing distance of S. In fact, the swing leg only swings a step length of S / 2, and the other step length of S / 2 is provided by the support leg on the other side. To simplify the calculation, the influence of the support leg on the swing leg should be ignored. Therefore, under the condition of a fixed hip joint base, the actual swing step length of the swing leg is S / 2.

[0242] The quadruped robot needs to determine four parameters, A1, B1, B2 and B3, for the pre-phase motion reflex of single-leg swing. A1 is the knee joint rotation angle in the obstacle-lifting stage, and B1, B2 and B3 are the anterior and posterior hip joint rotation angles in the obstacle-hitting retreat stage, leg-lifting and contraction stage, and landing stage, respectively.

[0243] Corner A1 needs to satisfy the conditions of formula (5-6):

[0244]

[0245] Where L i —length of the leg at stage i;

[0246] h z —Height of obstacle, h z =100mm;

[0247] H0—the distance between the hip joint base and the ground;

[0248] yf —Knee angle at the time of impact.

[0249] The foot-lifting height set in formula (5-7) must be higher than the obstacle height. By default, the height of all obstacles does not exceed 100 mm. Therefore, setting A1 to 0.15 rad not only meets the obstacle crossing requirements but is also within the leg and foot working space.

[0250] The rotation angle B1 is a set angle and is regarded as a known quantity. In this paper, the rotation angle B1 is set to 0.1 rad.

[0251] The turning angle B2 can be derived from formula (5-10):

[0252]

[0253] B2=θ h1 +θ h2 -x f -B1 (5-10)

[0254] Where θ hi —Anterior and posterior auxiliary angle of the hip joint at stage i;

[0255] x f —The anterior-posterior angle of the hip joint when hitting the obstacle.

[0256] In the unchanged obstacle crossing stage, the knee joint remains unchanged, so it is easy to get θ h3 =θ h2 , so the corner B3 can be derived as:

[0257] B3=θ h1 +x f +B1+B2-θ h3 (5-11)

[0258] B. Post-swing phase barrier kicking reflex movement: The reflex movement is planned as 6 swing phase stages within 2 movement cycles: ① obstacle detection, ② backward obstacle avoidance, ③ emergency landing, ④ leg lifting and contraction, ⑤ unchanged obstacle crossing and ⑥ falling back to the ground. Taking the knee joint as an example, Figure 8 , Fig. 9 As shown, the foot end trajectory is still an approximate reference motion trajectory.

[0259] The movement process of the kicking reflex after the swing phase is:

[0260] S1: obstacle detection stage: when the foot detects the obstacle-kicking state in the post-swing phase, the obstacle-kicking leg performs the post-swing phase obstacle-kicking reflex movement; S2: obstacle-avoidance retreat stage: when the post-swing phase obstacle-kicking reflex movement is triggered, the knee joint does not move, and the front and rear joints of the hip move away from the obstacle to avoid the foot from colliding again;

[0261] S3: Emergency landing phase: Due to time constraints, the robot cannot complete the obstacle crossing action, so the robot's foot lands urgently and enters the support phase in advance. At this time, the length of the leg that touches the obstacle changes.

[0262] S4: Leg lifting and contraction phase: through the posterior and posterior swing of the hip joint and the upward lifting of the knee joint, the foot end is lifted by a scissor-like contraction action to facilitate the subsequent crossing of obstacles;

[0263] S5: Unchanged obstacle crossing stage: The hip joint swings forward, the knee joint maintains an unchanged angle, and the single-leg obstacle crossing action is completed;

[0264] S6: Landing phase: the knee joint falls back to the ground, while the front and rear joints of the hip continue to swing forward, and the scissor-like extension is used to ensure that the foot ends fall smoothly in the vertical direction. At this time, the length of the leg that touches the obstacle changes again.

[0265] The overall reflex movement is decomposed into the movement of the knee joint. The reflex movement in the first movement cycle includes the knee joint remaining unchanged, landing and lowering, and the support phase. The specific reflex movement formula is as follows:

[0266]

[0267] In the second movement cycle, the reflex movement of the knee joint is divided into the following stages: contraction and lifting, unchanged obstacle crossing, extension and fall, support phase, and leg length reset. In the leg length reset phase, in order to prevent the continuous accumulation of leg length changes when continuously touching obstacles, the specific reflex movement formula is as follows:

[0268]

[0269] To ensure the smoothness of the robot's motion, the swing amplitude adjustment caused by the change in leg length needs to be considered during the support phase of the reflex motion, see Fig.10 When the robot's support phase leg length increases from L0 to L', the original support phase swing amplitude A is maintained. h This will cause inconsistent stride lengths of the opposite legs, resulting in fuselage deviation. Therefore, the swing amplitude of the support phase needs to be changed from A to h Adjust to A h ' to ensure stride consistency.

[0270] In the first cycle of the hip anterior and posterior joint movement, the reflex action is divided into the following phases: obstacle avoidance and retreat, retreat and landing, and support phase. The formula for the specific reflex action is as follows:

[0271]

[0272] In the second movement cycle of the anterior and posterior hip joint, the reflex action is divided into the following stages: contraction and withdrawal, forward swing, extension forward swing, support phase and hip joint immobility; in the forward swing and extension forward swing phases, the change in step length caused by the landing phase of the previous cycle must be considered. At the same time, during the support phase, the impact of leg length changes on the swing amplitude is considered, and the leg length needs to be reset at the end of the support phase.

[0273] The specific reflex motion formula is as follows:

[0274]

[0275] G1=x f -D1-D2-D3+D4+D5 (6-5)

[0276] T1=(T sw -t f +T kong ) / 3 (6-6)

[0277] T2=2(T sw -t f +T kong ) / 3 (6-7)

[0278] The quadruped robot needs to determine the parameters of C1, C2, C3, D1, D2, D3, D4 and D5 when performing the obstacle-kicking reflex movement after a single-leg swing. C1, C2, C3 are the knee joint rotation angles in the backward obstacle avoidance stage, emergency landing stage and leg lifting and contraction stage respectively. D1, D2, D3, D4, D5 are the hip anterior-posterior joint rotation angles in the backward obstacle avoidance stage, emergency landing stage, leg lifting and contraction stage, unchanged obstacle crossing stage and falling back to the ground stage respectively. The diagram of the obstacle-kicking reflex movement after a single-leg swing is as follows Fig.11 , Fig.12 shown.

[0279] The rotation angles D1 and D2 are the set hip joint rotation angles and are considered as known quantities. In this paper, D1 and D2 are set to 0.1 rad.

[0280] The rotation angle C1 is derived from formula (6-8):

[0281] C1=θ k3 +y f (6-8)

[0282] θ h3 =θ h0 -(x f -D1-D2) (6-9)

[0283]

[0284] Where θ ki— Knee auxiliary angle at stage i.

[0285] The rotation angle D3 is derived from formula (6-11):

[0286] D3=θ h4 -θ h3 (6-11)

[0287]

[0288] Where S3 is the step length of the previous motion cycle.

[0289] The rotation angle C2 is derived from formula (6-15):

[0290] C2=θ k3 -θ k4 (6-15)

[0291]

[0292] The rotation angle D4 is derived from formula (6-17):

[0293] D4=θ h4 -θ h5 (6-17)

[0294]

[0295] The turning angle D5 is derived from formula (6-19):

[0296] D5=θ h5 -θ h6 (6-19)

[0297]

[0298] The turning angle C3 is derived from formula (6-22):

[0299] C3=θ k6 -θ k5 =θ k6 -θ k4 (6-22)

[0300]

[0301] C. Reflex movement after stepping on obstacles in the swing phase: To address the impact problem on the foot caused by stepping on obstacles, active compliance control needs to be introduced. Through active compliance control, the leg stiffness is adjusted to adapt to terrain changes, thereby ensuring stability in movement. To ensure that the robot can walk smoothly along the predetermined route in complex terrain, in addition to applying active compliance control to adapt to the terrain, it is also necessary to actively adjust the swing amplitude of the swing phase and the support phase according to the changes in leg length and step length to achieve coordinated movement of the four legs, thereby ensuring the stability of the robot's movement and enhancing its adaptability and stability in complex terrain.

[0302] Impedance control is a commonly used method in active compliance control. CPG, as a bio-inspired control mechanism, is mainly responsible for generating periodic joint control signals and outputting desired position information. Therefore, it is more appropriate to choose impedance control based on the position inner loop. This method allows the system impedance characteristics to be adjusted to respond to external forces, thereby slowing down or stopping joint movement, thereby achieving compliant interaction with raised terrain.

[0303] The second-order differential form expression of impedance control is as follows:

[0304]

[0305] F e =Mg / n (7-2)

[0306] Where K d — expected stiffness matrix;

[0307] B d — expected damping matrix;

[0308] M d — expected inertia matrix;

[0309] △F—deviation between actual force and expected force;

[0310] F r —actual force;

[0311] F e —Expectancy;

[0312] n—number of supporting legs;

[0313] M—the overall mass of the robot.

[0314] Equation (7-1) can be rewritten into a transfer function form, which belongs to a typical second-order system and can be adjusted according to the second-order system. The formula is as follows:

[0315]

[0316] Stiffness parameter K dIt mainly affects the steady-state output value of the system, while the damping parameter B d and inertia parameter M d It only affects the dynamic process of the system. When the robot steps on an obstacle, the main impact force is generated in the vertical direction of the foot. In order to prevent damage to the leg structure, the impact force needs to be reduced. When the mass is constant, the magnitude of the impact force is mainly determined by the acceleration. When the foot steps on an obstacle, it can be approximately considered that the knee joint provides the vertical displacement and velocity, so the acceleration of the knee joint is the main factor affecting the magnitude of the impact force. The planning parameter of the knee joint angular velocity in formula (2-1) is Taking the derivative we get It can be regarded as the relevant planning parameter of knee joint angular acceleration. The value of Fig.13 shown.

[0317] use The impedance stiffness can be more flexibly adapted to the foot impact, and the formula (7-3) is optimized to the formula (7-4).

[0318]

[0319] Where k is the stiffness coefficient, k = -5;

[0320] b—stiffness coefficient, b=20;

[0321] —When the i-th leg triggers the impedance The value of

[0322] Δθ ki —Impedance adjustment value.

[0323] The input in formula (7-4) is the vertical force component of the foot in the fixed world coordinate system, and the output is the angle change caused by impedance. At the end of the swing phase, the angle change caused by impedance will be maintained during the support phase. At the same time, because the raised terrain is only a local terrain, the impedance value is cleared at the end of the swing phase, so that the length of the leg that touches the obstacle is restored to the initial length, avoiding the inconsistency of the length of the four legs in subsequent movements. The control signal of the knee joint is as follows:

[0324]

[0325] The above impedance control can effectively adapt to the raised terrain, but the influence of leg length and step length changes on the swing amplitude must also be considered. Since the leg length reset time is reserved at the end of the stance phase, the front and back hip joints need to complete the stance phase swing before the reset time. The control signals of the front and back hip joints in the stance phase after correction are as follows:

[0326]

[0327] Where L sw —Leg length at the end of the swing phase.

[0328] Also considering that the expected step length is not achieved due to early landing, the control signal of the hip anterior and posterior joint swing phase in the next movement cycle is as follows:

[0329]

[0330] To facilitate calculation, the step length and time are simplified into a linear relationship in formula (7-7), and the proportion of the time of the trigger impedance to the entire swing phase time is used to approximate the step length that has been walked, so as to compensate for the step length in the next swing phase.

[0331] D. Single-leg stepping reflex exercise: The stepping reflex exercise can be planned into ① waiting for the ground contact, ② lifting the leg to leave, and ③ swinging forward to land. Fig.14 As shown in the figure, the process of stepping air reflex movement is:

[0332] S1: Ground contact waiting phase: To improve the robot's motion stability, when the foot end does not detect force at the end of the swing phase, the stepping reflex movement will not be triggered immediately. Instead, the ground contact waiting time will be increased. If the foot end still does not detect force during the waiting period, it is determined that the robot has stepped on the ground with one leg.

[0333] S2: Leg-lifting phase: When the robot triggers the stepping-in reflex, in order to avoid hitting an obstacle again, the robot must first keep the front and rear hip joints still and lift the knee joints upward to get the foot out of the pit.

[0334] S3: Forward swing and landing stage: After the foot leaves the pit, the robot swings forward at the hip joint and lands at the knee joint to cross the pit. In this process, the actual step length of the robot increases, so the leg length needs to be increased accordingly to ensure that the robot can stably contact the ground.

[0335] The overall reflex movement is decomposed into the movement of the knee joint. The reflex movement includes waiting for the ground contact, lifting the knee joint, lowering the ground, the support phase and resetting the leg length. The reflex movement model of the knee joint is as follows:

[0336]

[0337] Where E1, E2—changes in knee joint angle.

[0338] The overall reflex motion is then decomposed into hip anterior and posterior joint motion, which includes ground contact waiting, hip anterior and posterior joint immobility, forward swing over obstacles, support and back swing, and leg length reset. During the support phase, the robot's leg length changes again, so the swing amplitude change caused by the leg length change must be considered. The hip anterior and posterior joint reflex motion model is as follows:

[0339]

[0340] Where F1 is the change in the anterior-posterior hip joint angle.

[0341] The values ​​of T1, T2, and T3 are as follows:

[0342]

[0343] When crossing the pit, the obstacle leg takes an extra step of S4. Therefore, in order to ensure the symmetry of the gait, the step length should be adjusted in the next swing phase from S / 2 to S / 2-S4, as shown in Fig.15 shown.

[0344] The formula for the anterior-posterior hip joint in the next swing phase is as follows:

[0345]

[0346] Illustration of the reflex movement of stepping into the air, such as Fig.16 The quadruped robot needs to determine the parameters E1, E2, and F1 for the air-stepping reflex movement, where E1 and E2 are the knee joint rotation angles in the leg-lifting phase and the front-swing landing phase, respectively, and F1 is the hip front-back joint rotation angle in the front-swing landing phase.

[0347] This description assumes that the pit size does not exceed 100mm, so the robot's single leg swing forward again to cross the obstacle distance S4 = 110mm. Then F1 can be derived from formula (4-53):

[0348]

[0349] F1=θ h0 -θ h2 +θ h1 (8-8)

[0350] The rotation angle E1 is a settable angle and is considered a known quantity. In this paper, E1 is set to 0.1 rad.

[0351] The value of the rotation angle E2 can be derived from formula (8-10):

[0352]

[0353] E2=θ k3 -θ k0 +E1 (8-10)

[0354] In this embodiment, the motion control of the robot on multiple asymmetric complex roads is taken as an example. Obstacles with heights of 70 mm, 60 mm, and 40 mm, and a pit with a width of 100 mm are respectively set in front of the four legs of the robot. Fig.17 As shown, the robot's motion parameters are set to a motion speed of 200 mm / s, a gait period of 1.7 s, a foot lift height of 45 mm, and an intermittent Trot gait.

[0355] exist Fig.17 The force images of the left front leg and left hind leg in (b) and (c) show that the absolute value ratio of the force component in the forward direction to the vertical direction is greater than the static friction coefficient, which triggers the kicking reflex movement in the post-swing phase and the pre-swing phase, respectively. Fig.17 The force image of the right hind leg in (d) shows that no force is detected at the foot end, triggering the step-off reflex movement. Fig.17 In (e), the correlation ratio of the right front leg is lower than the static friction coefficient, thus triggering the right front leg stepping reflex movement. Fig.17 (a) and Fig.17 (f) By comparison, it can be seen that the fuselage does not have obvious yaw after amplitude correction.

[0356] The joint control signal curve of complex terrain obstacle crossing motion after amplitude correction is as follows: Fig.18 As shown in the figure. There are four groups of signals in the figure, representing the left front (LF), right front (RF), right rear (RH) and left hind leg (LH), where each leg is composed of three joint control signals, namely the hip lateral swing joint, the hip front and rear joint and the knee joint. As can be seen from the figure, at t = 2.57 seconds, the robot's left front leg triggered the swing back phase kick barrier reflex movement. At t = 3.15 seconds, the robot's left hind leg triggered the swing front phase kick barrier reflex movement. At t = 6.15 seconds, the robot's right hind leg triggered the stepping air reflex movement. At t = 6.98 seconds, the robot's right front leg triggered the swing back phase step barrier reflex movement.

[0357] The change of Euler angle of the robot's center of mass when crossing complex terrain obstacles, such as Fig.19 shown.

[0358] The center of mass displacement curve of the quadruped robot's complex terrain obstacle crossing motion, Fig. 20 , Fig.21 , Fig. 22 The displacement of the center of mass in the lateral, vertical and forward directions in the world coordinate system are shown respectively.

[0359] Seek Fig. 20 The difference between the mean value of the mass center displacement curve in the lateral direction and the initial mass center is 5.07 mm. Fig.21 It was observed that the maximum fluctuation of the robot's center of mass during the obstacle crossing process did not exceed 16 mm, and the vertical fluctuation rate of the center of mass was 2.1%. Fig. 22In the figure, the displacement speed of the center of mass of the robot in the forward direction is 187.5 mm / s, which is close to the expected speed.

[0360] comprehensive Figure 19-22 The analysis shows that even in the face of multiple asymmetric complex terrains, the robot can maintain the expected speed and move smoothly along the predetermined straight path after amplitude correction, significantly reducing the need for subsequent robot correction. In addition, the robot's center of mass displacement curve during continuous obstacle crossing also demonstrates the stability of the robot's movement and the effectiveness of the model.

[0361] The above description is only a preferred feasible embodiment of the present invention, and does not limit the scope of rights of the present invention. All equivalent changes made using the contents of the present specification and its drawings are included in the scope of rights of the present invention.

Claims

1. A quadruped robot stable obstacle crossing control method, characterized in that: The structure of the quadruped robot is a front elbow and rear knee joint, comprising the following steps: Step 1: Use the DH method to establish the kinematic model of the four legs of the quadruped robot and obtain the force direction of the foot end in the world fixed coordinate system; Step 2: Establish a CPG motion control model based on the Hopf oscillator, and control the rhythmic motion of the quadruped robot through the standard oscillation signal output by the CPG motion control model to achieve walking on a flat road; Step 3: Based on the robot's obstacle crossing requirements, an intermittent Trot gait with a load factor β of 0.7 is selected during the obstacle crossing period; Step 4: The single-leg phase of the quadruped robot is divided into three stages: the pre-swing phase, the post-swing phase, and the support phase. The pre-swing phase and the post-swing phase are separated by the midpoint of the single-leg swing phase. Combining the three types of obstacle-touching force states, namely, kicking obstacles, stepping on obstacles, and stepping into the air, it is concluded that the robot's obstacle-touching state is divided into four categories: kicking obstacles in the pre-swing phase, kicking obstacles in the post-swing phase, stepping on obstacles in the post-swing phase, and stepping into the air. The judgment formula for the single-leg phase stage of the quadruped robot is as follows: In the formula, sng i —The phase phase judgment sign function of the i-th leg; where sng i When the value is 1, it is the pre-swing phase, sng i When the value is -1, it is the swing phase, sng i When the value of is 0, it is the support phase; The force state of the foot touching the obstacle is judged as follows: The formulas for the robot's foot-end kicking and stepping on obstacles are as follows: Where F yi —The vertical force on the foot of the ith leg in the world fixed coordinate system; F xi —The forward force on the foot of the ith leg in the world fixed coordinate system; μ—static friction coefficient of the ground; If the function f(F xi , F yi ) is 1, it is in the kicking state; if f(F xi , F yi ) is -1, it is an obstacle state; if f(F xi , F yi ) is 0, and the motion phase is the transition from the robot swinging back phase to the support phase, and the foot-end three-dimensional force sensor does not detect any force, then the robot is in the foot-end empty state; Step 5: Perform robot reflex motion planning for each obstacle-touching state, and model the reflex motion of stepping on the obstacle after swinging. Specifically, introduce the impedance control method. The second-order differential form of impedance control is as follows: F e =Mg / n Where K d — expected stiffness matrix; B d — expected damping matrix; M d — expected inertia matrix; △F—deviation between actual force and expected force; F r —actual force; F e —Expectancy; n—number of supporting legs; M—the overall mass of the robot; Rewrite the above formula into a transfer function form, which belongs to a typical second-order system. It is tuned according to the second-order system, and the formula is as follows: Stiffness parameter K d It mainly affects the steady-state output value of the system, while the damping parameter B d and inertia parameter M d It only affects the dynamic process of the system; Planning parameters for knee joint angular velocity in CPG motion control model Take the derivative is considered as the relevant planning parameter of the knee joint angular acceleration; use The impedance stiffness is formed and the above formula is optimized into the following formula: Where k is the stiffness coefficient; b—Stiffness coefficient 2; —When the i-th leg triggers the impedance The value of Δθ ki —Impedance adjustment value; The input in the above formula is the vertical force component received by the foot end in the fixed world coordinate system, and the output is the angle change caused by impedance. At the end of the swing phase, the angle change caused by impedance will be maintained during the support phase. At the same time, because the raised terrain is only a local terrain, the impedance value is cleared at the end of the swing phase, so that the length of the leg that touches the obstacle is restored to the initial length; the control signal of the knee joint is as follows: Where θ ki — auxiliary angle of the knee joint at stage i; T—gait cycle; T st —Support item cycle; t—current time; The front and rear hip joints complete the swing of the stance phase before the reset time. The control signals of the front and rear hip joints in the stance phase after correction are as follows: Where L sw — leg length at the end of the swing phase; x0—the distance between the single-leg base coordinate system and the center of mass coordinate system; A h —Amplitude of hip anteroposterior motion; T sw —Oscillating term period; Taking into account the situation where the expected step length is not achieved due to early landing, the control signal of the hip anterior and posterior joint swing phase in the next movement cycle is as follows: Where t f —The moment of hitting an obstacle during the movement cycle; For ease of calculation, the step length and time are simplified into a linear relationship in the above formula, and the proportion of the time of the trigger impedance to the entire swing phase time is used to approximately estimate the step length that has been walked, so as to compensate for the step length in the next swing phase.

2. The quadruped robot stable obstacle crossing control method according to claim 1, characterized in that: The movement process of the kicking reflex in the pre-swing phase is: S1: Backward obstacle avoidance: When the foot detects a collision in the pre-swing phase, the knee joint is kept stationary and the hip joint swings backward to escape from the obstacle; S2: Leg lifting contraction: by swinging the hip joint backward and the knee joint upward, a scissor contraction is performed to lift the foot end to cross the obstacle; S3: Unchanged obstacle crossing: The hip joint swings forward and the knee joint does not move, thus completing the obstacle crossing action with one leg; S4: Landing back: By swinging the hip joints forward and the knee joints downward, perform a scissor-like extension movement so that the foot ends land smoothly in a basically vertical direction.

3. The method for controlling a quadruped robot to smoothly cross obstacles according to claim 2, characterized in that: The pre-swing phase barrier kicking reflex movement is decomposed, wherein the movement of the knee joint includes the knee joint remaining unchanged, contraction and lifting, the knee joint remaining unchanged over the obstacle and extension and falling back, and the knee joint control signal is as follows: Where t f —The moment of hitting an obstacle during the movement cycle; y f —The angle value of the knee joint when it hits the obstacle; x f —The angle value of the anterior and posterior hip joint when it hits the obstacle; t—current time; A1—change in knee joint angle; T1, T2, T3—Obstacle crossing stage time; T0—total obstacle crossing time; The kicking reflex movement in the pre-swing phase is decomposed, where the movement of the front and rear joints of the hip includes retreating to avoid obstacles, contracting and retreating, swinging forward and extending and swinging forward. The control signals of the front and rear joints of the hip are as follows: Where x f —The angle value of the anterior and posterior hip joint when it hits the obstacle; B1, B2, B3—changes in the anterior-posterior angle of the hip joint; The formulas for time T0, T1, T2, and T3 are as follows: T0=T sw +T kong =T / 2 Where T is the gait cycle; T sw —Oscillating term period; T kong —Each commutation transition time; A1 is the knee angle in the stage of lifting the foot to cross the obstacle, B1, B2, and B3 are the hip anterior and posterior angles in the stage of retreating after hitting the obstacle, the stage of lifting the leg and contracting, and the stage of falling back to the ground. The angle A1 needs to meet the following formula: Where L i —length of the leg at stage i; h z —Height of obstacle, h z =100mm; H0—the distance between the hip joint base and the ground; The rotation angle B1 is a set angle and is considered as a known quantity; The rotation angle B2 is derived from the following formula: B2=θ h1 +θ h2 -x f -B1 Where θ hi —Anterior and posterior auxiliary angle of the hip joint at stage i; x f —Anterior-posterior angle of the hip joint when hitting the obstacle; In the unchanged obstacle crossing stage, the knee joint remains unchanged, so it is easy to get θ h3 =θ h2 , so the corner B3 is derived as follows: B3=θ h1 +x f +B1+B2-θ h3 4. The method for controlling a quadruped robot to smoothly cross obstacles according to claim 3, characterized in that: The movement process of the kicking reflex after the swing phase is: S1: Obstacle detection: When the foot detects the obstacle-kicking state in the post-swing phase, the obstacle-kicking leg performs the post-swing phase obstacle-kicking reflex movement; S2: Backward obstacle avoidance: When the swing phase kicking reflex is triggered, the knee joint does not move, and the hip joints move backward to get rid of the obstacle; S3: Emergency landing: Due to time constraints, the robot cannot complete the obstacle crossing action, so the robot's foot lands urgently and enters the support phase in advance. At this time, the length of the leg that touches the obstacle changes. S4: Leg lifting and contraction: through the hip joint swinging back and forth and the knee joint lifting, the foot end is lifted by a scissor contraction action to facilitate the subsequent crossing of obstacles; S5: Unchanged obstacle crossing: The hip joint swings forward, the knee joint maintains the same angle, and completes the single-leg obstacle crossing action; S6: Falling back to the ground: The knee joint falls back to the ground, while the front and rear joints of the hip continue to swing forward, and the scissor-like extension is used to ensure that the foot ends fall smoothly in the vertical direction. At this time, the leg length of the leg that touches the obstacle changes again.

5. The quadruped robot stable obstacle crossing control method according to claim 4, characterized in that: The post-swing phase kicking reflex movement is decomposed, wherein the reflex movement of the knee joint in the first movement cycle includes the knee joint remaining unchanged, landing and lowering, and the support phase. The specific reflex movement formula is as follows: In the formula, y f —The angle value of the knee joint when it hits the obstacle; t f —The moment of hitting an obstacle during the movement cycle; T—gait cycle; In the second movement cycle, the reflex movement of the knee joint is decomposed into contraction and lifting, unchanged obstacle crossing, extension and fall, support phase and leg length reset. In the leg length reset phase, in order to prevent the continuous accumulation of leg length changes when continuously touching obstacles, the specific reflex movement formula is as follows: In the support phase of the reflex motion, the swing amplitude adjustment caused by the change of leg length is introduced. When the support phase leg length of the robot increases from L0 to L', the support phase swing amplitude is adjusted from A to L. h Adjust to A h ' to ensure stride consistency; The swing phase kicking reflex movement is decomposed, in which the hip front and back joint movement in its first cycle, the reflex action is decomposed into obstacle avoidance retreat, retreat landing and support phase. The specific reflex movement formula is as follows: x f —The angle value of the anterior and posterior hip joint when it hits the obstacle; In the second movement cycle of the anterior and posterior hip joints, the reflex action is decomposed into contraction and withdrawal, forward swing, extension forward swing, support phase and hip joint immobility. In the forward swing and extension forward swing phases, the change in step length caused by the landing phase of the previous cycle is considered. At the same time, during the support phase, the effect of leg length changes on the swing amplitude is considered, and the leg length needs to be reset at the end of the support phase. The specific reflex movement formula is as follows: G1=x f -D1-D2-D3+D4+D5 T1=(T sw -t f +T kong ) / 3 T2=2(T sw -t f +T kong ) / 3 Among them, C1, C2, and C3 are the knee joint rotation angles in the backward obstacle avoidance stage, emergency landing stage, and leg lifting and contraction stage respectively; D1, D2, D3, D4, and D5 are the hip anterior and posterior joint rotation angles in the backward obstacle avoidance stage, emergency landing stage, leg lifting and contraction stage, unchanged obstacle crossing stage, and falling back and landing stage respectively; the angles D1 and D2 are the set hip joint rotation angles, which are regarded as known quantities, and D1 and D2 are set to 0.1rad. The angle C1 is derived by the following formula: C1=θ k3 +y f θ h3 =θ h0 -(x f -D1-D2) Where θ ki — auxiliary angle of the knee joint at stage i; θ hi —Anterior and posterior auxiliary angle of the hip joint at stage i; Corner D3 is derived from the following formula: D3=θ h4 -θ h3 Where S3 is the step length of the previous motion cycle; The rotation angle C2 is derived from the following formula: C2=θ k3 -θ k4 The rotation angle D4 is derived from the following formula: D4=θ h4 -θ h5 The rotation angle D5 is derived from the following formula: D5=θ h5 -θ h6 The rotation angle C3 is derived from the following formula: C3=θ k6 -θ k5 =θ k6 -θ k4 6. The quadruped robot stable obstacle crossing control method according to claim 3 is characterized in that: The single-leg stepping reflex movement process is: S1: Ground contact waiting: When the foot end does not detect force at the end of the swing phase, the stepping reflex movement is not triggered immediately, but the ground contact waiting time is increased. If the foot end still does not detect force during the waiting period, it is determined that the robot has stepped on the ground with one leg; S2: Lifting the leg to escape: When the robot triggers the stepping reflex movement, in order to avoid hitting the obstacle again, the robot first keeps the front and rear hip joints still and lifts the knee joints to get the foot out of the pit terrain; S3: Forward swing and landing: After the foot leaves the pit, the hip joint swings forward and the knee joint drops, which increases the leg length and allows the foot to land on the ground, thus crossing the pit terrain.

7. The quadruped robot stable obstacle crossing control method according to claim 6, characterized in that: The single-leg stepping reflex movement modeling is specifically as follows: the single-leg stepping reflex movement is decomposed, and the knee joint reflex movement included includes ground contact waiting, knee joint lifting, landing and lowering, support phase and leg length reset. The knee joint reflex movement model is as follows: Where E1, E2—changes in knee joint angle; The single-leg stepping reflex motion is decomposed. The hip anterior and posterior joint reflex motion includes waiting for touchdown, immobilization of the hip anterior and posterior joints, swinging forward to overcome obstacles, swinging back after support, and resetting the leg length. During the support phase, the robot's leg length changes again. Therefore, considering the swing amplitude change caused by the leg length change, the hip anterior and posterior joint reflex motion model is as follows: Where F1 is the change in the joint angle of the anterior and posterior hip joint; x f —The angle value of the anterior and posterior hip joint when it hits the obstacle; The values ​​of T1, T2, and T3 are as follows: In the next swing phase, the step length is adjusted from S / 2 to S / 2-S4. The formula for the anterior and posterior hip joint in the next swing phase is as follows: E1 and E2 are the knee joint rotation angles in the leg lifting and separation phase and the front swing landing phase respectively, and F1 is the hip anterior-posterior joint rotation angle in the front swing landing phase.

Images