A combined active and passive method for online foothold generation and adjustment of bipedal robots

Abstract

This invention discloses a method for online footing point generation and adjustment of a bipedal robot that combines active and passive methods. The method includes: determining the planned zero-moment point trajectory and the planned center-of-mass trajectory; designing a virtual inverted pendulum center-of-mass tracking controller; controlling the desired zero-moment point trajectory to ensure the robot's desired center-of-mass trajectory tracks the actual zero-moment point position and calculates external forces, applying them to the virtual inverted pendulum for state updates; optimizing the desired footing point position based on the estimated center-of-mass trajectory for the next two steps and the planned footing point position through online ankle trajectory tracking control, ensuring the robot's desired ankle position follows the calculated desired joint angle, which is then applied to the bipedal robot. This invention achieves a larger and more flexible footing point adjustment effect while ensuring real-time computation, resulting in better disturbance suppression and balance restoration.

Description

Technical Field

[0001] This invention belongs to the field of bipedal robot technology, specifically relating to a method for generating and adjusting the online footing of a bipedal robot that combines active and passive methods. Background Technology

[0002] To enable robots to walk stably, a stable sequence of footholds (p) and a center-of-mass trajectory (CoM) are required, along with various control mechanisms to maintain balance under external disturbances. Early trajectory planning methods, based on template models, significantly reduced computational load and enabled offline generation of stable trajectories. However, when the robot encounters external disturbances, additional control mechanisms are needed to adjust the footholds; these methods are termed offline trajectory generation plus online control. With the development of control and computer technologies, numerous control methods based on whole-body motion control and online foothold generation have emerged; these methods are termed online trajectory generation. Furthermore, compliant control mechanisms are crucial for bipedal robots to better absorb external disturbances and reduce the burden on balance control.

[0003] Existing methods combining offline trajectory planning and online control lack sufficient balance adjustment capabilities when dealing with large impact disturbances. This is because the motion generated by this method always revolves around the original trajectory, failing to produce significant adjustment amounts. Consequently, the landing adjustment amount is small and inflexible when facing large impacts. Existing online landing point planning methods based on whole-body motion control are only applicable to torque-controlled robots, not position-controlled robots, limiting their scope of use. This is because such methods cause jumps in the desired trajectory of the center of mass when adjusting the landing point, which position-controlled robots cannot withstand. Existing methods such as compliant drive control are effective in absorbing external impacts and improving robot stability, but they are not unified or integrated with various landing point control algorithms. Therefore, the compliant adjustment amount generated by them affects the landing point control effect. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for generating and adjusting the online footing of a bipedal robot that combines active and passive approaches. This method enables the robot to absorb some of the disturbance through a compliant response of its torso when dealing with external impacts and disturbances. At the same time, it can update the footing position online, ensuring real-time calculation while producing a larger and more flexible footing adjustment effect, thus achieving better disturbance suppression and balance restoration.

[0005] The present invention achieves the above-mentioned technical objectives through the following technical means.

[0006] A method for generating and adjusting the online foothold of a bipedal robot that combines active and passive methods:

[0007] Perform gait settings to determine the planned zero-moment point trajectory Pz PG And the planned centroid trajectory P c PG ;

[0008] Design a virtual inverted pendulum center-of-mass tracking controller to control the desired zero-torque point trajectory P. z ref The desired centroid trajectory P of the robot. c ref Tracking the centroid trajectory P c PG Based on the actual zero torque point position P z sen And the expected zero-moment point trajectory P z ref Calculate the external forces and apply them to the virtual inverted pendulum to update the state;

[0009] According to the expected centroid trajectory P c ref Estimate the centroid trajectory P within the next two steps. c est ;

[0010] From the estimated centroid trajectory P c est Planning the centroid trajectory P c PG and the planned landing site location Optimize the desired landing point location By using online ankle trajectory tracking control, the robot can achieve the desired ankle position. Follow the expected landing point

[0011] Using inverse kinematics, based on the desired ankle position And the expected centroid trajectory P c ref Calculate the desired joint angle q cmd And it applies to bipedal robots.

[0012] A further technical solution is that the virtual inverted pendulum center-of-mass tracking controller is:

[0013]

[0014] Where: P z ref K is the zero torque point of the virtual inverted pendulum. cp Here, ω is the feedback coefficient, and ω is the natural frequency of the virtual inverted pendulum. For the desired center-of-mass velocity, To plan the center of mass velocity.

[0015] A further technical solution involves using the following formula for updating the state of a virtual inverted pendulum:

[0016]

[0017] in: The desired center of mass acceleration.

[0018] A further technical solution is that the estimated centroid trajectory P c est Specifically, it refers to the state of the center of mass. As the initial state, let P Step To estimate the centroid state of the landing point within the next two steps, the centroid state is determined by substituting the following formula into the remaining time T of the current step. span The calculation yielded:

[0019]

[0020] in: Let be the expected centroid locus at time t. To determine the location of the zero torque point in this step, The initial value of the planning centroid at the current moment. Let be the initial value of the expected centroid at the current moment, and:

[0021]

[0022]

[0023]

[0024]

[0025] A further technical solution is that the desired landing point location satisfies:

[0026]

[0027] Where: P Step Let Y be the landing point, and let Y represent the divergence component of the motion. PG denoted by , where W represents the divergent component of the planned motion, and W represents the optimization weight coefficient.

[0028] A further technical solution is that the planned zero-moment point trajectory P z PG Including the zero-moment point trajectory in the x-direction Locus of zero moment point in the y-direction and:

[0029]

[0030] The beneficial effects of this invention are as follows:

[0031] (1) The online landing point optimization of the present invention provides an active balance adjustment means for the robot, and the disturbance compliance effect reflected by the virtual inverted pendulum state update provides a passive balance adjustment means for the robot. By adopting a combination of active and passive landing point adjustment control, the landing point control and the torso compliance control are unified, which can resist greater external disturbances and achieve better balance adjustment effect.

[0032] (2) This invention introduces a virtual inverted pendulum, which combines foot point control and torso compliance control. When adjusting the foot point, there is no change in the center of mass trajectory. It can realize online foot point planning and adjustment algorithms on robots with position control, which is flexible and takes into account both real-time performance and flexibility. Attached Figure Description

[0033] Figure 1 This is a block diagram of the method for generating and adjusting the online landing point of a bipedal robot that combines active and passive methods, as described in this invention. Detailed Implementation

[0034] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but the scope of protection of the present invention is not limited thereto.

[0035] like Figure 1 As shown, the present invention provides a method for generating and adjusting the online footing point of a bipedal robot using a combination of active and passive methods, comprising:

[0036] (1) Set the gait according to the instructions from the upper level to obtain the planned landing point position. optimization Get the desired landing point location This leads to the planned zero-moment point (ZMP) trajectory P. z PG Planning the centroid (CoM) trajectory P c PG The upper-level instructions are user-defined parameters, including the stepping cycle, step width, and step length.

[0037] The robot walks by alternating support from its left and right feet. When the left foot is supporting the robot, the ZMP (Zero-Minute Motion) trajectory needs to be planned directly below the robot's left ankle; when the right foot is supporting the robot, the ZMP trajectory needs to be planned directly below the robot's right ankle. Therefore, vector operations can be used to easily and quickly plan the ZMP trajectory for alternating left and right foot support. Define the vector:

[0038] V s =[1 1 ... 1] T ∈R n (1)

[0039] in, It is the duration of the current step, Tc It is the robot's control cycle;

[0040] Then, the desired landing location is set. The future ZMP trajectory can be calculated using the following formula:

[0041]

[0042] Where, N sp Support signals: 1 represents left foot support, -1 represents right foot support, L s,wid L is the length of a step. s,len Let N be the step width; thus, the ZMP trajectory in the x-direction for the first N steps. ZMP trajectory in the y-direction It can be represented as:

[0043]

[0044] Subsequently, based on and The planned centroid trajectory for the first N steps is calculated using online centroid generation methods such as predictive control (the specific calculation process is based on existing technology).

[0045] (2) Establish a virtual inverted pendulum model to simulate the robot's motion state, and then design a virtual inverted pendulum center-of-mass tracking controller to control the desired ZMP trajectory P. z ref This enables the robot to achieve its desired CoM trajectory P. c ref Capable of tracking and planning CoM trajectory P c PG .

[0046] (3) Based on the actual ZMP position P z sen And the expected ZMP trajectory P z ref The external forces are calculated and applied to a virtual inverted pendulum for iterative updates to achieve a compliant effect; the process of calculating the external forces is an existing technology.

[0047] Design of a virtual inverted pendulum center-of-mass tracking controller and method for updating the state of the virtual inverted pendulum's center of mass:

[0048] Traditional online landing point generation algorithms regenerate the centroid trajectory based on the new landing point, leading to abrupt changes in the centroid trajectory. The key to ensuring the robot's centroid trajectory doesn't change abruptly after adjusting the landing point is to establish a virtual inverted pendulum model. Subsequently, feedback control is needed to stabilize the virtual inverted pendulum model. The design of a virtual inverted pendulum centroid tracking controller is as follows:

[0049]

[0050] Among them, P z ref It is a virtual inverted pendulum ZMP, which is then executed by the robot as the expected ZMP; K cp For feedback coefficients; Let g be the natural frequency of the virtual inverted pendulum, g be the acceleration due to gravity, and z be the natural frequency of the pendulum. c The height of the center of mass of the virtual inverted pendulum;

[0051] The dynamic update of the virtual inverted pendulum can be expressed by the following equation:

[0052]

[0053] Based on the deviation between the actual ZMP position and the desired ZMP position, an external disturbance is introduced into the virtual inverted pendulum. Through the change in the acceleration of the virtual inverted pendulum's center of mass, its center of mass exhibits a compliant effect to the external disturbance.

[0054]

[0055] in, This refers to the amount of external disturbance. External disturbance feedback coefficient;

[0056] Will Effect on Updating formula (5), we can obtain the dynamic update formula for the virtual inverted pendulum with external disturbance:

[0057]

[0058] Among them, P z sen The robot's actual ZMP position can be calculated from the data read back from the force sensor.

[0059] (4) Based on the generated expected CoM trajectory P c ref Estimate the estimated CoM trajectory P within the next two steps c est ;

[0060] Method for estimating the center of mass state of a virtual inverted pendulum:

[0061] When optimizing the next landing point, it is necessary to consider the current state of the virtual inverted pendulum's center of mass and the desired ZMP position P. z refEstimate the centroid state after executing the next step. This estimation can be divided into two parts: first, estimating the centroid state at the next step; and second, estimating the centroid state within the next two steps. In practical applications, due to the geometric limitations of the robot's feet, the robot cannot achieve P beyond the foot's support range. z ref Therefore, P needs to be added. z ref The scope is limited; at the same time, P z ref The state of the center of mass will change due to the influence of the virtual inverted pendulum's center-of-mass state control, and is a time-varying quantity; the above two points will affect the estimation of the center-of-mass state. The analytical solution for the expected ZMP position at time t can be expressed as:

[0062]

[0063] Subsequently, the planned centroid trajectory P at time t c PG And the expected centroid trajectory P c ref The analytical solution can be expressed as:

[0064]

[0065] Establish boundary conditions:

[0066]

[0067] in, The initial value of the planned CoM at the current moment is obtained from the predictive control. The expected initial value of CoM at the current moment is obtained from the formula for updating the state of the center of mass of the virtual inverted pendulum.

[0068] Given the ZMP location for this step of planning Combining equations (7), (8), and (9), the analytical solution for the state of the center of mass of the virtual inverted pendulum can be expressed as:

[0069]

[0070] in:

[0071]

[0072]

[0073]

[0074] Substituting equation (11) into the remaining time T for this step span The state of the centroid in the next step can then be calculated. Then, starting with X as the initial state and P... Step Estimate the centroid state of the landing point within the next two steps.

[0075]

[0076] Among them, t n This indicates any point in time from the moment the foot lands on the ground until two steps in the future;

[0077] make:

[0078] This is for use when optimizing the landing point later, n = 1, 2, 3...N.

[0079] (5) Based on the estimated CoM trajectory P c est Planning CoM trajectory P c PG and the planned landing site location Optimize the desired landing point location The goal is to make P c est Try to follow P c PG ;

[0080] The methods for optimizing the desired landing point and for ankle trajectory tracking are detailed below:

[0081] After the next foot lands, N sampling points are taken between the next two steps to optimize the landing point. The motion divergence component (DCM) between the 1st and Nth sampling points can be expressed as:

[0082]

[0083] The centroid state is mapped to the DCM using matrix C = [1 1 / ω], and the planned DCM values ​​at the sample points are considered:

[0084]

[0085] Design and optimize the cost function, and calculate the optimal landing point.

[0086]

[0087] Where: W represents the optimization weight coefficient;

[0088] Equation (14) can be written in QP (quadratic programming) form:

[0089]

[0090] in:

[0091] Solving for the results

[0092] (6) By using online ankle trajectory tracking control (existing technology), the desired ankle position of the robot is achieved. Follow the expected landing point

[0093] (7) Based on the desired ankle position And the expected centroid trajectory P c ref The desired joint angle q is calculated using inverse kinematics (existing technology). cmd And it applies to bipedal robots.

[0094] The embodiments described above are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments. Any obvious improvements, substitutions or modifications that can be made by those skilled in the art without departing from the essence of the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for generating and adjusting the online foothold of a bipedal robot that combines active and passive methods, characterized in that: Perform gait settings to determine the planned zero-moment point trajectory. and planning the centroid trajectory ; Design a virtual inverted pendulum center-of-mass tracking controller to control the trajectory of the desired zero-torque point. To make the robot's desired centroid trajectory Tracking the centroid trajectory of the planning Based on the actual zero torque point location And the expected zero moment point trajectory Calculate the external forces and apply them to the virtual inverted pendulum to update the state; Based on the expected centroid trajectory Estimate the centroid trajectory within the next two steps. ; From the estimated centroid trajectory Planning the centroid trajectory and the planned landing site location Optimize the desired landing point location By using online ankle trajectory tracking control, the robot can achieve the desired ankle position. Follow the expected landing point ; Using inverse kinematics, based on the desired ankle position And the expected centroid trajectory Calculate the desired joint angle And it applies to bipedal robots; The virtual inverted pendulum center-of-mass tracking controller is: in: It is the zero torque point of the virtual inverted pendulum. Here, ω is the feedback coefficient, and ω is the natural frequency of the virtual inverted pendulum. For the desired centroid velocity, To plan the centroid speed; The estimated centroid trajectory Specifically, it refers to the state of the center of mass. As the initial state, with To estimate the centroid state of the landing point over the next two steps, the centroid state is calculated by substituting the following formula into the remaining time of the current step. The calculation yielded: in: Let be the expected centroid locus at time t. To determine the location of the zero torque point in this step, The initial value of the planning centroid at the current moment. Let be the initial value of the expected centroid at the current moment, and: 。 2. The method for generating and adjusting the online footing point of a bipedal robot combining active and passive methods according to claim 1, characterized in that, The formula for updating the state of the virtual inverted pendulum is: in: To achieve the desired center of mass acceleration, This represents the external disturbance feedback coefficient.

3. The method for generating and adjusting the online footing point of a bipedal robot combining active and passive methods according to claim 1, characterized in that, The desired landing point location satisfies: in: Let Y be the landing point, and let Y represent the divergence component of the motion. PG denoted by , where W represents the divergence component of the planned motion, and W represents the optimization weight coefficient.

4. The method for generating and adjusting the online footing point of a bipedal robot combining active and passive methods according to claim 1, characterized in that, The planned zero-moment point trajectory Including the zero-moment point trajectory in the x-direction Locus of zero moment point in the y-direction ,and: 。

Images