Skip to content

Industrial Robot using LQR controller to balance a linear inverted pendulum.

Notifications You must be signed in to change notification settings

wumin199/Industrial_Robot_Linear_Inverted_Pendulum

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation

To display the formula correctly, please view this page under Chrome with MathJax extension installed

Industrial Robot Linear Inverted Pendulum*

workflow

Status Input => Kalman Filter ==> Linear Inverted Pendulum Control(LQR) ==> Robot Planning ==> Robot & Inverted Pendulum Execution ==> Status Output

Kalman Filter

本案例中用到卡尔曼滤波的目的是,估算LQR的输入的$\dot{θ}$。因为我们没有倒立摆的摆动加速度测量传感器。而$\dot{x}$,是直接用TCP值计算的,因为编码器的精度很高。

Linear Inverted Pendulum Control(LQR)

Algorithm: LQR

直线一阶倒立摆系统研究_尹贻晓.caj
LQR(Linear quadratic regulator, 线性二次型调节器)的控制对象是现代控制理论中以状态空间的形式给出的线性系统,它的目标函数是对象状态和控制输入的二次型函数。状态反馈控制器K由权重矩阵Q和R唯一确定,而线性二次型调节器最优化设计就是要求设计处的状态反馈控制器K要使得二次型目标函数J取得最小值,所以参数Q和R的选择非常重要。

基于dSPACE仿真平台的一阶直线倒立摆控制研究_任玲
本章以系统全状态变量反馈为基本策略,设计LQR控制器(线性二次型最有控制算法)的基本思想为:在满足一定的性能指标要去下,利用状态反馈方法,再消耗能量金小的情况下,使得系统状态的动态性能最佳、状态误差最小。
LQR最优控制算法属于线性反馈的结构,在实际倒立摆控制系统中得到了广泛的应用,具有一定的鲁棒性,动态特性较好。

总结
LQR是能量最优算法,J包含两项=状态误差累计项+控制量的动态性能项目(具体见下文) 输入是全状态反馈输入量(即下文的大X),输出是倒立摆的控制量F,这个F可以让整个J取得最小值。虽然是能量最低,但是通过调试K参数,可以调试上升时间等。简单说就是针对J优化,构建汉密尔顿方程H,然后对H控制量u求偏导数,并令偏导数=0,即可求得控制量u(F),此时的J也是最小的。

Input: realTime x, $\dot{x}$, θ, $\dot{θ}$ and reference/aimed x0, $\dot{x}0$, θ0, $\dot{θ}0$ (to check later)

Output: F(or call it u)(optimal F to let J minimal)

Notice : F is proportional to acceleration

Ignore air resistance and cart floor friction, we get "cart-pole" model:

$$

\ddot{x} = \frac{(J + ml^2)F + (ml^2+J)mlsin(θ)(\dot{θ})^2-m^2l^2gsin(θ)cos(θ)}{(J+ml^2)(M+m)-m^2l^2cos(θ)} \ (1)

$$

$$

\ddot{θ} = -\frac{mlcos(θ)F+m^2l^2(\dot{θ})^2sin(θ)cos(θ)-(M+m)mglsin(θ)}{(J+ml^2)(M+m)-m^2l^2cos(θ)} \ (2)

$$

Let's say θ is small and linearize it as:

$\begin{cases}θ \ is \ small \ but \ not \ zero \\sin(θ)\approxθ\\cos(θ)\approx1\\(\dot{θ})^2\approx0\end{cases}$

back substitute the above assumptions to equation(1) and (2), we get

$$

\ddot{x} = \frac{(J + ml^2)F -m^2l^2gθ}{(J+ml^2)(M+m)-m^2l^2} \ (3)

$$

$$

\ddot{θ} = -\frac{mlF-(M+m)mglθ}{(J+ml^2)(M+m)-m^2l^2} \ (4)

$$

if $J=\frac{1}{3}ml^2$, then we can get

$$

\ddot{x} = \frac{4F-3mgθ}{4M+m} \ (5)

$$

$$

\ddot{θ} = \frac{3(M+m)gθ-3F}{(4M+m)l} \ (6)

$$

Linear Inverted Pendulum State Function:

$\begin{cases}\dot{X} = AX+BU\\Y = CX+DU\end{cases}$

where:

$\begin{cases}X=[x, \dot{x},θ, \dot{θ}]\\\dot{X}=[ \dot{x}, \ddot{x},\dot{θ}, \ddot{θ}]\\Y=[x, θ]\\U = u \ or \ F\end{cases}$

if we use equations (3) and (4), we can get State Fuctions:

$$ \dot{X}= \begin{bmatrix} \dot{x} \\ \ddot{x} \\ \dot{θ} \\ \ddot{θ} \end{bmatrix} =

\begin{bmatrix} 0 & 1 & 0 & 0 \\ 0 & 0 & \frac{-m^2l^2g}{(J+ml^2)(M+m)-m^2l^2} & 0 \\ 0 & 0 & 0 & 1 \\ 0 & 0 & \frac{(M+m)mgl}{(J+ml^2)(M+m)-m^2l^2} & 0 \end{bmatrix}

\begin{bmatrix} x \\ \dot{x} \\ θ \\ \dot{θ} \end{bmatrix} +

\begin{bmatrix} 0 \\ \frac{J+ml^2}{(J+ml^2)(M+m)-m^2l^2} \\ 0 \\ \frac{-ml}{(J+ml^2)(M+m)-m^2l^2} \end{bmatrix}F \ (7) $$

$$ Y= \begin{bmatrix} x \\ θ \end{bmatrix}=

\begin{bmatrix} 1 & 0 & 0 & 0 \\ 0 & 0 & 1 & 0 \end{bmatrix}

\begin{bmatrix} x \\ \dot{x} \\ θ \\ \dot{θ}+ \end{bmatrix} +

\begin{bmatrix} 0 \\ 0 \end{bmatrix}F \ (8) $$

if we use equations (5) and (6), we can get State Fuctions:

$$

\dot{X}= \begin{bmatrix} \dot{x} \\ \ddot{x} \\ \dot{θ} \\ \ddot{θ} \end{bmatrix} =

\begin{bmatrix} 0 & 1 & 0 & 0 \\ 0 & 0 & \frac{-3mg}{4M+m} & 0 \\ 0 & 0 & 0 & 1 \\ 0 & 0 & \frac{3(M+m)gθ}{(4M+m)l} & 0 \end{bmatrix}

\begin{bmatrix} x \\ \dot{x} \\ θ \\ \dot{θ} \end{bmatrix} +

\begin{bmatrix} 0 \\ \frac{4}{4M+m} \\ 0 \\ \frac{-3}{(4M+m)l} \end{bmatrix}F \ (9)

$$

$$ Y= \begin{bmatrix} x \\ θ \end{bmatrix}=

\begin{bmatrix} 1 & 0 & 0 & 0 \\ 0 & 0 & 1 & 0 \end{bmatrix}

\begin{bmatrix} x \\ \dot{x} \\ θ \\ \dot{θ}+ \end{bmatrix} +

\begin{bmatrix} 0 \\ 0 \end{bmatrix}F \ (10) $$

In program, we use equations (7) and (8).

Robot Planning

LQR计算出来的是u或者说加速度,需要积分到p(xyz),然后在笛卡尔直线约束下,下发点位给机器人执行。中间涉及到的是:逆解为J1J6,然后笛卡尔约束下的下发J1J6。

About

Industrial Robot using LQR controller to balance a linear inverted pendulum.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published