Adaptive Sliding Mode Control of 2 DOF Robotic Arm
2DOF Robot Arm Model on Simulink
$$
\ddot{x} = H^{-1} (\tau - C \dot{x} - F_c sgn(\dot{x}) - V \dot{x})
$$
Where:
$$
\tau = \begin{bmatrix}
\tau_1 & \tau_2
\end{bmatrix}^T
$$
$$
x = \begin{bmatrix}
x_1 & x
\end{bmatrix}^T
$$
$$
\ddot{x} = \frac{d}{dx}\dot{x} = \frac{d^2}{dx}x
$$
$$
H = \begin{bmatrix}
M_{11} & M_{12} \\
M_{21} & M_{22}
\end{bmatrix}
$$
where:
$$ M_{11} = a_1 + 2 a_3 cos(x_2) - 2 a_4 sin(x_2) $$
$$ M_{12} = M_{21} = a_2 + a_3 cos(x_2) + a_4 cos(x_2) $$
$$ M_{22} = a_2 $$
$$
C = \begin{bmatrix}
-c\dot{x}_2 & -c(\dot{x}_1 + \dot{x_2}) \\
c\dot{x}_1 & 0
\end{bmatrix}
$$
where:
$$ c = a_3 sin(x_2) - a_4 cos(x_2) $$
Estimated parameters:
$$
a = \begin{bmatrix}
a_1 \\
a_2 \\
a_3 \\
a_4
\end{bmatrix} = \begin{bmatrix}
3.3 \\
0.97 \\
1.4 \\
0.6
\end{bmatrix}
$$
Friction terms:
$$
F_c = \begin{bmatrix}
F_{c1} & 0 \\
0 & F_{c2}
\end{bmatrix}
$$
where:
$$ F_{c1} = F_{c2} = 5$$
and
$$
V = \begin{bmatrix}
V_{1} & 0 \\
0 & V_{2}
\end{bmatrix}
$$
where:
$$ V_1 = 5.5 $$
and
$$ V_2 = 2.7 $$
Given initial conditions:
$$\begin{aligned}
x_1 = 0.3 \space rad \\
x_2 = 0.5 \space rad
\end{aligned}$$
$$
Y = \begin{bmatrix}
y_{11} & y_{12} & y_{13} & y_{14} \\
y_{21} & y_{22} & y_{23} & y_{24}
\end{bmatrix}
$$
where:
$$\begin{aligned}
y_{11} &= \ddot{x}_{r1} \\\
y_{12} &= \ddot{x}_{r2} \\\
y_{21} &= 0 \\\
y_{22} &= \ddot{x}_{r1} + \ddot{x}_{r2} \\\
y_{13} &= (2\ddot{x}_{r1} + \ddot{x}_{r2}) \cos(x_{2}) - (\dot{x}_{2} \dot{x}_{r1} + \dot{x}_{1} \dot{x}_{r2} + \dot{x}_{2} \dot{x}_{r2}) \sin(x_{2}) \\\
y_{14} &= (2\ddot{x}_{r1} + \ddot{x}_{r2}) \sin(x_{2}) - (\dot{x}_{2} \dot{x}_{r1} + \dot{x}_{1} \dot{x}_{r2} + \dot{x}_{2} \dot{x}_{r2}) \cos(x_{2}) \\\
y_{23} &= \ddot{x}_{r1} \cos(x_2) + \dot{x}_1 \dot{x}_{r1} \sin(x_2) \\\
y_{24} &= \ddot{x}_{r1} \sin(x_2) + \dot{x}_1 \dot{x}_{r1} \cos(x_2)
\end{aligned}$$
Estimated values of unmodeled dynamics:
$$ b^{-1} = 5 H $$
$$ f = 5 (\ddot{x}_d - 2 \lambda \dot{e} - \lambda^{2} e) $$
Adaptive sliding surface definition
$$
s = \dot{e} + 2 \lambda e + \lambda^{2} \int^{t}_{0}{e}
$$
Given control law:
$$ u = u_a + u_{eq} $$
such that:
$$ u_a = \hat{b}^{-1} (Y \hat{a} - k_D s) $$
$$ u_{eq} = \hat{b}^{-1} (\ddot{x}_d - \hat{f} - 2 \lambda \dot{e} - \lambda^{2} e) $$
The update of estimated parameters:
$$ \dot{\hat{a}} = - \Gamma Y^{T} s $$
therefore:
$$ a = \int{\dot{\hat{a}}}\space dt $$
Used limited output integrators in the arm model to be between $$ -pi $$ and $$ +pi $$.
Used a small regularization term:
epsilon = 1e-6; % Small regularization term
H_reg = H + epsilon * eye(size(H));
H_inv = inv(H_reg);
In addition to using pseudoinverse if matrix is close to singular or badly scaled.
Handling NaN and inf Values
Implemented an algorithm to detect NaN and inf values, by using these replacements:
nan_replacement = eps; % A very small positive number
inf_replacement = realmax; % The largest positive floating-point number
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent