Two joint arm to point control
This is two joint arm to a point control simulation.
This is a interactive simulation.
You can set the goal position of the end effector with left-click on the plotting area.
Inverse Kinematics for a Planar Two-Link Robotic Arm
A classic problem with robotic arms is getting the end-effector, the mechanism at the end of the arm responsible for manipulating the environment, to where you need it to be. Maybe the end-effector is a gripper and maybe you want to pick up an object and maybe you know where that object is relative to the robot - but you cannot tell the end-effector where to go directly. Instead, you have to determine the joint angles that get the end-effector to where you want it to be. This problem is known as inverse kinematics.
Credit for this solution goes to: https://robotacademy.net.au/lesson/inverse-kinematics-for-a-2-joint-robot-arm-using-geometry/
First, let’s define a class to make plotting our arm easier.
%matplotlib inline
from math import cos, sin
import numpy as np
import matplotlib.pyplot as plt
class TwoLinkArm:
def __init__(self, joint_angles=[0, 0]):
self.shoulder = np.array([0, 0])
self.link_lengths = [1, 1]
self.update_joints(joint_angles)
def update_joints(self, joint_angles):
self.joint_angles = joint_angles
self.forward_kinematics()
def forward_kinematics(self):
theta0 = self.joint_angles[0]
theta1 = self.joint_angles[1]
l0 = self.link_lengths[0]
l1 = self.link_lengths[1]
self.elbow = self.shoulder + np.array([l0*cos(theta0), l0*sin(theta0)])
self.wrist = self.elbow + np.array([l1*cos(theta0 + theta1), l1*sin(theta0 + theta1)])
def plot(self):
plt.plot([self.shoulder[0], self.elbow[0]],
[self.shoulder[1], self.elbow[1]],
'r-')
plt.plot([self.elbow[0], self.wrist[0]],
[self.elbow[1], self.wrist[1]],
'r-')
plt.plot(self.shoulder[0], self.shoulder[1], 'ko')
plt.plot(self.elbow[0], self.elbow[1], 'ko')
plt.plot(self.wrist[0], self.wrist[1], 'ko')
Let’s also define a function to make it easier to draw an angle on our diagram.
from math import sqrt
def transform_points(points, theta, origin):
T = np.array([[cos(theta), -sin(theta), origin[0]],
[sin(theta), cos(theta), origin[1]],
[0, 0, 1]])
return np.matmul(T, np.array(points))
def draw_angle(angle, offset=0, origin=[0, 0], r=0.5, n_points=100):
x_start = r*cos(angle)
x_end = r
dx = (x_end - x_start)/(n_points-1)
coords = [[0 for _ in range(n_points)] for _ in range(3)]
x = x_start
for i in range(n_points-1):
y = sqrt(r**2 - x**2)
coords[0][i] = x
coords[1][i] = y
coords[2][i] = 1
x += dx
coords[0][-1] = r
coords[2][-1] = 1
coords = transform_points(coords, offset, origin)
plt.plot(coords[0], coords[1], 'k-')
Okay, we now have a TwoLinkArm class to help us draw the arm, which we’ll do several times during our derivation. Notice there is a method called forward_kinematics - forward kinematics specifies the end-effector position given the joint angles and link lengths. Forward kinematics is easier than inverse kinematics.
arm = TwoLinkArm()
theta0 = 0.5
theta1 = 1
arm.update_joints([theta0, theta1])
arm.plot()
def label_diagram():
plt.plot([0, 0.5], [0, 0], 'k--')
plt.plot([arm.elbow[0], arm.elbow[0]+0.5*cos(theta0)],
[arm.elbow[1], arm.elbow[1]+0.5*sin(theta0)],
'k--')
draw_angle(theta0, r=0.25)
draw_angle(theta1, offset=theta0, origin=[arm.elbow[0], arm.elbow[1]], r=0.25)
plt.annotate("$l_0$", xy=(0.5, 0.4), size=15, color="r")
plt.annotate("$l_1$", xy=(0.8, 1), size=15, color="r")
plt.annotate(r"$\theta_0$", xy=(0.35, 0.05), size=15)
plt.annotate(r"$\theta_1$", xy=(1, 0.8), size=15)
label_diagram()
plt.annotate("Shoulder", xy=(arm.shoulder[0], arm.shoulder[1]), xytext=(0.15, 0.5),
arrowprops=dict(facecolor='black', shrink=0.05))
plt.annotate("Elbow", xy=(arm.elbow[0], arm.elbow[1]), xytext=(1.25, 0.25),
arrowprops=dict(facecolor='black', shrink=0.05))
plt.annotate("Wrist", xy=(arm.wrist[0], arm.wrist[1]), xytext=(1, 1.75),
arrowprops=dict(facecolor='black', shrink=0.05))
plt.axis("equal")
plt.show()
It’s common to name arm joints anatomically, hence the names shoulder, elbow, and wrist. In this example, the wrist is not itself a joint, but we can consider it to be our end-effector. If we constrain the shoulder to the origin, we can write the forward kinematics for the elbow and the wrist.
Since the wrist is our end-effector, let’s just call its coordinates \(x\) and \(y\). The forward kinematics for our end-effector is then
A first attempt to find the joint angles \(\theta_0\) and \(\theta_1\) that would get our end-effector to the desired coordinates \(x\) and \(y\) might be to try solving the forward kinematics for \(\theta_0\) and \(\theta_1\), but that would be the wrong move. An easier path involves going back to the geometry of the arm.
from math import pi
arm.plot()
label_diagram()
plt.plot([0, arm.wrist[0]],
[0, arm.wrist[1]],
'k--')
plt.plot([arm.wrist[0], arm.wrist[0]],
[0, arm.wrist[1]],
'b--')
plt.plot([0, arm.wrist[0]],
[0, 0],
'b--')
plt.annotate("$x$", xy=(0.6, 0.05), size=15, color="b")
plt.annotate("$y$", xy=(1, 0.2), size=15, color="b")
plt.annotate("$r$", xy=(0.45, 0.9), size=15)
plt.annotate(r"$\alpha$", xy=(0.75, 0.6), size=15)
alpha = pi-theta1
draw_angle(alpha, offset=theta0+theta1, origin=[arm.elbow[0], arm.elbow[1]], r=0.1)
plt.axis("equal")
plt.show()
The distance from the end-effector to the robot base (shoulder joint) is \(r\) and can be written in terms of the end-effector position using the Pythagorean Theorem.
\(r^2\) = \(x^2 + y^2\)
Then, by the law of cosines, \(r\)2 can also be written as:
\(r^2\) = \(l_0^2 + l_1^2 - 2l_0l_1\cos(\alpha)\)
Because \(\alpha\) can be written as \(\pi - \theta_1\), we can relate the desired end-effector position to one of our joint angles, \(\theta_1\).
\(x^2 + y^2\) = \(l_0^2 + l_1^2 - 2l_0l_1\cos(\alpha)\)
\(x^2 + y^2\) = \(l_0^2 + l_1^2 - 2l_0l_1\cos(\pi-\theta_1)\)
\(2l_0l_1\cos(\pi-\theta_1) = l_0^2 + l_1^2 - x^2 - y^2\)
\(\cos(\theta_1) = \frac{x^2 + y^2 - l_0^2 - l_1^2}{2l_0l_1}\)
which leads us to an equation for \(\theta_1\) in terms of the link lengths and the desired end-effector position!
\(\theta_1 = \cos^{-1}(\frac{x^2 + y^2 - l_0^2 - l_1^2}{2l_0l_1})\)
This is actually one of two possible solutions for \(\theta_1\), but we’ll ignore the other possibility for now. This solution will lead us to the “arm-down” configuration of the arm, which is what’s shown in the diagram. Now we’ll derive an equation for \(\theta_0\) that depends on this value of \(\theta_1\).
from math import atan2
arm.plot()
plt.plot([0, arm.wrist[0]],
[0, arm.wrist[1]],
'k--')
p = 1 + cos(theta1)
plt.plot([arm.elbow[0], p*cos(theta0)],
[arm.elbow[1], p*sin(theta0)],
'b--', linewidth=5)
plt.plot([arm.wrist[0], p*cos(theta0)],
[arm.wrist[1], p*sin(theta0)],
'b--', linewidth=5)
beta = atan2(arm.wrist[1], arm.wrist[0])-theta0
draw_angle(beta, offset=theta0, r=0.45)
plt.annotate(r"$\beta$", xy=(0.35, 0.35), size=15)
plt.annotate("$r$", xy=(0.45, 0.9), size=15)
plt.annotate(r"$l_1sin(\theta_1)$",xy=(1.25, 1.1), size=15, color="b")
plt.annotate(r"$l_1cos(\theta_1)$",xy=(1.1, 0.4), size=15, color="b")
label_diagram()
plt.axis("equal")
plt.show()
Consider the angle between the displacement vector \(r\) and the first link \(l_0\); let’s call it \(\beta\). If we extend the first link to include the component of the second link in the same direction as the first, we form a right triangle with components \(l_0+l_1cos(\theta_1)\) and \(l_1sin(\theta_1)\), allowing us to express \(\beta\) as
\(\beta = \tan^{-1}(\frac{l_1\sin(\theta_1)}{l_0+l_1\cos(\theta_1)})\)
We now have an expression for this angle \(\beta\) in terms of one of our arm’s joint angles. Now, can we relate \(\beta\) to \(\theta_0\)? Yes!
arm.plot()
label_diagram()
plt.plot([0, arm.wrist[0]],
[0, arm.wrist[1]],
'k--')
plt.plot([arm.wrist[0], arm.wrist[0]],
[0, arm.wrist[1]],
'b--')
plt.plot([0, arm.wrist[0]],
[0, 0],
'b--')
gamma = atan2(arm.wrist[1], arm.wrist[0])
draw_angle(beta, offset=theta0, r=0.2)
draw_angle(gamma, r=0.6)
plt.annotate("$x$", xy=(0.7, 0.05), size=15, color="b")
plt.annotate("$y$", xy=(1, 0.2), size=15, color="b")
plt.annotate(r"$\beta$", xy=(0.2, 0.2), size=15)
plt.annotate(r"$\gamma$", xy=(0.6, 0.2), size=15)
plt.axis("equal")
plt.show()
Our first joint angle \(\theta_0\) added to \(\beta\) gives us the angle between the positive \(x\)-axis and the displacement vector \(r\); let’s call this angle \(\gamma\).
\(\gamma = \theta_0+\beta\)
\(\theta_0\), our remaining joint angle, can then be expressed as
\(\theta_0 = \gamma-\beta\)
We already know \(\beta\). \(\gamma\) is simply the inverse tangent of \(\frac{y}{x}\), so we have an equation of \(\theta_0\)!
\(\theta_0 = \tan^{-1}(\frac{y}{x})-\tan^{-1}(\frac{l_1\sin(\theta_1)}{l_0+l_1\cos(\theta_1)})\)
We now have the inverse kinematics for a planar two-link robotic arm. If you’re planning on implementing this in a programming language, it’s best to use the atan2 function, which is included in most math libraries and correctly accounts for the signs of \(y\) and \(x\). Notice that \(\theta_1\) must be calculated before \(\theta_0\).