Scribd
Upload
51 views18 pages

Chapter 4 - Inverse Kinematics

The document outlines the process of building an arm robot capable of picking and placing items, detailing steps for determining joint positions using inverse kinematics (IK). It discusses various methods for solving IK, including closed form, algebraic, geometric, and numerical solutions, as well as the use of neural networks for predicting joint configurations. The document emphasizes the importance of finding solutions for multiple joint configurations to achieve the desired end-effector position.

Uploaded by

Husein Yusuf
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
51 views18 pages

Chapter 4 - Inverse Kinematics

The document outlines the process of building an arm robot capable of picking and placing items, detailing steps for determining joint positions using inverse kinematics (IK). It discusses various methods for solving IK, including closed form, algebraic, geometric, and numerical solutions, as well as the use of neural networks for predicting joint configurations. The document emphasizes the importance of finding solutions for multiple joint configurations to achieve the desired end-effector position.

Uploaded by

Husein Yusuf
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

Arm Inverse Kinematics

Amanuel M
SiTE - AAiT
Objective
Our end goal:- Build an arm robot that can pick up an item and places it in another
place

● Step 1 - Determine the current or starting state


● Step 2 - Determine a new joint position that will put the end-effector at a
desired place
End Effectors
● The end-effector will have a the tool-frame or end-effector frame
● The unit vectors along the x, y, and z axes are labeled as n, s, and a,
respectively.
○ a is the approach direction
○ s direction is the sliding direction
○ n is the direction normal to the plane formed by a and s

a
s
End Effectors
End Effectors - on a 3DoF
End Effectors
End Effectors
End Effectors
Inverse Kinematics

Compute the FK:

Θ: 30”
R = 1.0

Compute the IK:

X = .866
Y = 0.5
R = 1.0

Find Θ
Inverse Kinematics
Compute the FK:

Θ1: 30”
L1 = 1.0
Θ2: 30”
L2 = 1.0
Find x, y
Compute the IK:

X = 1.5
Y = 1.5
L1 = 1.0
L2 = 1.0

Find Θ 1 and Θ2
Inverse Kinematics
● Existence of a solution
● Multiple Solutions

(x’,y’)
Desired
Location
Inverse Kinematics
Methods

● Closed Form
○ Algebraic Solution
○ Geometrical Solution
● Numerical Solutions
○ Gradient Descent
○ Cyclic Coordinate Descent
○ Neural Networks
○ Any Optimization Method
Inverse Kinematics - Geometric Solution
Directly find analytical solution that provides an explicit formula for the joint
variables in terms of the end-effector pose.

(x,y)
Inverse Kinematics - Geometric Solution
Formulate the IK for 3-dof arm non-planar robot
Inverse Kinematics - Algebraic Solution
Directly find analytical solution that provides an explicit formula for the joint
variables in terms of the end-effector pose.

(x,y)
Inverse Kinematics - Numerical Solution
● Iteratively minimize the error from Red to Green
vector
● Minimize the distance between P0 and P1 (the
current status vs the target status)
● Gradient Descent
P0(x0,y0)
○ Formulate the current status using FK
Error ○ Formulate the error using L2 norm
○ Find the gradient of the error with respect to the angles
P1(x1, y1) ○ For all angle, update them in a direction where gradient slope
is the lowest.
Inverse Kinematics - Numerical Solution
● Gradient Descent for a single variable
● Gradient Descent for a single link arm
● Gradient Descent for a two link arm

Use Coordinate Descent instead of the plain Gradient Descent


Inverse Kinematics - Neural Network
● From the simulation or an actual robot, collect pairs of joint - end-effector pose
pairs
● Train a giant neural network to predict the joints from a given end effector
position

You might also like