Forward and Inverse Kinematics for Robotic Manipulator

Predicting Forward and Inverse Kinematics for Robotic Manipulator

Motivation:

• To develop a general n-revolute robotic arm class in python.

• To implement the theory of kinematics, velocity analysis, dynamics, PID controller, and trajectory generation learned from the book of Craig.

• Solve inverse kinematics using Deep Neural Network, because of there being no general solution to solve for inverse kinematics of a non-intersecting wrist arm.

Outcome

• Developed a class of robotic arm, and PID controller. (robotic_arm.py, pid.py, pi.py)

• Used animation to demonstrate the inverse kinematics and motion of the arm.

• Developed a class of neural network, which can be used to generate NN of any number of nodes and hidden layers. Added sigmoid, linear, and RELU activation functions. (nn_class.py)

• Implemented the NN on a 2 link robotic arm. For 250 training data points, and NN of shape - [3, 5, 4] (nodes of Hidden layers 1,2, and 3) the error RMS for test data is about 1.05. (nn_class.py)

• Implemented Histogram of oriented gradients methods combined with SVM classifier to classify digits, could have used CNN but was contrained due to low processing power of my laptop. Anyway good project to learn the importance of gradients in face/edge detection. (hog_svm.py)