An analysis of different intelligent control techniques (evolutionary alg., reinf. learn., neural networks) applied to robot manipulator.

Here you can find the blogposts describing this work: https://medium.com/towards-data-science/making-a-robot-learn-of-to-move-intro-2bcf3c3330df.

WORK IN PROGRESS

Work by Norman Di Palo and Tiziano Guadagnino.

We explore the performance of different AI/Machine Learning techniques applied to trajectory tracking for a 3R robotic manipulator, whose dynamical model is unknown. Thus, the robot needs to learn of to precisely follow a trajectory to decrease the position and velocity error. Our baseline is a classic PD controller.

You can find the code for Evolutionary Algorithms applied to neural network controllers and reinforcement learning techniques such as Q-learning for Dynamic PD Tuning and Actor Critic for deterministic policy learning, and then Inverse Dynamic Model Learning by Feedback-Backpropagation.