IAAC – Master in Robotics and Advanced Construction
Software III Seminar
–
Faculty: Mateusz Zwierzycki
Faculty Assistant: Starsk Lara

AI CONTROLLER

Syllabus

This course will be an opportunity to experiment with various machine learning tools and algorithms available in Owl. It will start with a short lecture explaining the current and the past problems of AI in design. The presentation will introduce the notions of supervised, unsupervised and reinforcement learning. Using Owl, the participants will then see the Grasshopper-based workflow for neural networks (supervised learning), k-means clustering and t-SNE (unsupervised learning) and finally utilizing Q-Learning as an example of reinforcement learning. Introduction will end with an overview of Owl components used for image and data pre- and post-processing.

                                                                                                                                                                                              Ph @Mateusz Zwierzycki

Aims

After the introduction, the focus will be put on reinforcement learning in the context of one of the most popular algorithm – Q-Learning. Students will take an in-depth look at the inner mechanics of the algorithm, learn about reward engineering, exploration vs exploitation problem, Markov decision process etc. The main goal of the course is to replace a PID controller for a simple robot with a trainable Q-Learning agent.

Learning objectives

Additionally the participants will learn how to script using Owl (either with Python or C#), which will open up a range of possibilities to utilize other machine learning tools in Grasshopper. Over the course of the semester this knowledge shall be extended to application of QL in the students’ projects. Additional online consultations will be held to help the participants with that task.