How to train your drones

Due to advancements in processing power, computationally expensive machine learning algorithms are now becoming more readily accessible.

With this advancement there is now a means to investigate these computationally expensive methods and compare them. 

The aim of this project is to take a relatively complex offline machine learning algorithm known as "Deep Q Learning” and compare it against less computationally extensive learning algorithms. The main models that were compared are a basic Deep Q Network against an online method known as Regret Matching and also a method of combining a Deep Q Network with Regret Matching. 

The results from this project seemed to indicate that the regret matching method produced the best results and that the combination of regret matching and Deep Q Learning produced better results when compared against a basic Deep Q Network.

Theme
Transforming technologies

Booth
TT52

School
Electrical and Electronic Engineering

Exhibitors
Lachlan Glynn
Jake Miller

vote for this project: TT52

Back to project list