• Søren Skov
4. semester, Matematik-teknologi (cand.polyt.), Kandidat (Kandidatuddannelse)
The focus of this project is to train an agent to improve its behaviour of navigating in an indoor environment using visual input. This is done through the use of deep reinforcement learning trained on images to find a number of target positions. The work in this project is based on where an agent is trained on 100 million images to find 100 targets. The performance of this algorithm shows that there are room for improvements.

Therefore, in this project an analysis of what such as agent learns during the training is carried out to get an understanding of what the agent learns and how this might effect the performance. A way to visualize what the agent learns during the training is proposed to help this analysis.
Udgivelsesdatoaug. 2018
Antal sider95
ID: 280537410