University of Twente Student Theses
Development and optimization of a robot navigation algorithm combining deep reinforcement learning and state representation learning
Obbink, R. (2020) Development and optimization of a robot navigation algorithm combining deep reinforcement learning and state representation learning.
|
PDF
21MB |
| Abstract: | Reinforcement learning is a growing research field in robotics. It can help solve complex tasks that do not appear to have explicit programming solutions. These complex, high-level tasks often require high-dimensional input data such as camera images. One of the challenges that remains using end-to-end reinforcement learning with high-dimensional inputs is the enormous amount of (experience) data required to solve the task. In robotics, obtaining this data is often time-consuming and costly. State representation learning promises to solve this problem by introducing a network that maps the high-dimensional input data to a low-dimensional state prediction before it enters the reinforcement learning algorithm. This can speed up the learning process significantly and therefore reduce cost. This thesis features a 2D navigation problem of a mobile robot that is tasked to reach a target location based on camera images and laser data and aims to continue the research of state representation learning. This is done by investigating the effect of the state prediction’s dimension on performance and by conducting a more in-depth analysis regarding the physical relevance of the encoded state representation. Furthermore, it investigates how well the method can be applied in more realistic use-cases. This includes more realistic and more complex environments, multi-target tasks and investigating the possibilities of transfer learning from simulation to reality. |
| Item Type: | Essay (Master) |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 54 computer science |
| Programme: | Systems and Control MSc (60359) |
| Link to this item: | https://purl.utwente.nl/essays/80765 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page