Institutional Repository of Vilnius University
Investigation of perceiver networks for imitation learning in autonomous driving /
Abstract
dc:descriptionImitation learning algorithms are widely applied in autonomous vehicles field. They reach good results in lane following and obstacle avoidance tasks. However current approaches still struggle in urban environments with multiple dynamic objects and complex traffic rules. We discuss that improving image encoding methods could help to alleviate the issues related to static or dynamic object detection, traffic light and stop sign infractions. We propose to use perceiver network as image encoder and show that it reaches lower loss compared to the state-of-the-art model when pre-trained image encoders are considered and no fine-tuning is done. In the same setting, the proposed model also shows higher road completion percentage and lower infraction rate when test runs are made in simulated environment.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2022
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Žaltauskas, Augustas,
Subjects
dc:subject × 1Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- eng
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD146235685&prefLang=en_US