Digitala Vetenskapliga Arkivet

Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Reinforcement Learning based Motion Planning of Autonomous Ground Vehicles
Linköping University, Department of Electrical Engineering, Vehicular Systems.
2024 (English)Independent thesis Advanced level (degree of Master (Two Years)), 28 HE creditsStudent thesis
Abstract [en]

The thesis explores the application of deep reinforcement learning techniques to the task of motion planning of autonomous vehicles, where the data for training is generated by the vehicle executing actions through trial and error to learn a driving policy. The thesis work implements reinforcement learning algorithm called Soft-Actor Critic algorithm for motion planning in static environments and investigates the extension of the methodology to dynamic environments. The deep reinforcement learning framework is integrated in a driving simulator called CARLA, for training and testing. High dimensional action and state spaces inherent to autonomous vehicle motion planning, along with the requirement of large computational resources and training time, presents a challenge for applicability of the reinforcement learning framework. To ensure computational feasibility, given the available computational resources and reducing training times, a reduced action space is used. The thesis employs Soft-Actor Critic algorithm to learn the steering control for a fixed throttle value of the agent using the image from a front-facing camera on the ego vehicle as input, to solve the problem of motion planning in static environments.An extension of the method for dynamic environments can be to segregate the control into lateral control using pure-pursuit control and longitudinal control using the Soft-Actor Critic algorithm, among others.The observed results for various driving situations in the static environment demonstrate the ability to learn driving policies without pre-existing driving datasets. However, enhancements in diversity in training environment, better perception, hyper-parameter tuning of the policy and value networks, increased computational resources, and the integration of imitation learning can potentially lead to better results.

Place, publisher, year, edition, pages
2024. , p. 70
Keywords [en]
reinforcement learning autonomous vehicles robotics
National Category
Computer Systems
Identifiers
URN: urn:nbn:se:liu:diva-207856OAI: oai:DiVA.org:liu-207856DiVA, id: diva2:1901293
External cooperation
Scania CV AB
Presentation
2024-09-23, 13:15 (English)
Supervisors
Examiners
Available from: 2024-09-27 Created: 2024-09-26 Last updated: 2024-09-27Bibliographically approved

Open Access in DiVA

fulltext(6087 kB)1564 downloads
File information
File name FULLTEXT01.pdfFile size 6087 kBChecksum SHA-512
cd63ffdd26e695f1cb1ee79b2fad79473680655cfed2646f876b5f6c858e65e48f9d400579827e29aa50ec7aba0e60a78ec43a0b853bb691005d6f6520bc0509
Type fulltextMimetype application/pdf

By organisation
Vehicular Systems
Computer Systems

Search outside of DiVA

GoogleGoogle Scholar
Total: 1564 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

urn-nbn

Altmetric score

urn-nbn
Total: 719 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf