CARLA: Car Learning to Act - An Inside Out

Sumbal Malik, Manzoor Ahmed Khan, Hesham El-Sayed

Research output: Contribution to journalConference articlepeer-review

21 Citations (Scopus)

Abstract

Training autonomous vehicles require rigorous and comprehensive testing to deal with a variety of situations that they expect to undergo on roads in real-time. The physical testing of autonomous driving on roads has always been hindered; by the infrastructure costs, high-performance systems, sensors, communication devices, and jeopardizing the safety of people in the real world. That is where the testing in the simulation helps in filling the gap and democratizes autonomous driving research. There are plenty of simulators to test autonomous driving solutions, and CARLA is one of them. CARLA is a powerful simulator that encompasses tools to develop, train and validate the systems in controlled scenarios. This paper presents a detailed overview of the CARLA, discussing its simulation engine, client-server architecture, urban environment, and a variety of sensors. The wide range of features and modules of each CARLA release has also been reviewed helping the research community to choose the suitable CARLA release meeting best their requirements. Moreover, the comparative analysis of CARLA with well-known open-source and commercial simulators is also presented. The analysis shows that CARLA and LGSVL are currently state-of-the-art open-source simulators for end-to-end testing of autonomous driving solutions. Finally, we discuss the various research applications and use case scenarios where the CARLA has been implemented and extended by the research community.

Keywords

  • CARLA
  • LGSVL
  • autonomous driving
  • simulation

ASJC Scopus subject areas

  • General Computer Science

Fingerprint

Dive into the research topics of 'CARLA: Car Learning to Act - An Inside Out'. Together they form a unique fingerprint.

Cite this