Title:	Self-supervised learning for multimodal perception systems
DNr:	Berzelius-2024-45
Project Type:	LiU Berzelius
Principal Investigator:	Carl Lindström <carlinds@chalmers.se>
Affiliation:	Chalmers tekniska högskola
Duration:	2024-03-01 – 2024-09-01
Classification:	10207
Keywords:

Abstract

The advancement of autonomous vehicle technology has the potential to revolutionize the transportation industry and greatly improve road safety. However, the successful implementation of this technology depends heavily on the ability of autonomous vehicles to accurately and efficiently perceive their environment, often using a combination of camera and lidar sensors. Current perception systems, although sophisticated, still face significant challenges in detecting and interpreting complex traffic scenarios, particularly in tight areas and poor lighting conditions. This project aims to explore and develop novel deep learning-based perception systems that can overcome these challenges. Specifically, the project will investigate self- supervised learning strategies that leverage vast amounts of unlabeled data to develop more accurate and robust perception models. By reducing the reliance on manually annotated data, this approach can potentially speed up the development of more advanced perception systems, enabling autonomous vehicles to detect and respond to the surrounding environment more accurately in real-time. We believe self-supervised learning will be an essential element in future autonomous systems that can significantly boost their performance.