Proceedings of 2nd International Multi-Disciplinary Conference Theme: Integrated Sciences and Technologies, IMDC-IST 2021, 7-9 September 2021, Sakarya, Turkey

Research Article

Deep Learning for Self-Driving Vehicles

Download323 downloads
  • @INPROCEEDINGS{10.4108/eai.7-9-2021.2314949,
        author={Safa Jameel Dawood Al-Kamil and Mohammed Salah Al-Radhi},
        title={Deep Learning for Self-Driving Vehicles},
        proceedings={Proceedings of 2nd International Multi-Disciplinary Conference Theme: Integrated Sciences and Technologies, IMDC-IST 2021, 7-9 September 2021, Sakarya, Turkey},
        publisher={EAI},
        proceedings_a={IMDC-IST},
        year={2022},
        month={1},
        keywords={self-driving vehicles dsdnet deep learning prediction and planning},
        doi={10.4108/eai.7-9-2021.2314949}
    }
    
  • Safa Jameel Dawood Al-Kamil
    Mohammed Salah Al-Radhi
    Year: 2022
    Deep Learning for Self-Driving Vehicles
    IMDC-IST
    EAI
    DOI: 10.4108/eai.7-9-2021.2314949
Safa Jameel Dawood Al-Kamil1,*, Mohammed Salah Al-Radhi2
  • 1: Southern Technical University, Basra, Iraq
  • 2: Budapest University of Technology and Economics, Budapest, Hungary
*Contact email: safa.alkamil@stu.edu.iq

Abstract

Self-driving vehicles (SDV) and advanced safety features offering the greatest challenges and opportunities for Artificial Intelligence. The understanding of human intention is a very difficult task. So, the prediction of other drivers’ future behaviour is very important to perceive their past motion and analyse the interplay with other agents and process the information available from the scene. Automated driving systems (ADSs) promise a safe, comfortable, and efficient driving experience. We propose in this paper the Deep Structured Self-Driving Network (DSDNet), which performs object detection, motion prediction, and motion planning with a single neural network. However, we improve the deep structured energy-based model. Additionally, DSDNet explicitly exploits the predicted future distributions of actors to plan a safe manoeuvre. Experiments Results on a self-driving dataset show that our model improves the performance of detection, prediction, and planning significantly.