Edge Device Deployment of Multi-Tasking Network for Self-Driving Operations

TL;DR

This paper presents the deployment of a multi-tasking neural network on an embedded system for autonomous driving, enabling real-time perception tasks like object detection, segmentation, and lane detection.

Contribution

It introduces a multi-tasking network architecture optimized for embedded deployment, with extensive comparisons of backbone models for self-driving perception tasks.

Findings

Successful deployment on Nvidia Jetson Xavier NX

Comparable performance across different backbone networks

Real-time operation of multiple perception tasks

Abstract

A safe and robust autonomous driving system relies on accurate perception of the environment for application-oriented scenarios. This paper proposes deployment of the three most crucial tasks (i.e., object detection, drivable area segmentation and lane detection tasks) on embedded system for self-driving operations. To achieve this research objective, multi-tasking network is utilized with a simple encoder-decoder architecture. Comprehensive and extensive comparisons for two models based on different backbone networks are performed. All training experiments are performed on server while Nvidia Jetson Xavier NX is chosen as deployment device.