End-to-End Self-Driving Car Software Stack

Chris Sunny Thaliyath, SURYAKANTH V GANGASHETTY

Keywords: ADAS, LiDAR, Camera, Sensor Fusion, Au- tonomous Driving, Occupancy Grid, ROS

Abstract:

Abstract—This paper introduces move_car, a modular and
real-time Advanced Driver Assistance System (ADAS) stack
that integrates multi-modal perception, dynamic occupancy grid
mapping, hierarchical planning, and control for autonomous
navigation. The system fuses LiDAR and multi-camera inputs
through a CUDA-based BEVFusion approach, enabling robust
environment understanding via dynamic occupancy grids. A
Model Predictive Control (MPC) framework is employed in
closed-loop execution to ensure precise and safe trajectory
tracking.
The framework is trained on standard autonomous driving
datasets and evaluated within the CARLA simulator on an
NVIDIA RTX 3060 platform. Experimental results demonstrate
real-time performance and reliability. A comparative study
against open-source baselines highlights the effectiveness of
the proposed stack, and key limitations along with potential
directions for future research are discussed.

Submission Number: 2