Autonomous Driving Simulator Framework base on Low-cost 3D LIDAR

Abstract

Self-driving simulators have a higher cost-saving effect compared to reality in testing and verifying autonomous driving, but they require a lot of hardware devices due to high computer computation. In some cases, learning data necessary for autonomous driving may be obtained using a game. Games are used as low-cost simulators, but they are limited in simulating specific situations outside the game and obtaining necessary data. As another method, a simulation study of the virtual environment through a game engine is being conducted. However, in game engines, development costs are high because users have to model themselves. In particular, 3D LIDAR requires real-time acquisition of a precise distance of at least 10 Hz by shooting Ray at 360 degrees. Acquiring 3D LIDAR data in real time is a task that consumes a lot of graphics processing unit (GPU), so a low-cost 3D LIDAR simulator requires a low-cost 3D LIDAR simulation. In this paper, we propose a C++-based 3D LIDAR simulation framework using low computer operations. The proposed 3D LIDAR verified its performance in an unpaved map composed of many hills, and introduced a simple local path planning (LPP) generation method with 3D LIDAR generated in this paper by performance verification. We hope that the proposed 3D LIDAR framework will be actively used in autonomous driving fields that require low-cost real-time simulations.