Physical LiDAR Simulation in Real-Time Engine

TL;DR

This paper introduces a GPU-accelerated, physically-based LiDAR simulation within real-time game engines, enabling accurate and efficient sensor data generation for testing and development.

Contribution

It presents a novel real-time LiDAR simulation method based on physical properties and GPU acceleration, validated against real sensor data.

Findings

The simulation produces accurate depth and intensity data.

Validation shows high accuracy and precision.

Dependence on spectral material data affects results.

Abstract

Designing and validating sensor applications and algorithms in simulation is an important step in the modern development process. Furthermore, modern open-source multi-sensor simulation frameworks are moving towards the usage of video-game engines such as the Unreal Engine. Simulation of a sensor such as a LiDAR can prove to be difficult in such real-time software. In this paper we present a GPU-accelerated simulation of LiDAR based on its physical properties and interaction with the environment. We provide a generation of the depth and intensity data based on the properties of the sensor as well as the surface material and incidence angle at which the light beams hit the surface. It is validated against a real LiDAR sensor and shown to be accurate and precise although highly depended on the spectral data used for the material properties.