LiDAR Sensor Simulation in Unity: Real-Time Performance for Autonomous Systems and Virtual Environments
Аннотация
LiDAR sensor simulations within the virtual setting is a core enabling factor towards the advancement of automation in general as well as robotics in particular. In this paper, a GPU-implemented LiDAR sensor simulation environment based on the Unity simulation platform is proposed, especially focusing on real-time and high-quality simulation. The above proposed solution utilized Unity rendering pipeline as well as GPU parallelism to create realistic LiDAR point cloud that are customizable with parameters such as scan pattern, range, and noise models. In this framework, many improvements done to raycasting and the use of compute shaders lead to dramatic reduction time compared to conventional CPU methods. Numerous tests were performed to assess the framework acceptance, precision, and unless otherwise noted, scalability. Evaluation proves <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{4 x}$</tex> to <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{1 0 x}$</tex> point cloud generation efficiency for large scale scenes as compared to CPUs, with less than a 1% margin from real LiDAR data. As expected, real-time frame rates were maintained in dynamic scenes with greater than <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{1 M}$</tex> points per second, further proving the system's ability to handle denser geometries in dynamic scenes. These applications range from autonomous vehicle testing, robotics training to those that are used in Virtual Reality environments. The integration of the framework within Unity guarantees compatibility with dynamic environments, interaction, and different terrains and therefore, the proposed method is suitable for research and development and instructional applications. Future works will build and extend from this research to work on noise modeling beyond a simple noise model and the integration of multiple sensors in order to increase the flexibility and accuracy and applicability of the system.