Dust Cloud Generation from High Fidelity Sensor Simulation on HPC
Assessing the impact of dust on sensor performance is crucial for autonomous navigation of ground vehicles. The ability to evaluate sensor performance in dusty environments in a simulated setting would prove invaluable for navigational planning in extreme environments. ERDC/GSL has been engaged in the development of modeling and simulation technology of sensors, which include lidar, cameras, and GPS. These capabilities are included in the VANE/ESE software. The aim of this work is to incorporate dust into our simulations to achieve high fidelity measurements of sensor performance in realistic conditions.
Conventional CFD programs lack the ability to simulate the dust cloud generated by a moving vehicle. The MantaFlow fluid simulation framework has this capability but is typically run with the 3D modeling software Blender. Our first computational challenge was developing a method to run numerous instances of Blender, which is a GUI-based software, on DSRC systems in an efficient manner. A Python interface was developed to start multiple instances of Blender and allocate a portion of available CPU cores accordingly. This approach better utilizes each node and enables the generation of more data in the same amount of time.
Real-world tests of environmental impacts on sensor performance can be expensive, time- consuming, and in some cases may be impossible due to the controlled nature of certain environments. VANE/ESE, with its dust simulation support, offers an innovative solution that can replicate testing in a dust-laden environment, all while reducing costs and time investments. Our method of running an OTS software concurrently to saturate each node is likely applicable to other projects with software-related bottlenecks or tools that lack scalability.
PRESENTER
Lamb, Zachary
zachary.lamb@gdit.com
606-748-7649
HPCMP PET - GDIT
CO-AUTHOR
Nathan Bowman
nathaniel.bowman@gdit.com
Peilin Song
peilin.song@erdc.dren.mil
Ruth Cheng
ruth.c.cheng@erdc.dren.mil
CATEGORY
Artificial Intelligence / Machine Learning usage for HPC Applications
SYSTEMS USED
Carpenter, Narwhal, Nautilus
SECRET
No