Running simulations faster

[Oskar Brodin]

Hi,

I am working on a project where the time it takes for the simulation is important. I am simulating a car running in a straight line at constant speed for 3600 seconds (one hour). Is there any way to make this simulation fast? It currently takes about 6 minutes to do it, but for me that is sadly not sufficient.

Any tips?

/Oskar

[HUZAIFA MUSTAFA UNJHAWALA]

Hello Oskar,

If you do not need visualization and are happy with the vehicle operating on a rigid terrain, you could check out the library of low-fidelity vehicle models - https://github.com/uwsbel/low-fidelity-dynamic-models. Herein there are 3 models with varying fidelity, the 11 DoF model which is the fastest, the 18 DoF model and the 24 DoF model which is the slowest but most expressive. Each of these models take as input the same commands as a Chrono Vehicle - throttle, brake and steering. These models are much faster than real-time. For instance, the 18 DoF model on a commodity CPU is 3835 times faster than Real Time. So 3600 seconds should take about 1 second to simulate. The models are also available on Nvidia GPUs

where you could run upto 150,000 18 DoF models in real-time.

However, do note that these models are standalone and are not yet part of Project Chrono.

If you are looking for something within Project Chrono, you could improve computational speed by simplifying the vehicle model by using simplistic vehicle components. Here is a nice demo where you can play around with different fidelity's of vehicle sub-systems for a HMMWV vehicle model and understand the computational speed of each - https://github.com/projectchrono/chrono/blob/main/src/demos/vehicle/wheeled_models/demo_VEH_WheeledJSON.cpp

Hope I answered your question :)

[Oskar Brodin]

Perfect!

Thank you for this.

Is there any way of running this model at a constant speed? I saw something in the thesis about a predefined speed. Is this implemented? 

/Oskar