Replies: 2 comments 3 replies
-
Beta Was this translation helpful? Give feedback.
-
|
Convexifying objects is a resource intensive procedure. What I've done for some of my robots is to run CoACD outside of genesis and then update my models with the collision bodies. What you'll end up with is the visual meshes, which have collisions disabled, and the hidden convex collision bodies. Doing it this way will also let you adjust the CoACD parameters for each part -- some you might want higher detail, while others can be crude shapes. For my latest robot, I created a script that generates these and updates the model file automatically. In my case, not all bodies need to have collisions, so it skips those, which saves some of the simulation processing resources. |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
Hello, I have encountered some issues while uploading a complex URDF file to Genesis for visualization. Initially, the collision model appeared as a simplified, bulky shape. To address this, I changed the configuration by setting convexify=False.
My current hardware setup includes an RTX 4090 GPU and 32GB of RAM. However, due to the complexity of the structure, I experienced a memory error from NumPy, indicating insufficient RAM. I then tried running the process on a server equipped with 256GB of RAM, and found that about 60GB of RAM was required to handle the task.
Additionally, despite configuring GPU support, I noticed that the process heavily utilized CPU resources, while the GPU remained largely underutilized.
Is this the expected behavior for handling complex URDF models in Genesis?
Are there any settings I can modify to offload more of the computational workload from the CPU to the GPU?
Interestingly, I observed that the FPS was higher when running the process on the CPU instead of the GPU. Why might this be the case during the visualization process?
I would appreciate your guidance on these issues. Thank you!
Beta Was this translation helpful? Give feedback.
All reactions