[Question] Issues with Tet Mesh Generation (Inverted Elements) & Importing External Mesh Data for Deformable Bodies

[conner9403]

Environment

• Isaac Lab Version: 2.3.0
• Isaac Sim Version: 5.1.0 (pip installed)
• OS: Windows 11
• GPU: NVIDIA GeForce RTX 5090 Laptop GPU (Driver: 591.44)

Context

Hello, I am currently working on a project to optimize the topology of a robot finger using the Deformable Body features in Isaac Lab. My goal is to simulate a soft gripper (Fin Ray structure) and optimize its shape.

I am attempting to automatically generate a tetrahedral (Tet) mesh using create_auto_volume_deformable_hierarchy within the simulation environment.

Current Issue

When attempting to generate the Tet mesh from my CAD file, I consistently encounter "Inverted mesh (Inverted element)" errors, causing the simulation to fail or behave unstably.

I have verified the CAD model by performing a simulation in Ansys, where the meshing and simulation proceeded without any issues. Therefore, I believe the CAD geometry itself is valid.

Steps Taken

To resolve this, I have tried the following:

1. Adjusting Resolution: I attempted to generate the mesh with various resolution settings in create_auto_volume_deformable_hierarchy, but the inverted mesh issue persists.
2. GUI Generation: Instead of generating it via code, I tried creating the mesh using the Isaac Sim GUI and loading the resulting USD file. The same inverted mesh error occurs.
3. Hex Mesh Check: I confirmed that generating a Hexahedral (Voxel) mesh works perfectly fine with the same geometry. However, due to the nature of my optimization project, I specifically require a high-quality Tet mesh.

Questions

1. Feasibility: Is the current Deformable Body feature in Isaac Lab (PhysX) robust enough to handle complex shape optimization tasks involving repeated topological changes?
2. Improving Mesh Quality: Aside from the basic resolution settings, are there any advanced methods or workflows within Isaac Lab to generate higher-quality Tet meshes that avoid inverted elements?
3. External Import: Is there a way to bypass the internal mesher and directly import mesh data generated from external software (e.g., Gmsh, TetGen)? Specifically, I am looking for a way to load node/element data from files like .npz or standard mesh formats into the Deformable component.

Thank you for your help!

[jingchang0623-crypto]

Great proposal! We have been running llms.txt in production at our site (miaoquai.com) for over a month now, and our OpenClaw agents use it as the primary skill discovery mechanism.

Some practical observations from running it:

1. The format works surprisingly well — structured sections (## Docs / ## Skills / ## Tools) map perfectly to how agents reason. Our agents went from "cant find the right skill" to "immediately picks the right one" once we organized skills by domain.

2. Versioning is an underrated feature — we use semantic version tags in our llms.txt (## Skills v2.1 -> skills/) and agents naturally handle it. No schema changes needed.

3. Polyglot challenges — if your site has multilingual content (we support both zh-CN and en), the llms.txt should reflect that. We use separate llms-zh.txt / llms-en.txt files.

4. Size matters — our llms.txt has ~2000+ skill references. Some agents truncate. Adding a summary/priority section at the top helped agents prioritize.

We wrote about our implementation journey here: https://miaoquai.com/stories/llms-txt-implement.html — including the mistakes we made (missing trailing slashes, not updating after content changes).

Happy to share more about the monitoring setup if you are interested! We built a simple checker that validates llms.txt integrity every 6 hours.