Inverse Problem – Force/Pressure Estimation in SOFA

[angela241299]

Hello,

I am a student using SOFA for my university project. I am new to SOFA and have been working on estimating the forces/pressures that hands apply to a rigid articulated object. I understand this is not a direct simulation problem, as my goal is to compute forces/pressures from an already known sequence of motion.

To be more specific, I have hand and object meshes along with their interactions over a sequence of frames. Currently, I am using a FreeMotionAnimationLoop and setting the vertex positions at each iteration. For contact handling, I am using PenalityConstraintForceField, since it provides the highest forces in regions of interpenetration. In my case, due to imperfections in the meshes, interpenetrations occur frequently.

Could you please let me know whether SOFA is suitable for this type of force simulation? My current results correctly indicate contact regions, but I often observe zero force in areas of interpenetration—where I would expect the largest forces. Instead, the highest forces tend to appear in a ring around the contact region, i.e. just at the boundary of penetration.

Additionally, could you advise on the type of mesh (surface/volumetric) representation best suited for this purpose? My ultimate goal is to obtain per-vertex force values.

Thank you in advance for your guidance!
Angela

[hugtalbot]

Hi @angela241299

Thank you for your message. Interesting topic, SOFA is definitely suitable for this.

About the PenalityConstraintForceField, it is a method for contact response which defines springs on each contact point (output from the collision detection phase). As it is a spring-based response (weak formulation), the non-interpenetration is not guaranteed at the end of the time step. Using Lagrange multipliers would allow to better respect non-interpenetration.

About meshes, the mechanical representation should rely on volumetric meshes while the collision representation should rather rely on surface meshes.

I understand that you conducted an experiment with a hand manipulating a rigid articulated object. Is this correct? Would you have an image of your experiment? With these info, we could better advise on the methodology.

Hope this helps

[angela241299]

Hello @hugtalbot,

Thank you for your prompt reply!

My goal is to determine the force or pressure that hands apply on a rigid articulated object over a sequence of frames. I already know the exact motion, as well as the hand and object meshes for each frame, and based on their positions and contacts, I want to compute these force/pressure values.

Since the dataset provides only .obj files, which are not always watertight, I used pygalmesh to generate .vtk volumetric meshes for use in my simulation.

I’ve attached some examples to illustrate the problem:

The first example shows the left-hand interaction with a ketchup bottle in two consecutive frames. The first two images show the original frame from the dataset alongside the force I simulated. Ideally, the maximum force should occur in the interpenetration region, gradually decreasing around it (as shown in the first image). However, in the next frame, the left thumb still penetrates the object mesh, but the penetration area shows zero force.
The third image shows the first frame in SOFA during simulation. The last two images illustrate the left thumb penetrating the mesh from an internal perspective in these two consecutive frames.

The second example shows the following:

Images 1, 3, and 4 depict frame 51 from the dataset, my simulated forces, the same frame in SOFA, and the penetration region of the left thumb.

Images 2 and 5 depict frame 53 from the dataset, my simulated forces, and the same frame in SOFA (no penetration here, as the thumb is moving away from the object).

I am not sure why the simulated force is larger when there is no contact than when there is (though in this case it is an interpenetration area).

[bakpaul]

Hello !

Very nice project, I like your simulations ! If you want to have better rendering, don't forget that the community have propose a Unity-based rendering plugin: https://www.sofa-framework.org/applications/plugins/sofa-unity3d-basic-renderer/

For your contact issue I agree with Hugo, your problem here is that you are using springs that will allow interpenetration + you might have bad parameters that will deactivate contact points if you have too large interpenetration.

Now my suggestion :

1. I would reduce the load on collision detection by only using a triangle collision model on the object and a point collision model on the hand.
2. I would reduce the collision model of the hand by extracting the surface (I think this is already the case) and only the tip of the fingers to have faster collision detection
3. I would use a larger alarmDistance in the 'Proximity' object that you use in your collision pipeline along with a slightly higher contactDistance. This will insure that you'll not have pathces without forces in the middle
4. Lastly, if penality are still not good enough for you I would swith to Lagrangian-based constraints. See this scene for an example on how to do this : https://github.com/sofa-framework/sofa/blob/master/examples/Demos/fallingBeamLagrangianCollision.scn

Tell me how it goes !

And by the way, if you are able to do it, you can join us for the 2025 edition of the SOFA Week on November 24-28! This is a good opportunity to show your work with SOFA and discuss with the community.

[github-actions[bot]]

⚠️ ⚠️ ⚠️
@angela241299
Feedback has been given to you by the project reviewers, however we have not received a response from you. Without further news in the coming weeks, this discussion will be automatically closed in order to keep this forum clean and fresh 🌱 Thank you for your understanding