Yesterday, 06:41 PM
Thank you for your detailed explanation!
Thanks to your explanation, I now understand that OnCollision only returns the normal contact value produced by the final substep. I also understand that, when the contact force is nearly constant, scaling the result of the final substep by the number of substeps can provide an acceptable approximation of the result over the full physics step.
However, in my tests, this approximation increasingly underestimated the normal contact force as the number of substeps was increased.
Therefore, I tried an implementation-specific measurement approach using the internal contact data from the Burst backend. I recorded the contact state at each substep boundary: immediately before each substep, and after the final substep using OnCollision.
For each contact that could be matched between two consecutive boundary snapshots, I calculated the change in the reported normal contact value and treated it as Δλ.
Since the positional correction associated with the normal multiplier is divided by the number of remaining substeps, I estimated its contribution to the particle-side normal impulse as follows:
ΔJ = n Δλ / (k h)
where n is the contact normal, k is the number of remaining substeps, and h is the substep duration.
I then summed these vectors over all contacts and substeps, and divided the result by the full physics-step duration to obtain an estimated average normal contact force.
In my static tests, where the contacts remained stable, this method produced values closer to the analytical result and reduced the dependence of the measured force on the number of substeps.
I understand that this is an implementation-specific reconstruction rather than an officially supported measurement method. Since you explained that each substep is an independent constraint evaluation cycle, I am uncertain whether taking the difference between consecutive substep values is conceptually valid.
Could you please let me know if there are any issues with this approach? Specifically, is using Δλ / (k h) appropriate for estimating the particle-side normal impulse contribution, or could the independent evaluation of each substep make this reconstruction invalid?
Kind regards,
Thanks to your explanation, I now understand that OnCollision only returns the normal contact value produced by the final substep. I also understand that, when the contact force is nearly constant, scaling the result of the final substep by the number of substeps can provide an acceptable approximation of the result over the full physics step.
However, in my tests, this approximation increasingly underestimated the normal contact force as the number of substeps was increased.
Therefore, I tried an implementation-specific measurement approach using the internal contact data from the Burst backend. I recorded the contact state at each substep boundary: immediately before each substep, and after the final substep using OnCollision.
For each contact that could be matched between two consecutive boundary snapshots, I calculated the change in the reported normal contact value and treated it as Δλ.
Since the positional correction associated with the normal multiplier is divided by the number of remaining substeps, I estimated its contribution to the particle-side normal impulse as follows:
ΔJ = n Δλ / (k h)
where n is the contact normal, k is the number of remaining substeps, and h is the substep duration.
I then summed these vectors over all contacts and substeps, and divided the result by the full physics-step duration to obtain an estimated average normal contact force.
In my static tests, where the contacts remained stable, this method produced values closer to the analytical result and reduced the dependence of the measured force on the number of substeps.
I understand that this is an implementation-specific reconstruction rather than an officially supported measurement method. Since you explained that each substep is an independent constraint evaluation cycle, I am uncertain whether taking the difference between consecutive substep values is conceptually valid.
Could you please let me know if there are any issues with this approach? Specifically, is using Δλ / (k h) appropriate for estimating the particle-side normal impulse contribution, or could the independent evaluation of each substep make this reconstruction invalid?
Kind regards,