[BUG FIX] Fix missing buoyancy on rigid bodies submerged in SPH fluid.#2857
Conversation
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: c75d2106f8
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
|
Please add a minimal unit (integration) test with analytical validation. |
189e249 to
cdce737
Compare
duburcqa
changed the title
duburcqa
changed the title
|
@codex review |
…pling (Genesis-Embodied-AI#2070) The SPH-rigid coupling only transferred force when a particle was actively colliding (rvel_normal_magnitude < 0). This meant stationary submerged objects never experienced upward buoyancy force from static fluid pressure. Fix: When a particle overlaps with the rigid body (signed_dist < 0) and has positive pressure, directly transfer the pressure force to the rigid body using the SPH pressure, particle volume, and support radius. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…ancy (Genesis-Embodied-AI#2070) DFSPH never writes particles.p (unlike WCSPH which uses equation of state), so the pressure-based buoyancy force transfer added in the previous commit was a no-op for DFSPH scenes. Fix by adding a kernel that derives the implicit pressure from the density solve residual: p = -(drho - 1.0) * dfsph_factor * rho0 after the density solve converges and writes it to particles.p. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…yancy (Genesis-Embodied-AI#2070) Validates that submerged rigid bodies receive physically meaningful upward force from the SPH-rigid coupling and that WCSPH pressure fields follow the hydrostatic gradient. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
…e integral and accumulate the DFSPH implicit pressure over the density solve.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: cdce737db3
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
cdce737 to
69f7ea6
Compare
3b800fc to
16ab62a
Compare
|
@codex review |
|
Codex Review: Didn't find any major issues. Another round soon, please! Reviewed commit: ℹ️ About Codex in GitHubCodex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback". |
Introduces LegacyCouplerOptions.sph_akinci_boundary (default False) for WCSPH boundary-particle buoyancy. Stock Genesis-Embodied-AI#2857 coupler path is unchanged. Includes coupling tests, underwater examples, and UNDERWATER.md.
2 participants
Description
Rigid geoms now receive the static pressure force of the surrounding SPH fluid. Each particle within kernel support of a geom surface applies the ghost-mirrored pressure force
2 * p * V * sigma(signed_dist)along the surface normal, wheresigmais the closed-form plane integral of the cubic spline kernel. Sincesigmaintegrates to 1/2 across the support band, a covering layer of particles transmits exactlypper unit area, so Archimedes buoyancy is recovered with no tuning constant. The particle receives the opposite reaction, keeping the exchange momentum-conserving. Fixed links are exempt: they cannot respond to the force, and fluid resting on them keeps behaving exactly as before.DFSPH scenes get a physically meaningful
particles.pby accumulating the per-iteration stiffness of the density solve, which matches the equation-of-state pressure that WCSPH already provides. In a settled column the bottom pressure matches the exact column weight per unit area to within 0.2%.Hibernated links now wake when they receive a coupling force from any solver: couplers write forces directly into the coupling-force accumulator, bypassing the wake-aware public force API, and forward dynamics only consumes the forces of awake entities, so a sleeping floater would stay frozen while the fluid keeps receiving the opposite reaction.
Related Issue
Resolves #2070
Motivation and Context
Submerged rigid bodies never floated: the coupling only transferred momentum from particles approaching the surface, and the pressure solve holds fluid nearly static, so static pressure was never transmitted. On main, a rigid ball sinks at every density from 200 to 1500 kg/m^3 in a rho=1000 fluid. With this change the flotation threshold lands near the fluid rest density: between 800 and 1000 under DFSPH, between 500 and 800 under WCSPH (its stiff equation of state under-reads pressure near boundaries; exact Archimedes under WCSPH would require boundary-particle sampling, out of scope for the legacy coupler).
How Has This Been / Can This Be Tested?
tests/coupling/test_sph_rigid.py::test_rigid_flotation_follows_density_ratio: a rho=200 ball must rise and a rho=1500 ball must sink, both pressure solvers, n_envs 0 and 2. Fails on main.tests/particles/test_sph.py::test_pressure_carries_column_weight: bottom-band pressure of a settled column must carry the analytically exact column weight per unit area and decrease with height. The DFSPH case fails on main.tests/coupling/test_sph_rigid.py::test_coupling_force_wakes_hibernated_link: a hibernated resting ball must wake up and float when fluid is teleported over it. Fails with the wake-up scan disabled, with the ball frozen at its resting height.tests/particles/,tests/coupling/andtests/rigid/test_islands.pysuites pass locally, including the water-settling check oftest_deformable_parallel, which is bit-identical to main thanks to the fixed-link exemption.Checklist:
Submitting Code Changessection of CONTRIBUTING document.