Skip to content

Unify fluid trajectory recording and Gym replay - #1160

Merged
yuecideng merged 1 commit into
feat/dexuni-fluid-prototypefrom
codex/unified-fluid-replay
Nov 11, 2026
Merged

yuecideng merged 1 commit into
feat/dexuni-fluid-prototypefrom
codex/unified-fluid-replay

Conversation

@yuecideng

Copy link
Copy Markdown
Contributor

Description

Unify fluid trajectories with the existing Gym replay modes. kinematic and control restore recorded robot, rigid-body and full particle positions/velocities without advancing physics; dynamic restores the initial particle state and then replays recorded controller actions through DexUni. The separate FluidReplayEnv and its synthetic-only tests are removed.

Accepted Task Program pours can opt into --record-trajectory --record-fluid-state. Particle frames are stored in arena-local coordinates in bounded, content-verified sidecars (32 MiB per environment row), aligned to the initial, control and terminal trajectory boundaries. Fluid replay validates qualification, program hash, scene particle counts and sidecar contents before constructing the World. Existing non-fluid replay modes retain their behavior.

Depends on #790 (feat/dexuni-fluid-prototype).

Type of change

  • Enhancement (non-breaking change which improves existing functionality)

Validation

  • 73 focused replay/fluid/runner tests and 42 agent-context tests passed; static Task Program deployment inspection, Black 26.10.1, context validation and API docs coverage (4,777/4,777) passed.
  • Real one-environment FastRT Task Program source passed physical acceptance and recorded 751 control steps with a 1.3 GiB particle sidecar. The complete kinematic and dynamic CLI replays both completed 751 steps. A real frame-200 kinematic state check had maximum particle position/velocity error 0; a 10-step dynamic check wrote fluid state only at reset and subsequently changed particle state through the solver.
  • Representative four-environment I/O measurement at 76,607 particles and 747 frames per row: 5.49 GB stored, 7.65 s bounded-chunk write, 3.83 s sequential read and 220 MiB peak process RSS. This is a synthetic I/O measurement, not a fluid-solver benchmark; reads may benefit from filesystem cache. The real recording's one-run wall time was 117.75 s. A no-recording comparison was stopped because other GPU jobs contended for the device, so no recording-overhead claim is made.

Checklist

  • Ran black ..
  • Reviewed and updated affected agent context and task documentation.
  • Checked public API documentation coverage.
  • Reviewed the necessity of added tests: new sidecar contracts need independent corruption, row-alignment and coordinate tests; existing replay tests were extended for mode behavior.

@yuecideng yuecideng added enhancement Improvements to existing capabilities; use feature for new capabilities. gym Environment lifecycle, managers, configuration composition and registration. data-pipeline Data collection, ODS, sampling, recording, persistence, conversion and analysis. labels Nov 11, 2026
@yuecideng
yuecideng merged commit 65e437d into feat/dexuni-fluid-prototype Nov 11, 2026
3 checks passed
@yuecideng
yuecideng deleted the codex/unified-fluid-replay branch November 11, 2026 12:24
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

data-pipeline Data collection, ODS, sampling, recording, persistence, conversion and analysis. enhancement Improvements to existing capabilities; use feature for new capabilities. gym Environment lifecycle, managers, configuration composition and registration.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant