Repository navigation
Adapt EmbodiChain to DexSim 0.6.0rc1 on Python 3.12 - #1124
Conversation
Update Newton solver, contact query, USD loading and rendering adapters for the Python 3.12 DexSim release. Align regression coverage and project context with the new contracts.
Switch the main workflow and shared Conda activation to py312. Keep the supported Python 3.11 compatibility lane, remove Python 3.10 installation checks, and raise the package minimum to 3.11 because DexSim 0.6.0rc1 has no cp310 wheel.
[High impact] The PR appears safe to merge, with non-blocking feedback about the CI guide and the removed physical-outcome test.
|
Remove CI planner and example fallbacks for Python 3.10, update metadata tests, and align user-facing support documentation with the Python 3.11 minimum.
Regenerate Python 3.11/3.12 Docker locks for DexSim 0.6.0rc1, sync curated architecture evidence, and migrate shipped soft-body demos to grouped DexUni options with per-object contact materials.
Give the Python 3.12 simulation test container a dummy SDL video driver and private runtime directory so DexSim can initialize its native renderer without a display server.
Temporarily hide visual materials from DexSim NoRender native calls, flush deferred simulator cleanup after locomotion subprocesses, and remove the flaky ice retention smoke test as requested.
| @@ -112,48 +112,6 @@ def compute_ik(pose, **kwargs): | |||
| torch.testing.assert_close(plan(moved_bin, tilted_grasp), moved_bin @ baseline) | |||
|
|
|||
|
|
|||
There was a problem hiding this comment.
Removing test_default_scoops_and_retains_ice drops the only check that the simulated scoop actually retains ice. The remaining tests use fake objects to check placement and commanded motion, so they still pass if a DexSim change breaks the physical grasp. Keep this slow GPU check, or replace it with an equivalent physical-outcome test.
Prompt To Fix With AI
This is a comment left during a code review.
Path: tests/sim/test_scoop_ice_demo.py
Line: 114
Comment:
**Empty scoops go unchecked**
Removing `test_default_scoops_and_retains_ice` drops the only check that the simulated scoop actually retains ice. The remaining tests use fake objects to check placement and commanded motion, so they still pass if a DexSim change breaks the physical grasp. Keep this slow GPU check, or replace it with an equivalent physical-outcome test.
---
For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!
|
|
||
| CI defaults to the published internal `v4-py311-py312` image | ||
| and activates `py311`. The image reference and activation command are explicit | ||
| and activates `py312`. The image reference and activation command are explicit |
There was a problem hiding this comment.
docker/README.md says CI activates py312, but the latest change switches both the shared action and the main test job back to py311. Contributors following this guide will use a different Python and dependency set from CI. docker/VALIDATION.md also still says main CI uses Python 3.12. Update both statements to match the workflows, or restore the intended Python 3.12 jobs.
Prompt To Fix With AI
This is a comment left during a code review.
Path: docker/README.md
Line: 193
Comment:
**CI guide names wrong Python**
`docker/README.md` says CI activates `py312`, but the latest change switches both the shared action and the main test job back to `py311`. Contributors following this guide will use a different Python and dependency set from CI. `docker/VALIDATION.md` also still says main CI uses Python 3.12. Update both statements to match the workflows, or restore the intended Python 3.12 jobs.
---
For each issue above, determine whether it is valid and should be fixed. If so, fix it directly.
Description
Pin
dexsim_engine==0.6.0rc1and adapt EmbodiChain to the released DexSim contracts. The change updates DexUni solver defaults and options, URDF continuous-joint loading, USD material export, renderer teardown, contact solver reports and validity masks, and robot asset lookup. Relevant implementation details were selected from #829 and verified against the installed release wheel.Run the primary CI jobs on Python 3.12. DexSim 0.6.0rc1 publishes Python 3.11 and 3.12 wheels, so the package minimum is now 3.11 and the Python 3.10 installation matrix is removed. Python 3.11 compatibility coverage remains in the dataset tooling workflow.
The two local Python 3.10 Conda environments were removed as requested; that cleanup is outside this PR.
Dependencies:
dexsim_engine==0.6.0rc1from the configured private package index.Type of change
Validation
pip checkpasses.pytest -q tests/— 13,308 passed, 592 skipped, 93 deselected. The deselected cases are marked slow by repository configuration; GPU-marked cases are skipped by default and are being run separately.black .andblack --check .pass (2,267 files); API docs coverage passes (4,391/4,391 exports);actionlintpasses for changed workflows; agent context check andgit diff --checkpass.Coverage additions protect changed solver and contact contracts, native source names, USD degree-to-radian stiffness, and raw MuJoCo parameter lowering. Tests using removed query/capture behavior were updated to the 0.6 contract. Simulation and sensor context was updated; affected robot, visualization, environment, RL, and GenSim context remains accurate.
Checklist
black .with the required version 26.10.1.