Repository navigation
Conversation
Migrate DexUni example settings to the current namespaced solver interface and preserve the existing cloth contact values. Update the Newton config regression test and document the new mapping contract.
Port the retained MJVBD V2 integration consumers to the current DexUni public interface, while keeping AutoSolver as the Newton default. Preserve namespaced solver options, tutorial calibration, reset/mimic handling, and current task configuration.
Allow geometry-only ContactQuery data for DexUni VBD paths and full impulse/friction data for pure MuJoCo. Add capability tests and GPU end-to-end coverage.
…e-adaptation # Conflicts: # embodichain/lab/sim/cfg/simulation.py # tests/sim/test_cfg.py
Expose pre-build Newton articulation and collision options, runtime-control registration, authored tetrahedral meshes and direct nodal-state writes. Support cloth rest angles and pneumatic cavities through native descriptors, and forward renderer synchronization and graph-capture settings. Keep native follower drives and mimic constraints by default; retain the Default-parity mimic stabilizer as an explicit opt-in. Include focused tests and the owning simulation context documentation. Validation: 718 CPU/context tests passed (2 GPU cases skipped in that run); 3 explicit-tetrahedra/nodal-state GPU tests and both W1 mimic modes passed. Black 26.3.1 (black .), API docs coverage (3269/3269), context checks and whitespace checks passed. Robot/visualization context ownership remains accurate; changed contracts are documented in simulation-system context. Technical-report experiments, output artifacts and the local integration report are intentionally excluded.
Replace the handwritten routing environment with configuration-owned registration and a bounded 49-stage program. Reuse PickUp contact grasps, sampled native motions and the standard Gym bridge; keep DexUni physics and strict measured final acceptance. Scope material references and slip evidence to each acknowledged execution and preserve inactive integration identities. Validation: 2,468 related CPU regressions; 204 observer lifecycle rechecks; two NoRender CUDA spool/contact qualifications; deployment inspector, API coverage, context checks and Black 26.10.1. The paired seed-73 FastRT run remains unsuccessful at strict insertion: the baseline stops on slip, while the Task Program stops on final lowering TCP error.
KeqiYe
force-pushed
the
feat/franka-rope-wrap
branch
from
November 16, 2026 06:52
c911dc2 to
8536d99
Compare
4 tasks done
KeqiYe
force-pushed
the
codex/cable-object
branch
from
November 19, 2026 07:27
d2a1829 to
5517b73
Compare
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
This PR adds
FrankaRopeWrap-v1as a configured Task Program on top of #834 (codex/cable-object). The robot picks up the free end from the floor, winds it around a single post through a commanded 465°, tightens it and finishes with the gripper closed. The scene imports the 97 original centerline points from the packagedcable.usdathrough Gym's USD scene entry point. The task fixes segment zero, applies collision filter span 2 and its calibrated material/contact parameters before spawn, and grasps the free end using bilateral finger contact.Ten semantic stages lower to the standard
MoveJointsatomic skill. The planner samples the full TCP orbit and solves continuous IK on the flat-pad Franka model. Explicitly authored velocity targets preserve the incoming joint command across stage boundaries. The Gym contact sensor observes four physics samples per 60 Hz control interval. Physical acceptance uses the measured rope geometry, finger contacts and evaluation windows.Run the configured task and record a trajectory with the standard entry point.
python -m embodichain run-example sim/demo/franka_rope_wrap --output /tmp/franka-wrapruns the same Task Program and exports the plan, 19 individual check results and measured frames. It also supports custom scenario files, offline planning and a diagnostic time limit.USD migration on the same rc2 wheel
The procedural Task Program at
48d803c61and the USD deployment at27652ab93each ran the full 36-second episode on the same RTX 5090 D v2 using DexSim0.6.0rc2(builddbaca660, Python 3.12, Newton 1.7.0). All 97 centerline points, the anchor, the post center and the planned motion arrays matched exactly. Both runs used the same acceptance thresholds and completed 10 stages and 2,160 control steps.Earlier Task Program qualification
The original #923 head
c911dc21and the configured task ran on the same RTX 5090 D v2 with Python 3.12, Newton 1.7.0 and DexSim 0.6.0rc2 (builddbaca660471d5aaa7e750093e7fe1c207465c4af, wheel SHA-256d450987f763942a1ffc0435710cdabd6ceaffb9e4a309e5f324a7ef88e928f83). Both this branch and its #834 base now pin this release candidate.The recorded 2,160 position/velocity commands match the original command stream element for element. The standard recorder saved the full episode. Evaluating the saved terminal observation reproduced all 19 results. Offline replay restored measured cable states and rendered both cameras at 5, 7, 28 and 36 seconds.
Duration checks
--seconds 35.999executed exactly 2,159 steps and recorded 2,160 states through 35.98333 s. The process exited 1 withpartial,passed=falseandcompleted=false. A sub-frame budget executed zero steps and exported only the settled initial frame.Checks