Repository navigation
fix(libero): hold LIBERO end-effector position during wrist and pitch rotation (v0.1) - #240
Merged
Sonorous281 merged 6 commits intoOct 14, 2026
Conversation
… rotation Both primitives copy the end-effector xyz on entry and command translation as clip((hold_xyz - current_xyz) / action_scale, -1, 1), with action_scale defaulting to 0.05 m per unit action. Rotation uses clip(clipped_radians / rot_scale, -1, 1) with rot_scale defaulting to 0.5 rad per unit action, and rotate_pitch also writes action[5] from the wrapped yaw error against the yaw at entry. The result adds converged, max_pos_drift_m, and position_hold_ok, which is true when the largest xyz displacement from the entry pose, including the final pose, is within pos_tol (default 0.03 m); a missed hold stays a normal result with no error key. The planner tool descriptions and schemas expose the new arguments and result fields, and tests/unit_tests/robots/libero/test_libero_rotation_hold.py adds an offline fake controller (0.05 m and 0.5 rad per unit action) covering command scale, translation back toward the entry xyz, pitch yaw correction, hold failure, and an exhausted step budget. Fixes RLinf#234
panhe1818
requested review from
jx-qiu,
qurakchin and
wilburx813
and removed request for
jx-qiu and
wilburx813
October 12, 2026 03:38
Collaborator
Port the rotate_wrist/rotate_pitch position hold onto the @tool interface: the hold arguments are documented in the method docstrings, results are returned as ToolResult, and the LIBERO tool schema fixture records the new arguments.
Documentation build overview
17 files changed ·
|
Documentation build overview
17 files changed ·
|
Contributor
Author
|
@wilburx813 thanks, merged main in 54ddb47 and ported the change onto the new @tool interface. rotate_wrist and rotate_pitch now take pos_tol / action_scale / rot_scale as documented Args and return the hold fields in the ToolResult, and I updated the libero rotate_wrist/rotate_pitch entries in tests/unit_tests/robots/fixtures/tool_schemas.json to match. the rotation hold test now reads result.data and checks the new schema args. ran tests/unit_tests/robots/libero plus test_tool_schema_contracts.py locally: 77 passed. ruff check and format are clean on the touched files. |
wilburx813
reviewed
Oct 13, 2026
qurakchin
changed the title
wilburx813
approved these changes
Oct 14, 2026
Sonorous281
approved these changes
Oct 14, 2026
Elessar123
reviewed
Oct 14, 2026
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
4 participants
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.
Description
Both primitives copy the end-effector xyz on entry and command translation as clip((hold_xyz - current_xyz) / action_scale, -1, 1), with action_scale defaulting to 0.05 m per unit action. Rotation uses clip(clipped_radians / rot_scale, -1, 1) with rot_scale defaulting to 0.5 rad per unit action, and rotate_pitch also writes action[5] from the wrapped yaw error against the yaw at entry. The result adds converged, max_pos_drift_m, and position_hold_ok, which is true when the largest xyz displacement from the entry pose, including the final pose, is within pos_tol (default 0.03 m); a missed hold stays a normal result with no error key. The planner tool descriptions and schemas expose the new arguments and result fields, and tests/unit_tests/robots/libero/test_libero_rotation_hold.py adds an offline fake controller (0.05 m and 0.5 rad per unit action) covering command scale, translation back toward the entry xyz, pitch yaw correction, hold failure, and an exhausted step budget.
rotate_wrist and rotate_pitch can let the grip site walk several centimeters during the rotation, far enough for a grasped object to slip out, and still return as a normal completion. The result only carried the final angle error plus the environment terminated and truncated flags, so a caller had no signal that the position hold had failed. Each step sent a zero translation and set the rotation component to the clipped radian error divided by 0.10, then clipped to [-1, 1]. A zero translation leaves the position goal on the latest observed pose, so drift is not pulled back to the pose from the start of the call, and dividing a 0.10 rad clip by 0.10 requests a full-scale action (0.5 rad on the controller these calls use) instead of the clipped increment. rotate_pitch also left the yaw channel at zero, and the loop stopped on angle error without measuring xyz displacement from the entry pose.
Fixes #234
Testing
The recorded Python test run found no tests to execute for the touched modules.
Manual verification
Covered in the summary above.
Checklist
Left unticked: this change to
robots/libero/tools.pytouches no documentation.Not claimed: no local test run is recorded for
robots/libero/tools.py.AI was used for assistance.