Skip to content

Fix Wipe table height drifting across resets#816

Merged
Abhiram824 merged 2 commits into
ARISE-Initiative:masterfrom
HaozheZhang6:fix/wipe-table-height-drift
Jul 11, 2026
Merged

Fix Wipe table height drifting across resets#816
Abhiram824 merged 2 commits into
ARISE-Initiative:masterfrom
HaozheZhang6:fix/wipe-table-height-drift

Conversation

@HaozheZhang6

Copy link
Copy Markdown
Contributor

What this does

Fixes #815 (🐛 Bug).

In Wipe, self.table_offset[2] += self.delta_height ran on every _load_model while delta_height was sampled only once (guarded by is None). Since table_offset is a persistent array, the table height accumulated a fixed delta each reset (a linear ramp), and table_height_std produced no per-episode variation after the first episode.

Now the base offset is stored once at init and each reset sets table_offset[2] = base + N(table_height, table_height_std), so the height is re-sampled per episode and never drifts.

How it was tested

Added test_wipe_table_height_no_drift in tests/test_environments/test_wipe_table_height.py: with table_height_std=0.02 the resets no longer ramp, stay centered on the base height, and vary per episode; with the default std=0.0 the height stays fixed (no regression).

Before: table_offset[2] over 6 resets ramped 0.930 -> 0.945 -> ... -> 1.004. After: bounded and varying, e.g. 0.926, 0.923, 0.867, 0.912, ....

How to checkout & try?

pytest -sx tests/test_environments/test_wipe_table_height.py

@Abhiram824

Copy link
Copy Markdown
Member

Hi, change looks good. can you remove the extra test file and I will approve

@HaozheZhang6

Copy link
Copy Markdown
Contributor Author

Removed, thanks!

@Abhiram824
Abhiram824 self-requested a review July 11, 2026 22:44
@Abhiram824
Abhiram824 merged commit 5ce6643 into ARISE-Initiative:master Jul 11, 2026
0 of 2 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

2 participants