Skip to content

feat: update deps + release v0.2.0 - #1

Merged
sebcrozet merged 6 commits into
mainfrom
rapier-up
Oct 27, 2025
Merged

feat: update deps + release v0.2.0#1
sebcrozet merged 6 commits into
mainfrom
rapier-up

Conversation

@sebcrozet

Copy link
Copy Markdown
Member

No description provided.

@sebcrozet
sebcrozet merged commit b798d97 into main Oct 27, 2025
5 checks passed
haixuanTao referenced this pull request in haixuanTao/nexus Jul 14, 2026
emit_limit_constraint always emitted a kind=1 row and computed its
O(ndofs) LU back-solve column, even when the joint sits strictly inside
its bounds — where the row has zero positional bias AND both impulse
clamps at 0, so it can never apply any impulse. Every PGS sweep then
carried #joints dead rows per env per substep.

Skip emission when !min_enabled && !max_enabled: the slot stays kind=0
(pre-zeroed by the discovery walk), which compute_joint_columns_par and
both PGS sweeps already skip. Constraints are rebuilt from current
coords each substep, so a joint reaching its bound re-emits next
substep exactly as before. Result-identical (verified: bit-identical
training stats on the zealot biped).

Measured (zealot biped trainer, 5090): gpu_mb_init_solve_joint_with_bias
was the #1 GPU kernel for a 25-joint humanoid (28% — O(joints) rows x
O(ndofs) back-solve each, per substep); with the skip, full-body-G1
iteration throughput +16%, 12-DOF biped +5%.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016Dbpu5NhqXK2Xt3YzUbJRM
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant