Skip to content

perf(mpr): reduce iteration count + tolerance for RL training workload #84

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
zhihuidu-amd wants to merge 1 commit into from
Closed

perf(mpr): reduce iteration count + tolerance for RL training workload #84

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion genesis/engine/solvers/rigid/collider/collider.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -86,7 +86,7 @@ def __init__(self, rigid_solver: "RigidSolver"):

self._mc_perturbation = 1e-3 if self._solver._enable_mujoco_compatibility else 1e-2
self._mc_tolerance = 1e-3 if self._solver._enable_mujoco_compatibility else 1e-2
self._mpr_to_gjk_overlap_ratio = 0.25
self._mpr_to_gjk_overlap_ratio = 0.1 # OPT: fewer GJK fallbacks
self._box_MAXCONPAIR = 16
self._diff_pos_tolerance = 1e-2
self._diff_normal_tolerance = 1e-2
Expand Down
4 changes: 2 additions & 2 deletions genesis/engine/solvers/rigid/collider/mpr.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,8 +14,8 @@ def __init__(self, rigid_solver):
# It has been observed in practice that increasing this threshold makes collision detection instable,
# which is surprising since 1e-9 is above single precision (which has only 7 digits of precision).
CCD_EPS=1e-9 if gs.qd_float == qd.f32 else 1e-10,
CCD_TOLERANCE=1e-6,
CCD_ITERATIONS=50,
CCD_TOLERANCE=1e-4, # OPT: 100x looser, sufficient for RL training
CCD_ITERATIONS=5, # OPT: G1 locomotion converges in <5 iterations (was 50)
)
self._mpr_state = array_class.get_mpr_state(self._solver._B)

Expand Down