You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
As shown in MJX 3.2.7, ELLIPSOID and CYLINDER are implemented but only collide with other primitives. If I use mesh for wheeled robots, the GPU memory consumption will be incredible for Rough environments with hfield. If I only use the empty world with flat plane, the GPU memory consumption is also not reasonable.
Flat Plane
The GPU memory will take around 20GB, much bigger than other examples (e.g., GO1) that take no more than 5GB.
Rough Hfield
W external/xla/xla/hlo/transforms/simplifiers/hlo_rematerialization.cc:3021] Can't reduce memory use below 17.42GiB (18704526550 bytes) by rematerialization; only reduced to 168.35GiB (180768929689 bytes), down from 169.22GiB (181694524993 bytes) originally
The text was updated successfully, but these errors were encountered:
The feature, motivation and pitch
As shown in MJX 3.2.7, ELLIPSOID and CYLINDER are implemented but only collide with other primitives. If I use mesh for wheeled robots, the GPU memory consumption will be incredible for Rough environments with hfield. If I only use the empty world with flat plane, the GPU memory consumption is also not reasonable.
Alternatives
No response
Additional context
Details for my settings. I use Mujoco-Playground and parallel 8192 wheeled robots. I use train_jax_ppo.py for training.
The GPU memory will take around 20GB, much bigger than other examples (e.g., GO1) that take no more than 5GB.
W external/xla/xla/hlo/transforms/simplifiers/hlo_rematerialization.cc:3021] Can't reduce memory use below 17.42GiB (18704526550 bytes) by rematerialization; only reduced to 168.35GiB (180768929689 bytes), down from 169.22GiB (181694524993 bytes) originally
The text was updated successfully, but these errors were encountered: