-
Hi, I've been trying to replicate one of the environments from Data-Efficient Hierarchical Reinforcement Learning in brax, specifically Ant Push, which is an Ant env with some walls and a movable block, that has to be pushed out of the way to reach the goal. I've created the following
This is basically exactly the same as After some tinkering I've noticed that removing I've tested it for positional pipeline, and the behavior is somewhat similar, but the block is ejected into air, instead of wobbling (and removing My understanding is that I would appreciate some help with this issue, as I am not very experienced in mujoco and brax, so it is quite hard for me to understand really what is the problem here. |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 4 replies
-
I'd recommend trying this with the |
Beta Was this translation helpful? Give feedback.
I'd recommend trying this with the
MJX
pipeline - thespring
pipeline is difficult to tune, especially for large inertia ratios (small ant vs big block).