Getting Started
Environment IDs
Complete list of registered environment IDs in SO101-Nexus.
SO101-Nexus environments are registered with Gymnasium when you import a backend module. Import so101_nexus.mujoco for the single-environment MuJoCo tasks and so101_nexus.warp for the GPU-parallel batched tasks before calling gymnasium.make() or gymnasium.make_vec().
import so101_nexus.mujoco # registers MuJoCo envs
import so101_nexus.warp # registers Warp envs (requires so101-nexus[warp])Listing Environments
from so101_nexus.env_ids import all_registered_env_ids
for env_id in all_registered_env_ids():
print(env_id)All Registered Environment IDs
all_registered_env_ids() returns the IDs for whichever backends you have imported. With the MuJoCo backend imported it returns the five MuJoCo IDs:
| Environment ID | Task |
|---|---|
MuJoCoPickLift-v1 | Pick up and lift an object |
MuJoCoPickAndPlace-v1 | Pick up and place an object at a target |
MuJoCoTouch-v1 | Touch an object on the table |
MuJoCoLookAt-v1 | Orient the end-effector toward a target |
MuJoCoMove-v1 | Move the TCP in a cardinal direction |
Warp Environment IDs
The Warp backend registers five batched vector environments. Import so101_nexus.warp (from so101-nexus[warp]) and create them with gymnasium.make_vec():
| Environment ID | Task |
|---|---|
WarpPickLift-v1 | Pick up and lift an object |
WarpPickAndPlace-v1 | Pick up and place an object at a target |
WarpTouch-v1 | Touch an object on the table |
WarpLookAt-v1 | Orient the end-effector toward a target |
WarpMove-v1 | Move the TCP in a cardinal direction |
import gymnasium as gym
import so101_nexus.warp # noqa: F401
envs = gym.make_vec("WarpTouch-v1", num_envs=4096, device="cuda")