Overview

SO101-Nexus is split into two backend packages. You install only the one you need -- the shared so101-nexus-core package is pulled in automatically as a dependency.

• so101-nexus-mujoco -- MuJoCo-based environments for the SO-101 arm. All five tasks are available.
• so101-nexus-maniskill -- ManiSkill-based environments for both the SO-100 and SO-101 arms. All five tasks are available with GPU-accelerated batched simulation.

Importing a backend package registers its environments with Gymnasium:

import so101_nexus_mujoco      # registers MuJoCo envs
import so101_nexus_maniskill   # registers ManiSkill envs

After importing, create environments with gymnasium.make().

When to Use Each Backend

MuJoCo is the right choice when you want:

• Fast single-environment simulation with a live viewer
• Custom mesh objects (MeshObject)

ManiSkill is the right choice when you want:

• GPU-accelerated batched simulation via num_envs for faster training
• SO-100 arm support alongside SO-101

Both backends support all five task types (PickLift, PickAndPlace, Reach, LookAt, Move).

Next Steps

• Installation -- install your chosen backend
• MuJoCo Quickstart -- run your first environment
• ManiSkill Quickstart -- run batched environments