SO101-Nexus

SO101-Nexus provides Gymnasium-compatible simulation environments for the SO-100 and SO-101 robot arms.

There are very few standardized simulation environments available for these arms. SO101-Nexus addresses this by providing fully customizable, vision-and-language-grounded environments across multiple physics backends, enabling consistent experimentation and benchmarking. Robust policies should generalize across simulators before being deployed in the real world -- sim-to-sim transfer helps identify policies that rely on simulator-specific artifacts.

SO101-Nexus also provides a foundation for training text-conditioned embodied policies via curriculum learning, with environments that expose primitives such as object localization and grasping.

Key Features

• Standard Gymnasium API -- works with any RL library that supports Gymnasium
• Multiple backends -- MuJoCo for fast single-env workflows, ManiSkill for GPU-accelerated batched simulation
• Five manipulation tasks -- pick-and-lift, pick-and-place, reach, look-at, and move
• Configurable scenes -- swap objects, add distractors, randomize colors, tune rewards
• YCB object support -- use real-world object meshes from the YCB dataset
• Sim-to-real transfer -- environments match the physical SO-100 and SO-101 arms

Quick Examples

Pick up and lift an object in five lines:

import gymnasium as gym
import so101_nexus_mujoco

env = gym.make("MuJoCoPickLift-v1", render_mode="human")
obs, info = env.reset()

for _ in range(1000):
    action = env.action_space.sample()
    obs, reward, terminated, truncated, info = env.step(action)
    if terminated or truncated:
        obs, info = env.reset()

env.close()

Use a banana from the YCB dataset instead of the default cube:

import gymnasium as gym
import so101_nexus_mujoco
from so101_nexus_core import PickConfig, YCBObject

config = PickConfig(objects=YCBObject(model_id="011_banana"))
env = gym.make("MuJoCoPickLift-v1", config=config, render_mode="human")

Add distractors and randomize colors for domain randomization:

import gymnasium as gym
import so101_nexus_mujoco
from so101_nexus_core import PickConfig, CubeObject, YCBObject

config = PickConfig(
    objects=[
        CubeObject(color="green"),
        CubeObject(color="blue"),
        YCBObject(model_id="011_banana"),
        YCBObject(model_id="058_golf_ball"),
    ],
    n_distractors=2,
    ground_colors=["gray", "white", "black"],
    robot_colors=["yellow", "orange"],
)
env = gym.make("MuJoCoPickLift-v1", config=config, render_mode="human")

Tasks

SO101-Nexus ships five manipulation primitives that cover the fundamentals of robotic grasping and movement:

• PickLift -- grasp an object and lift it above a height threshold
• PickAndPlace -- pick up a cube and place it at a marked target location
• Reach -- move the end-effector to a target position in space
• LookAt -- orient the end-effector to gaze at a target object
• Move -- move the end-effector to a target position with tight tolerance

All five tasks are available on both the MuJoCo and ManiSkill backends. ManiSkill additionally supports GPU-accelerated batched simulation and SO-100 arm variants. Genesis support is on the roadmap.

Get Started

• Installation -- install the backend you need with pip or from source
• MuJoCo Quickstart -- run your first MuJoCo environment
• ManiSkill Quickstart -- run batched GPU-accelerated environments
• Working with Objects -- use cubes, YCB objects, or custom meshes
• Customizing Environments -- tune rewards, cameras, spawn regions, and more