gym-aloha

What it is and what it does

gym-aloha wraps the ALOHA dual-arm robotic manipulation system as a Gymnasium environment, allowing reinforcement learning researchers and practitioners to train agents on two concrete manipulation tasks: TransferCubeTask (pick and transfer a cube between grippers) and InsertionTask (grasp and insert a peg into a socket). The environment is built on MuJoCo physics simulation and dm-control, providing continuous 14-dimensional action spaces (joint positions and gripper states for two arms) and multi-modal observations (joint states, camera images from multiple angles, and environment state). Rewards are structured as sparse milestones (1–4 points per task) based on task progress.

The package is designed for researchers developing imitation learning, reinforcement learning, or other control algorithms on realistic manipulation tasks. It supports configurable observation types (pixels or pixels with agent position), image resolution, and rendering modes. GPU-accelerated rendering is available but requires careful EGL configuration. The environment is in alpha status and actively maintained, with support for Python 3.10 through 3.14.

Use it for:

• Train reinforcement learning agents on dual-arm manipulation tasks with sparse reward signals and visual observations.
• Develop imitation learning pipelines using the ALOHA environment as a standardized benchmark.
• Prototype robot control policies in simulation before deployment to physical ALOHA hardware.
• Research multi-modal learning approaches combining joint state, velocity, and camera image observations.
• Benchmark manipulation algorithms on structured insertion and transfer tasks with configurable difficulty.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Provides a Gymnasium environment for the ALOHA dual-arm robotic manipulation system, enabling reinforcement learning agents to train on cube transfer and peg insertion tasks with physics simulation via MuJoCo.

Yes, if you are working on reinforcement learning or robot control research and need a standardized ALOHA simulation environment. The package has low install friction, permissive licensing, no known vulnerabilities, and active maintenance. Caveat: requires Python 3.10+, and GPU rendering requires careful MuJoCo/EGL setup; CPU rendering will work but may be slow for large-scale training.

Install

gym-aloha on PyPI

pip install gym-aloha

uv add gym-aloha

poetry add gym-aloha

Installing gym-aloha

Before you install

Low friction—pure Python wheel with four runtime dependencies (dm-control, gymnasium, imageio, mujoco). Actively maintained as of 65 days since release. Python 3.10+ required; MuJoCo itself may require GPU/EGL setup for optimal rendering performance.

License in practice

Apache-2.0 permissive license allows use in commercial and private projects with minimal restrictions; attribution required.

Quickstart

pip install gym-aloha

import gymnasium as gym
import gym_aloha

env = gym.make("gym_aloha/AlohaInsertion-v0")
observation, info = env.reset()
action = env.action_space.sample()
observation, reward, terminated, truncated, info = env.step(action)
env.close()

Requires Python 3.10+. MuJoCo rendering may silently fall back to CPU unless EGL is properly configured on GPU systems.

Verify before relying

• Whether the package has active community support or issue response time beyond the maintenance status flag.
• Performance characteristics (steps/sec, memory usage) for typical training workloads.
• Compatibility with specific versions of dm-control, gymnasium, or mujoco beyond the declared runtime dependencies.

Package facts

Evidence: gym_aloha-0.1.4-py3-none-any.whl

Keywords: robotics, deep, reinforcement, learning, aloha, environment, gym, gymnasium, dm-control, mujoco

Tags

More Build Tools packages

Further reading