Skip to main content
Eden

Embodied-AI Development Environments Nexus

License Python Genesis 1.4.6 Docs

Installation · Quick start · Gallery · Documentation · Citation

Eden teaser: a G1 skateboarding, a G1 filling a cup at a tap, two hands playing piano, sixteen hands reorienting cubes, G1 squads crossing terrains (stairs, pallets, stepping stones and a step down), a Franka writing EDEN on a whiteboard, a close-up of fin-ray fingers grasping an apple, a UR5e inserting a peg in a workshop (OmniReset), a lineup of humanoids, quadrupeds and arms, a LimX Oli with two Wuji hands moving a chair, a batched splat render of sixteen envs, and a lineup of dexterous hands

Eden is a robot-learning framework built on genesis-world. Write any robot-learning task as one typed config, train it across thousands of GPU-parallel environments, then run the same policy on a real robot.

Highlights

  • One typed config per task: scene, MDP and learning runner in one Pydantic object; variants become CLI flags.
  • 245 composable MDP terms, searchable with eden terms --grep <text>; any decorated function becomes a new term.
  • 47 robots and 11 composite rigs: humanoids, quadrupeds, arms and dexterous hands.
  • 66 tasks, from locomotion and piano to calligraphy, plus robosuite, LIBERO, SIMPLER and PolaRiS ports.
  • Multi-physics: every Genesis solver, plus Mochi for coupled rigid, cloth, soft-body and rod dynamics.
  • Sensing and rendering: RGB-D, lidar, tactile, IMU and audio; batched Gaussian-splat scenes and path tracing.
  • Motion data and teleop: retarget human motion, teleoperate, and record HDF5 or LeRobot datasets.
  • Train and deploy: rsl-rl, rl_games, skrl or SB3 on multi-GPU DDP, then eden deploy to hardware or MuJoCo.
  • 21-policy zoo: SONIC, BeyondMimic, π0.5, GR00T, … with eden inference --policy <name>.
  • Built for coding agents: strict configs, error messages that name the fix, and a shipped AGENTS.md.

Installation

Eden needs Python 3.12+ and runs on Linux, macOS (Apple silicon) and Windows. Linux with an NVIDIA GPU is the primary platform; macOS and Windows are best-effort, and a few extras need more: Nyx path tracing has no macOS build, Gaussian-splat rendering needs CUDA, and the Wuji Glove SDK is Linux-only. We use uv:

git clone https://github.com/embodied-ai-nexus/Eden.git
cd Eden
uv sync --all-extras --all-groups # the groups add the extras' git-only packages
source .venv/bin/activate         # Windows: .venv\Scripts\activate

uv sync alone installs the core; the individual extras (vis, nyx, batch-splat, gello, pico, wuji, …) are listed in the installation guide.

Quick start

eden list                                     # every registered task and its variant flags
eden train --task reacher                     # train a Yam arm to reach, 4096 envs
eden inference --task reacher --checkpoint <path/to/model.pt> --vis
eden inference --policy sonic --vis           # roll out a published policy, no task needed
eden terms --grep friction                    # find a term and its parameters

Sim-to-sim, sim-to-real and export:

eden deploy --task <task> --checkpoint model.pt --mujoco    # sim-to-sim check
eden deploy --task <task> --checkpoint model.pt             # real hardware
eden export --task <task> --checkpoint model.pt             # ONNX / TorchScript + policy_spec.json

A task is one config

@TASK_REGISTRY.register(modifiers=(ReacherDifficultyMod,))  # → eden train --task reacher --difficulty hard
class ReacherConfig(EdenRLConfig):
    env_options = EnvOptions(num_envs=4096, episode_length_s=9.0, sim_dt=0.01, ...)
    scene_options = en.scenes.Empty(robot=en.robots.Yam())
    observation_options = ObservationManagerOptions(
        policy=ObservationGroupOptions(
            dofs_pos=observations.DofsPos.configure(entity_name="robot"),
            dofs_vel=observations.DofsVel.configure(entity_name="robot"),
        ),
    )
    action_options = ActionManagerOptions(
        joint_pos_controller=actions.ImplicitPDController.configure(
            entity_name="robot", dofs_name="joint*", scale=0.05,
        ),
    )
    command_options = CommandManagerOptions(reaching_target=UniformT3Command.configure(...))
    reward_options = RewardManagerOptions(
        reaching=RewardTerm.configure(func=reaching_reward, weight=1.0, params={...}),
        action_rate=RewardTerm.configure(func=rewards.action_rate_l2, weight=-0.01),
    )
    termination_options = TerminationManagerOptions(
        timeout=DoneTerm.configure(func=terminations.time_out, time_out=True),
    )
    runner_options = RslRlOnPolicyRunnerOptions(...)

The full version is eden/tasks/benchmark/reacher/config.py, the template every task follows.


Sensor
Sensors: a Go2 with LiDAR in a scanned tunnel

Engine
Multi-modality: playing Chopin with sound synthesised from the pressed keys

Engine
Beyond rigid: a G1 fills a glass with water at a kitchen tap (1.5x)

Rendering
Batch rendering: batched splat + heterogeneous sim render
A lineup of Eden's built-in robots moving in place: humanoids, quadrupeds, a wheeled quadruped and arms
Entities
Embodiments: 47 robots and 11 rigs out of the box, with 950+ objects
Nine published controllers in Eden: SONIC, HUSKY, BeyondMimic, TWIST2, ProtoMotions, AMO and HOMIE on a G1, MetaMotivo on an SMPL humanoid, and SimToolReal on a KUKA iiwa14 with a SharPa hand
Policies
Policy zoo: published controllers run as shipped: eden inference --policy <name>
Nine motion datasets replayed in Eden: AMASS, LAFAN1 and SEED retargeted to a G1 (dance, fall and get-up, crawl), PHUMA kung fu on a G1, Weave G1 with Inspire hands lifting a table, OMOMO carrying a suitcase, HHI-Assist caregiver helping a person sit up in bed, ARCTIC hands opening a box, and DexYCB grasps of a drill, a mustard bottle and a mug
Data
Motion datasets: shipped with loaders for 12+ datasets, shared one motion format.

Data
Cross-embodiment retargeting: eden retarget for retargeting the data
Human-object demonstrations retargeted with the object: OMOMO box carrying onto a G1 and a LimX Oli with Wuji hands (OTRetarget), and DexYCB grasps onto an XHand and a SharpaWave hand
Data
HOI retargeting: retarget with an object for a rig and hands
massively parallel training of G1 in EDEN
Training
Train at scale: eden train --task <task> --num_envs 4096, across GPUs with DDP
Three policies in MuJoCo sim-to-sim: a Unitree G1 boxing under SONIC, a LimX Oli walking across boards and a Wuji Hand 2 spinning a pen
Deploy
Sim-to-sim: eden deploy --task <task> --checkpoint model.pt --mujoco
The same three policies on the real robots: the G1 in a gantry frame, the Oli on a safety tether and the Wuji Hand 2 on its wrist mount
Deploy Real
Sim-to-real: eden deploy --task <task> --checkpoint model.pt

RL
Motion tracking: native motion manager stack

RL
Human–object interaction: a G1+Inspire and a Oli+Wuji move OMOMO objects

RL
Multi-agent RL: flexible observation, action, reward design

RL
Soft-body manipulation: a trained policy wipes a bar with a foam sponge (1.5x)

Scripted
Cloth manipulation: two arms fold a towel in half (3x)

Scripted
Cable routing: a cable routed round two spools into a port (8x)

Scripted
Calligraphy: 森羅万象 in kaisho with a modeled Chinese brush (24x)

Scripted
Wet-acrylic painting: The Starry Night with the wet paint blending (90x)

Benchmark
Benchmark suites: LIBERO, robosuite, SIMPLER, PolaRiS, etc., reimplemented

Documentation

The documentation covers concepts, every guide and the full API. To build it locally:

uv sync --inexact --extra docs
uv run mkdocs serve        # → http://127.0.0.1:8000/Eden

Good places to start: task configs, writing terms, entities, training runs and deployment. Runnable scripts live in examples/ (eden examples list), each with its own README.

Contributing

Issues and pull requests are welcome. Before you open a PR:

pre-commit install                     # ruff + formatting on every commit
uv run pytest tests/<path>/test_X.py   # tests mirror eden/; run the files you touched

AGENTS.md holds the rules every change follows (and is what coding agents read first); docs/guides/testing.md explains how tests are written.

Contributors

A huge thank you to everyone whose early feedback and contributions have helped shape this project!

Kashu7100
Kashu7100
lalit-jayanti
lalit-jayanti
Milotrince
Milotrince
pcseanlo
pcseanlo
thanyu-hub
thanyu-hub
hughperkins
hughperkins
jasonzhou32
jasonzhou32
ZuWang95
ZuWang95
YutoShibata07
YutoShibata07
yituoren
yituoren
Akihisa-Watanabe
Akihisa-Watanabe
dhruv0000
dhruv0000
ziyin-xiong
ziyin-xiong
yxKryptonite
yxKryptonite
duburcqa
duburcqa
YilingQiao
YilingQiao

Acknowledgments

Eden builds on these open-source projects:

  • genesis-world: high-performance cross-platform simulation backend.
  • quadrants: Python compiler that generates parallel kernels for Nvidia/Vulkan/Metal GPUs and x86/ARM CPUs.
  • IsaacLab and mjlab: reference manager-based environment implementations.
  • ManiSkill: reference holistic environment for embodied agents.
  • ProtoMotions and MimicKit: reference environments for human motion priors.
  • Holosoma and InstinctLab: reference end-to-end humanoid frameworks.

Citation

If you use Eden in your research, please cite:

@misc{Eden,
  author = {Eden Authors},
  title = {Eden: One Task Definition from Simulation to Real Robots},
  month = {November},
  year = {2026},
  url = {https://github.com/embodied-ai-nexus/Eden}
}
@misc{Genesis,
  author = {Genesis Authors},
  title = {Genesis: A Generative and Universal Physics Engine for Robotics and Beyond},
  month = {December},
  year = {2024},
  url = {https://github.com/Genesis-Embodied-AI/Genesis}
}

License

Eden is released under the Apache 2.0 License.

Metadata

Release files for eden-world 0.1.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for eden-world 0.1.0
File Size Uploaded Provenance Actions
eden_world-0.1.0.tar.gz 3.2 MB With provenance Details for eden_world-0.1.0.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for eden-world 0.1.0
File Interpreter ABI Platform Actions
eden_world-0.1.0-py3-none-any.whl Python 3 none any Details for eden_world-0.1.0-py3-none-any.whl

Total release size: 6.8 MB

Release files / eden_world-0.1.0.tar.gz

Download URL eden_world-0.1.0.tar.gz
Size 3.2 MB
Tags Source
SHA-256 checksum
How to use checksums
87e8390e2bcefad789637172c6268470e491f79543e50fec5c9ce9d855e33ff1
BLAKE2b-256 checksum
How to use checksums
da90f3554a833b6e8e201454a44f3200cddfa54ff74980edf803c16888f00ab6
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 12, 2026.

Transparency log

Release files / eden_world-0.1.0-py3-none-any.whl

Download URL eden_world-0.1.0-py3-none-any.whl
Size 3.7 MB
Tags Python 3
SHA-256 checksum
How to use checksums
f796458383b8138cd759f999b7edbe5db96e924d921639e25c4ed0f341e03c4c
BLAKE2b-256 checksum
How to use checksums
7f791033b94981dff33fe9121c753a30e56bea3b5456fd241a6ba8567d098973
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
Yes
Uploaded via twine/7.0.0 CPython/3.13.16

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 12, 2026.

Transparency log

Security for eden-world 0.1.0

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

Full provenance

Both files in this release have been uploaded with provenance metadata.

Select from 2 files

Attestation for eden_world-0.1.0.tar.gz

PyPI Publish Attestation

PyPI verified that this artifact, at this checksum, originated from the publisher listed below.

Signed by GitHub Actions, verified by PyPI on Nov 12, 2026.

Transparency log

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Release history Release notifications | RSS feed

This release

0.1.0 This release

2 release files Full provenance

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page