Real-to-sim robot simulation platform · Preparing for beta
Photorealistic robot simulation,
built in your browser
Bring real environments into simulation with Gaussian Splatting, share them with anyone, and deploy them as cloud gyms — pip install digiarena, five lines of code to connect your policy.
From real scenes to public benchmarks
Build scenes
In-browser editor: 3DGS scene reconstruction, image-to-mesh generation, lighting matched to the captured scene — no local GPU required.
Share
Scenes live under the {owner}/{scene} namespace. Public scenes can be browsed, forked, and reused by anyone.
Deploy as gym
Spin up a cloud simulation session in one click. Connect remotely with the digiarena SDK and a token — standard Gym interface.
Benchmark
Admins group public scenes into benchmark collections with server-authoritative scoring — an evergrowing robotics benchmark.
Five lines to a cloud gym
digiarena is a thin wrapper over world-engine-client: a standard reset/step 5-tuple, numpy-only with zero heavy dependencies. Deploy, connect, spectate, and teardown all go through the platform REST API — you can curl it without the SDK.
“Google Docs is to Microsoft Word” — what DigiArena is to desktop simulation software.
# pip install digiarena
import digiarena as da
gym = da.deploy(scene="alice/kitchen_001")
env = gym.connect()
obs, info = env.reset(seed=0)
obs, r, term, trunc, info = env.step(action)Featured public scenes
Community-shared simulation environments, ready to use.