Build simulation infrastructure for physical AI

Manifold Policy Evaluation

Evaluate any policy on any benchmark with one line.

LEARN MORE GITHUB

Stardust Synthetic Data

Train and test your perception models with photorealistic, multi-modal sensor data.

GET ACCESS

We’re getting AI into the physical world

Codifying the hardest tasks in simulation

Turn the world’s most demanding physical tasks into high-fidelity, sensor-accurate simulation.

Building with domain experts

Requirements come from real users operating in your domain.

Making simulation accessible to all

Open tools and synthetic data so any team can train, test, and evaluate physical AI.

We started where the stakes are highest, in production with the world’s most demanding teams.

NASA

BACKED BY

A living library of physical work, across every domain

MARITIME

SPACE

AERIAL

OFF-ROAD

ROBOTICS

OPERATIONS

MANIFOLD

The open-source evaluation platform for robotics

Run any policy on any benchmark. Scale to a thousand rollouts with a single CLI command.

manifold-cli

$ manifold run █

GITHUB READ THE DOCS

One harness for every major simulator

ISAAC LABMUJOCOMANISKILLGENESISSTARDUST

// evaluation today

✕A single eval rollout takes 24 hours or more
✕Every team reinvents the harness for every policy and every benchmark
✕No CI. Reproducibility is nearly impossible

PLACEHOLDER · RERUN VISUAL interactive 3D rollout playback

01One click, any benchmarkLIBERO, RoboCasa, or your own scenarios

02Live metricserror analysis as the run progresses

03Living leaderboarda standard, not a snapshot

04Open sourcethe standards layer can’t be proprietary

Manifold run detail: score, per-task pass rates, and clustered failure analysis

STARDUST

Synthetic data for perception and autonomy

Define a scenario in Python, or just describe it to your coding agent (Claude Code, Codex, any LLM). Stardust renders the rest: diverse, labeled, sensor-accurate data with rich time-series metadata, no 3D expertise required.

scene.ipynb

import bbi world = bbi.World()                      # initialise world stateworld.spawn("container_ship", quantity=12)   # place assetsworld.spawn("buoy", quantity=30, scatter=True) world.ocean(sea_state=4)                 # change the seaworld.weather(fog=0.3, time="dusk")      # change the weathercam = world.camera(preset="maritime_eo") # sensor preset imgs = world.render(frames=4000)         # render + auto-labelimgs.download(annotations=["bbox", "segmentation"])

BOOK A DEMO READ THE DOCS

SENSORS & LABELS

Multi-domain, photorealistic data matching your operational requirements

One scene, rendered across modalities in perfect registration, with ground-truth labels generated automatically for every frame.

RGB Photorealistic visible spectrum

IR Infrared and thermal bands

DEPTH Per-pixel depth and range

SEGMENTATION + BBOXES Pixel-perfect masks and 2D/3D boxes

NEURAL RENDERING

Achieve realism and diversity with AI post-processing

A learned post-processing pass closes the domain gap and multiplies diversity. One scene, every condition: weather, lighting, and time of day, all sensor-true, so models transfer cleanly to the real world.

Synthetic maritime scene, clear conditions — CLEAR

Synthetic maritime scene, overcast conditions — OVERCAST

Synthetic maritime scene, fog conditions — FOG

Synthetic maritime scene, dawn conditions — DAWN

Synthetic maritime scene, dusk conditions — DUSK

Synthetic maritime scene, rain conditions — RAIN

3D ASSETS

A 1,000+ object 3D asset library, and counting

Drag and drop from over a thousand production-ready, physically-accurate 3D assets, each tagged with real-world dimensions. Need something rare? Generate new assets with AI on demand, or request a custom asset from our 3D team.

Build physical AI at the speed of compute

From synthetic data to policy evaluation, Bifrost is the infrastructure for AI in the physical world. Tell us what you’re building.

EXPLORE MANIFOLD BOOK A DEMO