Accelerate Robotics Research
An evaluation platform that delivers insights in 30 mins
Trusted by teams building mission critical physical AI systems
Massively Parallelized GPU Evaluation
Run thousands of GPU-accelerated simulations in parallel and get your results in minutes, not overnight.
Atomic Subgoals & Custom Rubrics
Go past pass-fail metrics and see exactly where your models break down, and where they hold up. Define your own rubrics to grade rollout quality.
Agentic Failure Analysis
Agents watch the rollouts, cluster the failures and rank them by impact, so you're not scrubbing through hundreds of videos to find the pattern yourself.
Benchmarks & Environments
Works with any simulator, benchmark or world model. We work with industrials to codify real world work into custom simulation tasks along with nuanced success rubrics.
Make Your Research Loop 100x Faster
We're AI and robotics researchers, graphics engineers and infrastructure engineers. We've spent the last five years building our own physical AI systems and helping some of the world's largest companies deploy theirs. Now we're making that far easier for everyone else.
What is Manifold?
Manifold is a robot policy evaluation platform. One harness runs your policy on every major simulator benchmark, including LIBERO, RoboCasa and your own scenarios, sharded across GPUs so results land in minutes instead of days.
Which models and benchmarks does Manifold support?
Manifold evaluates VLA and embodied AI policies, including OpenVLA, GR00T, pi-0 and Octo class models, on LIBERO, RoboCasa, RoboMimic and CALVIN, with more benchmarks and embodiments on the way.
