What we test, and what it actually measured
Short studies from our Scene→Sim→Skill pipeline: a real capture becomes a simulation, and a policy learns a skill inside it. One question per study, the numbers we got, and the parts that didn't work.
Research Studies
Study 04 · Sim→Skill A humanoid that wipes a counter with a cloth
A Unitree G1 learns an autonomous kitchen-wipe skill from 22 teleoperation demos. VBD deformable cloth, contact-rich manipulation, end-to-end pipeline — trained in a sim twin of a real kitchen.
Study 03 · Scene→Sim From an RGB video to a parametric kitchen
A real kitchen captured on a phone, turned into a watertight parametric CAD model — every cabinet, counter and shelf a separate solid — and loaded into FOI Sim as a training environment.
Study 02 · Sim→Skill A humanoid that walks — and dribbles — inside a Gaussian splat
A closed-loop Unitree G1 walking in FOI Sim native physics, dropped into a real garden reconstructed as a Gaussian splat. No animation, no keyframes — and the policy has never seen the ball.
Study 01 · Scene→Sim PPISP on a real 3-camera capture
NVIDIA just shipped PPISP. We put it through a clean ablation inside our Scene→Sim pipeline — same capture, same 30k iterations, one toggle. Drag to see the difference.
Reproduction Studies
Reproduction 002 · Open A humanoid pipettes in a lab. What does the video prove?
A viral clip carries a claim of 69% success across three tasks. It shows one task, once, with no tactile signal, no baseline and no primary source. We audit what the footage actually establishes — and open the experiment that would test it.
Reproduction 001 · Baseline We ran a published result again and got a different number
A released Diffusion Policy checkpoint reports 65.4% success on PushT. We evaluated the same checkpoint over 500 episodes and measured 62.0% — a gap of 1.6 standard errors. What we did, and what could explain it.
The Skill Layer for humanoid robots
Forenly AI turns real scenes into simulation where humanoids learn their skills — then transfers them onto the machine.