Datasets & Models
OpenVLA
One of the first open VLA models to be widely reproduced: 7B parameters, MIT licensed, and runnable on consumer GPUs once quantised. The distance between reported success rates and what different people get on real hardware is the most privately discussed and least publicly recorded fact in the field.
Where it sits on a build
- The road to a humanoid: why it is not the first machine you buyToward a humanoid: how the body goes together · №8·Toward a humanoid: system architecture (signals and power) · №2
Claim Check
Claim Check · B
How much of a LIBERO score measures the environment instead of the model
A benchmark is doing two jobs at once: measuring capability and being gamed. When randomisation is thin, the second on…
LIBERO benchmark · OpenVLA · 2026-09-28
Claim Check · B
'The simulation score is high, so it will work on hardware': where that breaks
Treat simulation as a filter, not a promise. A policy that scores well in sim is worth trying on hardware; do not copy…
MuJoCo · LIBERO benchmark · OpenVLA · 2026-09-28