PI’s π0/0.5 tops a robotics benchmark, with NVIDIA’s GR00T close behind

chris_j_paxton · x · 2026-07-30

A quoted robotics benchmark says π0/0.5 from Physical Intelligence wins overall, with NVIDIA Robotics’ GR00T not far behind.

It also notes that diffusion and ACT, the only non-VLA policies trained from scratch, are competitive with MolmoACT and SmolVLA. But the benchmark author cautions that headline scores do not tell the whole story, implying that capability comparisons across robot policies are still incomplete.

Related event: π0.5 Tops Robot Policy Benchmark Amid Evaluation Concerns(3 posts)→

Original post →

More from Embodied

Embodied channel →