Today the humanoid story is really a supply-chain and data story: Jabil weighs in on manufacturing pace while Mecka AI banks $60M to capture human motion for training. Elsewhere, Nvidia and Helm.ai push foundation models toward physical AI, ETH Zurich reimagines what a robot even needs to be, and defense autonomy makes an uneasy appearance.

In today’s roundup

• Jabil discusses the pace of humanoid robot development and production — The Robot Report
  A contract manufacturer's view on humanoid progress is a useful reality check, since building the things at scale is a different problem than demoing them.
• Nvidia's big bet on physical AI aims for safer robotaxis, humanoid robots — Ars Technica
  Nvidia is pitching a full-stack safety layer for physical AI, betting that robotaxis and humanoids need trustworthy foundations before they earn their keep.
• Helm.ai reaches $70M in signed commercial contracts for its foundation models — The Robot Report
  Helm.ai's $70M in signed contracts suggests its unsupervised 'deep teaching' approach to world models is finding real commercial buyers, not just research interest.
• This Disembodied Hand Is All the Robot You Need — IEEE Spectrum Robotics
  ETH Zurich turned an off-the-shelf hand into a standalone walking robot, a fun reminder that robots needn't inherit the human body plan.
• 5 days to TechCrunch Disrupt 2026: Don’t pay more at the door for your pass — TechCrunch Robotics
  Disrupt 2026 lands in San Francisco October 13-15, so grab a pass now if you'd rather not pay more at the door.
• Robot data startup Mecka AI nabs $60M from Sequoia — TechCrunch Robotics
  Sequoia's $60M bet on Mecka AI underscores that the scarce resource in humanoid training isn't hardware but good human motion data.
• Dassault-backed startup unveils strike weapon designed to hit targets up to 1,240 miles away — Interesting Engineering
  Harmattan AI's IRIFI pushes autonomous long-range strike to 1,240 miles, a sobering marker of where defense autonomy is heading.
• MimicX: Policy-in-the-Loop Supervision Refinement for Video-Driven Humanoid Motion Tracking — arXiv cs.RO
  MimicX tackles a real headache in video-driven humanoid learning: references that look plausible but fail under physics, refining supervision with the policy in the loop.

Read this roundup on RobotCafe →

The Daily Grind · RobotCafe · robotcafe.com