ACQUIRING TARGETS...

· DroidHunt news

Humanoid Robotics Morning Brew - October 16, 2026

  • Agility Robotics taps UK battery maker Nyobolt for Digit 5: a 9-minute charge delivers 90 minutes of runtime, a 10:1 run-to-charge ratio, enabling 20+ working hours a day.
  • Ex-Huawei Xiaoyi LLM chief Tang Duyu launches Guanxing Shuzhi, a startup betting on 'proactive' physical AI that decides when to act, not just how.

# Agility picks UK battery maker Nyobolt for Digit 5: 9-minute charge, 90-minute runtime, 10:1 run-to-charge

Agility Robotics announced (Oct 15) that Nyobolt, a Cambridge (UK) ultra-fast-charging battery maker founded in 2019, will supply Digit 5's battery system. The pack delivers 90 minutes of runtime and charges in 9 minutes — a 10:1 run-to-charge ratio, up from 2:1 on Digit 4 — enabling more than 20 productive hours in a 24-hour day, or over two full shifts. "For our customers, Digit's uptime is the business case," said chief business officer Daniel Diez. Agility cites more than 65,000 hours of real-world Digit operation. Early access to Nyobolt-equipped Digit 5 begins H1 2027, with broader availability over the following six to nine months (DroidHunt, Sept 16 — Digit 5 unveil).

Why it matters: Charging downtime is a direct tax on humanoid fleet economics. A 10:1 run-to-charge ratio effectively doubles useful deployment per robot and raises the bar on utilization claims for warehouse RaaS competitors.

Sources: prnewswire.com therobotreport.com

Tags: agility-robotics components us confirmed

# Ex-Huawei Xiaoyi LLM chief Tang Duyu launches Guanxing Shuzhi, betting on 'proactive' physical AI

QbitAI profiled (Oct 15) Guanxing Shuzhi (观行数智), the new startup of Tang Duyu — former Huawei AI chief expert and R&D head of the Xiaoyi multimodal LLM (level-21), previously at Microsoft Research Asia (CodeBERT, CodeGPT). The company is building "proactive physical AI": instead of executing human instructions, its dual-model ProAgent stack reads a continuous event stream — a lightweight StreamingJev module flags when intervention is needed, then a Response Model plans the action. It also released ProactiveXBench, a benchmark scoring timing, progress and outcome of unprompted interventions. Team-reported results claim ProAgent beats GPT-6 on intervention-timeliness metrics; the system is being validated on a Reachy Mini robot across 60 cases. Funding details were not disclosed.

Why it matters: The industry's models are converging on "how to execute"; "when to act" is largely unclaimed. Whether proactive agency is a real product or a research framing, a Huawei-grade LLM team taking it on marks a new front in embodied AI competition.

Sources: qbitai.com

Tags: guanxing-shuzhi foundation-models china reported

Watch next

  • Zhengzhou components expo closes Oct 17; Tesla Q3 earnings Oct 21 (Optimus commentary expected); Agility–CCXI shareholder vote Nov 2.
An unhandled error has occurred. Reload 🗙