ACQUIRING TARGETS...
· DroidHunt news
Agility Robotics unveiled Digit 5 on Sept 15 in Salem, Oregon — its next-generation humanoid and the first engineered for "cooperatively safe work at scale," designed to operate in close proximity to people without the physical barriers that have fenced off every prior deployment. The safety stack combines AI-based human detection (autonomously avoid, stop, or sit down), visual and auditory intent cues, and an independent safety controller. Agility is the first launch partner for NVIDIA's Halos for Robotics platform, pairing its systems with IGX Thor and Halos Core infrastructure. Hardware upgrades, built from three years of customer feedback: 40% more payload with repetitive 50 lb (22.7 kg) lifts — covering OSHA single-person lift tasks — via new legs and proprietary cycloidal actuators; 90-minute runtime with a 9-minute recharge (10:1 run-to-charge ratio, enabling 20+ productive hours/day); 1.81 m tall, 129 kg, 2.2 m reach; swappable ISO-flange end effectors. The installed base has logged 65,000+ hours at GXO, Amazon, Schaeffler and Toyota Motor Manufacturing Canada — at GXO, Digit moved ~100,000 totes at ~98% accuracy — and Digit 4 was the first humanoid to pass an independent field evaluation against OSHA-administered industrial safety standards on a customer production line.
Why it matters: Commercially, Agility cites $300M+ in multi-year orders (subject to contractual milestones), early access in H1 2027 and general availability by end-2027, with first availability outside North America (EU/UK); RoboFab is being retrofitted to build up to 10,000 units/year, with illustrative pricing of ~$8,500/month RaaS or ~$200,000 upfront. The launch lands days after its SPAC S-4 (covered in our Sept 11 digest) disclosed $1.8M of 2025 revenue against a $2.5B valuation — the $300M order book is the forward claim to watch. Agility is also a project leader on ANSI/A3 TR R15.108 and a contributor to ISO 25785-1, the first international humanoid safety standard. Safety certification, not intelligence, has been the binding constraint on humanoid deployment — Digit 5 is the first direct attack on it.
Sources: morningstar.com finance.yahoo.com the-decoder.com
Tags: agility-robotics product-launch safety nvidia deployment
Hangzhou-based Zeno Robotics (芝诺机器人, founded 2025) has closed a seed round of hundreds of millions of RMB, 科创板日报 reported Sept 16. Investors include Sunny Optical (舜宇光学), MetaX (沐曦股份), 中网银新, 赛意产业基金 and 和利资本. The company — which calls itself a foundation-model-first "physical intelligence" company — is already running overseas PoC (proof-of-concept) tests and targets scaled deployment starting H1 2027. Alongside the round it released Zeno-1, a 3B-parameter "collaborative intelligence" foundation model with a decentralized architecture: robots decide autonomously from vision and interaction history with no central controller, with cross-scenario, cross-task and cross-embodiment adaptation. Real-robot validation shows 10+ minutes of continuous operation across 8 subtasks, including multi-robot coordination.
Why it matters: The team is the tell: co-founder and chief scientist Zhi Weiming (植伟铭) is a tenure-track assistant professor at the University of Sydney; chief robot architect Zhang Ji (张楫) is the author of LOAM, the canonical lidar odometry-and-mapping algorithm; much of the core team comes from CMU's Robotics Institute. Context: TrendForce pegs China's 2026 humanoid market at ¥15B with 60%+ growth in 2027.
Sources: 163.com
Tags: zeno-robotics funding foundation-models embodied-ai china
XPeng is expanding its self-developed Turing AI silicon beyond cars into humanoid robots, treating the same chip family as shared compute for its broader Physical AI stack, Pandaily reports (building on DigiTimes and company materials). IRON configurations are described as running about three on-device Turing chips for roughly 2,250 TOPS of effective compute — enough, the company argues, to keep its VLA (vision-language-action) foundation model local, cutting latency and cloud dependence. The strategic logic is amortization: one silicon program across EVs and humanoids, reusing automotive-grade process, packaging, supplier relationships and driving-hardened software toolchains. IRON carries 76 degrees of freedom, 21 per hand. The move follows XPeng commissioning its automated IRON production line in Guangzhou (covered in our Sept 9 digest), with the company signaling a 1M-units-per-year ambition for 2030.
Why it matters: Chip-level vertical integration is becoming a differentiator among humanoid players. What to watch: whether advertised TOPS translate into sustained on-device VLA performance in third-party benchmarks, and whether Turing's toolchains hold up as robot workloads diverge from highway driving.
Sources: pandaily.com hannovermesse.de
Tags: xpeng chips supply-chain humanoid vla
Wuxi-based Quanzhibo (泉智博), an integrated joint-module supplier, has closed a Series B of hundreds of millions of RMB led by Yida Capital (毅达资本), 国泰海通开元 and 西证投资, with existing backers 光合创投 and 无锡新投 topping up. Founded in 2023, it has now raised eight rounds in under two years. The company develops all three reducer routes in-house — planetary, harmonic and cycloidal — plans million-unit annual capacity, and counts Zhiyuan, Leju, Songyan and Galaxea among its customers. Proceeds go to capacity ramp, next-generation high-torque-density joints and a global service network. Founder/CEO Chen Wankai (陈万楷) previously did robot R&D at CETC's marine IT institute and the Peking University IT institute in Zhejiang.
Why it matters: This is the second integrated joint-module mega-round in two days, after ENCOS's ¥300M+ Series B (covered in our Sept 15 digest) — capital keeps concentrating on component suppliers with proven volume capacity rather than full-robot brands.
Sources: m.sohu.com cbgc.scol.com.cn
Red Star Capital Bureau (红星资本局) tallies a September model blitz: Songyan Power launched its Scalabot embodied-tech sub-brand with the HERON-World Model on Sept 8, then on Sept 15 released HERON-CRA — focused on memory, reinforcement learning and cross-embodiment generalization, letting robots "remember the past and learn from mistakes." Zhiyuan shipped two models in one day (Sept 9): AGILE 2.0 (perception-control) and GE-Act 2.0 (world-action); Unitree fully open-sourced UnifoLM-WLA-1.0 on Sept 10 (covered in our Sept 11 digest). Routes are diverging: Zhiyuan is scaling real-world data from 300 to 30,000 hours to unlock fine manipulation (towel folding, cup nesting, flower arranging); Fourier is using large models for intent understanding and task planning — its GR-3 demo fetched water on the vague prompt "I'm thirsty."
Why it matters: The money explains the rush: of ~¥43.8B in China embodied-AI funding in H1 2026, brain-centric companies took ¥22.25B (50.8%) while body makers drew just ¥5.6B (12.8%). "Show muscle" demos are out; model capability is now the fundraising currency.
Sources: finance.eastmoney.com
The Mech-Mind vs. Galaxy General "fake revenue" feud (our Sept 11 digest) and the reported CSRC window-guidance tightening on humanoid IPOs (Sept 14 digest) have escalated: 观察者网 reported Sept 16 that IPO hopefuls Qianxun Intelligence (千寻智能), Galaxea (星海图) and Robot Era (星动纪元) face "key tests" this month, with at least one peer reportedly filing reports about them to listing reviewers. The content of the reports is not public — treat the referral claim as single-source and unverified. Market backdrop: Unitree closed Sept 16 at ¥474.28, a ~¥191.8B market cap — more than ¥250B evaporated from its post-IPO peak (the slide began with the break below ¥500 we noted Sept 11). Meanwhile 28+ embodied-AI companies are pursuing listings (7 formally accepted, 5 in tutoring or filing stages, 16 preparing), and H1 saw 288 funding events across 226 companies with ¥46B+ disclosed.
Why it matters: With regulators demanding sustainable revenue, the sector has tipped from land-grab to zero-sum — proving a rival's revenue is inflated makes clean revenue relatively more valuable. Expect more disclosure fights as IPO windows narrow.
Sources: guancha.cn znyj.ofweek.com
Tags: unitree regulation ipo china galbot
Zhiyuan partner Yao Maoqing (姚卯青) told 财联社/科创板日报 in an interview published Sept 15 that the company's cumulative robot production has gone from 6 units to 15,000 in three years, organized into seven deployment categories: line-side loading/unloading, industrial handling, logistics sorting, guided tours, service retail stations, security patrol and commercial cleaning. Yao's framing on the bubble debate: embodied AI today is roughly where NEVs (new-energy vehicles) were circa 2015 — fast-growing with capital still flowing in — and the whole sector's ~¥100B-scale funding remains small against the trillion-RMB annual AI-infrastructure capex of large-model companies.
Why it matters: Context to keep in view: Zhiyuan is also the largest supplier to the data-collection centers now under regulatory scrutiny (~200 units deployed there; our Sept 14 digest), so cumulative production is not the same as end-customer demand.
Sources: cls.cn
Tags: agibot production deployment china
← Newer: Sep 17, 2026 Older: Sep 15, 2026 →
Home Every digest Robot catalog Drones on DronePriceHunt Privacy & affiliate disclosure RSS