PRESENTED BY
Tesla / USPTO application excerpts, arranged for layout.
A $300 million robot order sounds straightforward. Agility’s investor day showed how much depends on what the robots can actually do. Each new use case Digit 5 proves releases another batch of orders. The session also put numbers on launch costs and a clearer timeline on availability.
That question—how to turn capability into dependable work—runs through this week’s issue. Musk explains why Optimus’s hands are so difficult to get right, Figure is reportedly in funding talks with Nvidia, and XPENG has set a European rollout target for IRON. Further down, our Aether and Monolithos interviews explore how robots can check their actions and retain lessons from experience.
IN THIS EDITION
Agility’s numbers: launch costs, delivery timing and the conditions behind its $300M order.
Hardware and capital: Musk on Optimus’s hands, Tesla’s patent, Nvidia–Figure talks and Microagi’s revenue claim.
The rundown: Unitree’s GPT-6 optimism, Boston Dynamics’ new CEO, Ultra’s $50M raise and XPENG’s Europe plans.
Our interviews: Aether on checking actions; Monolithos on remembering experience.
LEAD STORY
Agility says Digit 5’s parts will cost about $150,000 at commercial launch, versus roughly $125,000 for Digit 4. Extra sensors and safety computing help explain the increase. Its projected launch-era product margin under robotics-as-a-service is around 50%; the 70% figure depends on reaching 1,000 units a year and excludes corporate costs.
The unnamed customer behind the $300 million, 1,000-robot commitment receives robots in tranches as Digit proves agreed use cases. The contract covers three years of service spread over four, so revenue depends on deployments. Early-access units are now expected in the first half of 2027, with general availability described as late 2027 and early 2028.
One customer detail stood out: Schaeffler says it still needs guarding around its humanoids until they are certified to safety standards. That limits which jobs they can take on. Proving skills, lowering costs and meeting safety requirements all sit between the order book and a wider rollout.
ADVERTISEMENT · REFERRAL PARTNER
See China’s robotics ecosystem first-hand
Illustration from the organiser’s tour deck; not a confirmed tour location.
Experience China Speed · November 29–December 4, 2026
Join a six-day executive tour across Beijing, Guangzhou and Shenzhen, organised by Jee Chee Ren and guided by Rich Robinson. The programme spans AI, humanoid robotics, autonomous mobility and manufacturing, with company visits and executive briefings. Specific visits are subject to confirmation.
Humanoids Daily referral offer, October 12–26: five referred seats at US$8,500 and a further ten at US$9,000, against a US$10,000 standard rate. Availability and your final quote are confirmed by the organiser; full payment secures a discounted seat.
Humanoids Daily earns a commission on referred bookings. An enquiry does not reserve a seat or lock in a price. The tour is operated by Jee Chee Ren (Inner Circles Limited).
HARDWARE & CAPITAL
Tesla’s newly published application describes flexible sensor arrays that conform to fingers, palms and other curved surfaces. Much of it concerns how to manufacture them, including printing conductive patterns and molding layered sensors into shape.
At X Takeover on October 10, Musk put the challenge in everyday terms: a hand must handle a tiny screw, apply torque and stay precise. Touch sensors have to survive repeated bending, while rubber finger coverings must grip without slipping around the underlying structure—and still be replaceable when worn.
He said the initial production ramp would be slow and attributed a two-to-three-month setback last year to motor-magnet supply restrictions. The patent helps explain Tesla’s engineering work, but neither it nor Musk’s remarks establishes which sensor design Optimus uses or that the durability problems are solved.
The Information reports that Nvidia has discussed investing another $1 billion in Figure, which is reportedly seeking financing at about a $38 billion pre-money valuation. No deal has been announced. The talks follow Figure’s multibillion-dollar Nscale compute commitment and its Index human-video collection platform.
CEO Bercan Kilic says Microagi reached the milestone six months after launching Shift. The company labels it “ARR,” but its graphic defines it as an annualized run rate. It has not disclosed the calculation or the split between data and robot-deployment revenue. The claim signals commercial ambition, but does not tell us how many robots are working at customer sites.
PRESENTED BY ROBOSTRATEGY
Most of the companies we usually cover in this newsletter are still private, which puts them out of reach for public-market investors. RoboStrategy, Inc. ($BOT) is built to give you access to several of them.
Investing involves risk, including possible loss of principal. Read the prospectus before investing.
THE RUNDOWN
Unitree’s CEO grows more optimistic after GPT-6 tests. Wang Xingxing says tests across robots and tasks were encouraging and suggests a tipping point could arrive within one or two years in a fast scenario. He did not disclose tasks, model configuration or success rates.
Boston Dynamics hires former Amazon AI leader Rohit Prasad as CEO. His background spans Alexa and the Nova models. The appointment puts AI integration and commercial scale at the center of the next chapter for the maker of Atlas, Spot and Stretch.
Ultra raises $50M for warehouse robots. The Series A brings announced funding to $62M. Its Physical Intelligence-powered OP1 systems already pack customer orders, with human help when needed. Ultra says its robots have packed more than half a million orders.
XPENG targets Europe for IRON in H2 2027. The plan follows a late-2026 mass-production target and deliveries beginning in H1 2027. Initial uses center on sales centres and company sites, including tours, shopping assistance and patrol support. European pricing and launch countries remain unspecified; this is a rollout target, not a home-robot sales commitment.
OUR INTERVIEWS
A robot can finish a movement without achieving its goal. Aether’s CRIS-0 pairs a model that predicts the consequences of actions with an agent that checks the outcome, then retries or replans when needed.
In our written interview, Aether described using roughly 20,000 hours of pretraining data and only “hours” of robot-specific adaptation. The household demonstrations remain controlled tests, and the company identifies contact-rich manipulation and unfamiliar environments as major limitations.
When a human corrects a robot, what should the next attempt remember? Founder Silas Yu told us how Monolithos’s Robot Brain alpha records failures, corrections and outcomes, then supplies relevant experience to an existing planner or recovery system.
The memory can be revised when new evidence challenges an earlier lesson. But physical-robot results are still pending, and Yu says a performance advantage over well-built log retrieval or hand-written recovery rules has not yet been established. Monolithos is seeking robotics teams to test it inside their own software stacks.
ADVERTISEMENT · AFFILIATE PARTNER
In Munich on November 5–6?
Humanoid Europe 2026 brings robotics builders, researchers and industrial users together at the Holiday Inn Munich – City Centre to discuss moving from demos to deployment. If the dates fit, take a look at the programme and registration options.
Humanoids Daily earns a commission if you buy a ticket through our link.
This edition of Humanoids Daily was presented by RoboStrategy.
Seeing something before I do? Just hit reply — that's where a lot of these stories start.
P.A
Humanoids Daily
Know someone following the robotics industry? Forward them this briefing—it helps Humanoids Daily reach more people.
Received this from a friend? Subscribe to the free weekly briefing.