Dispatch #10: Hyundai Buys All of Boston Dynamics, Europe Gets a Humanoid Unicorn, and Tesla's Robotaxi Plateau

Welcome back to The Robotics Dispatch. Last time, at the mid-year mark (#9), we tallied $55.8 billion in robotics funding for the first half of 2026 and asked whether the money and the trip counts were as solid as the headlines made them sound. This fortnight brings a different kind of story: not a new fundraise chasing a valuation, but two of last year’s biggest robotics bets — one on hardware, one on industrial arms — settling into ownership, and a very public reminder that “deployed” and “growing” are not the same word.

The lead: Hyundai now owns all of Boston Dynamics

On July 16, Hyundai Motor Group confirmed it is buying SoftBank’s entire remaining stake in Boston Dynamics for $325 million, making the robot maker a wholly owned subsidiary for the first time (Bloomberg; Hyundai Motor Group newsroom). SoftBank had held onto roughly 10% of the company since Hyundai took an 80% controlling stake back in 2021, and it exercised a put option built into that original deal — a contractual right to force the sale — rather than continue holding a minority position (Yahoo Finance). The price values Boston Dynamics at about $3.3 billion, the same figure used in the 2021 deal; some analysts note that a fresh valuation today would likely be considerably higher, given how far Physical AI investment has run since then.

Hyundai’s own statement frames this less as a hardware acquisition and more as the last piece of an “End-to-End AI Robotics Value Chain,” combining Boston Dynamics’ engineering with Hyundai’s factories, supply chain, and mobility business under the banner “Partnering Human Progress.” Concretely, the company says it plans to put Boston Dynamics’ Atlas humanoid to work at its Metaplant America facility in Georgia starting in 2028, beginning with parts sequencing before expanding into broader manufacturing tasks by 2030 (Hyundai Motor Group newsroom). That is a notably longer runway than some competitors are promising — a useful reality check on how much of “humanoids in the factory” is happening now versus being planned for later this decade.

Three robotics deals confirmed or advanced this fortnight, compared by deal value. The ABB and Boston Dynamics figures are enterprise/company valuations from acquisitions already in progress; the Humanoid figure is a fresh venture-funding valuation.

A new humanoid unicorn, and it isn’t in California

Five days later, on July 21, the London-based startup Humanoid announced a $152 million Series A at a $1.35 billion post-money valuation, led by Prime Movers Lab with Schaeffler, Bosch, Fubon Financial Holding Venture Capital, and Aglaé Ventures also participating (Forbes; tech.eu). Founded by Artem Sokolov barely two years ago, the company bills itself as Europe’s first pure-play humanoid robotics unicorn, and its total raised now stands at $270 million.

Humanoid’s first product, the HMND 01, is a wheel-based (not bipedal) robot — a pragmatic choice that trades away leg mobility for a simpler, more stable platform to reach factories sooner, with a bipedal version reportedly still in development. It runs on the company’s in-house KinetIQ software stack, and the round follows what Humanoid calls the industry’s largest publicly announced commercial agreement to date: a deal with German auto-parts giant Schaeffler covering thousands of robot deployments. The company says the new capital will fund next-generation hardware, mass manufacturing of the wheeled platform, and commercial deployments starting in the fourth quarter of this year.

Tesla’s robotaxi map is growing; its trip count isn’t

On July 21, Tesla added Orlando and Tampa to its Robotaxi map, following earlier launches in Austin, Dallas, Houston, and Miami (Insurance Journal). Coming right before Tesla’s Q2 earnings call, the timing looked designed to project momentum. The numbers tell a flatter story.

Quarter	Paid Robotaxi miles added	Notes
Q1 2026	~900,000	Early expansion beyond Austin
Q2 2026	~900,000	Two new cities added, mileage unchanged

Roughly 500,000 of Q2’s miles came in April alone, with May and June combined adding only about 200,000 miles a month — a run-rate closer to 600,000 miles per quarter by the end of the period (Electrek). The unsupervised fleet sits at roughly 21 vehicles across four cities, down from a late-April peak near 25 in Austin alone. On the earnings call, Musk acknowledged the company had missed its own earlier expansion targets, attributing the pace to “rigorous safety testing” rather than a change in ambition (Insurance Journal). Every California ride, notably, still carries a Tesla employee as a safety monitor — which Electrek characterizes as “a supervised pilot with a paywall” rather than driverless service in the way Waymo or Baidu’s Apollo Go, covered in Dispatch #9, already operate.

Also this fortnight: the SoftBank–ABB Robotics deal flagged in Dispatch #9 as expected to close “mid-to-late 2026” took a concrete step forward — a four-bank syndicate (BNP Paribas, Goldman Sachs, HSBC, and Mizuho) finalized a $1.75 billion loan backing the $5.4 billion acquisition on July 22 (Briefs). Debt financing locking into place is usually one of the last steps before a deal closes.

Connecting the dots

Two of the ideas from the last three tutorials show up directly in these stories. ROS 2 services and actions are the request/response and long-running-task patterns you used to ask a node to do something and track its progress. Every “pick up this part” or “sequence these components” instruction Atlas will receive at Hyundai’s Metaplant, or that an HMND 01 will receive on a Schaeffler line, is exactly that pattern at industrial scale: a goal is sent, feedback streams back while the arm moves, and a result confirms success or failure — the same shape as the action you built by hand, just with a manipulator instead of a print statement.

ROS 2 lifecycle nodes and composition matter even more once a robot moves from a lab demo to a factory floor with real safety requirements. A deployed humanoid can’t just run — it needs clean, managed states: unconfigured while it boots, inactive while a human clears the area, active only once checks pass, and a controlled shutdown path if something goes wrong. That is precisely the lifecycle state machine from two weeks ago, and it is also, less visibly, what “rigorous safety testing” means for something like Tesla’s Robotaxi: not just better perception, but a rigorous state machine for when the vehicle is and is not allowed to act.

And the discipline behind recording and replaying data with rosbag2 is the quiet infrastructure underneath every fleet number in this Dispatch. Tesla’s 900,000-mile quarters, Boston Dynamics’ Atlas pilots, and Humanoid’s Schaeffler rollout all depend on capturing every sensor stream from every run so it can be replayed for debugging and fed back into training — the same rosbag2 record-and-play workflow you ran on a single robot, just multiplied across thousands of vehicles and machines.

Next week we start a new block of the tutorial track: Working with Point Clouds (Week 31), the first of four posts on perception. It’s a fitting next step — every robot in this Dispatch, wheeled or bipedal, depends on exactly that kind of 3D sensing to know what’s in front of it.

The takeaway

Two of this fortnight’s three stories are ownership changes, not new capability: Hyundai already controlled Boston Dynamics, and SoftBank was always going to close the ABB deal it announced last October. What’s genuinely new is Humanoid’s unicorn round and its emphasis on a wheeled platform, which is a tacit admission that legs are still the hardest, slowest part of humanoid robotics to get right — a deployable robot today looks more like a torso on wheels than a walking machine. Our read: the more interesting number in this Dispatch isn’t a valuation, it’s Tesla’s flat 900,000 miles quarter-over-quarter. Announcing new cities is cheap; growing utilization in the ones you already have is not, and that gap between geographic footprint and actual usage is likely to be the real story of the back half of 2026 for every company in this piece, not just Tesla.