Robotics Weekly Review 2026-07-26

Week In Review

This was a rare week in which every layer of the robotics stack — silicon, models, bodies, capital, and clinical evidence — moved at once. AMD’s launch of the Ryzen AI Embedded X100 and Kria AI SoM put a 128 GB unified-memory Strix Halo class SoC into a package rated for -40 to 105 °C and ten years of continuous operation, giving robot builders a credible x86 alternative to NVIDIA’s Jetson line for on-board physical-AI compute. That silicon lands into a market where models trained to run on it are, for the first time, generalizing across hardware: Generalist’s GEN-1 foundation model now transfers a single set of learned sensorimotor policies to five-fingered hands, parallel grippers, and specialized tools trained on more than half a million hours of real-world interaction data.

The bodies to run those models on are proliferating too. UK-based Humanoid became Europe’s first pure-play humanoid unicorn with a $152 million Series A, Chinese leader Unitree began a coordinated three-continent commercial launch of the H1 Pro, and NEURA Robotics opened its Aachen “gym” for physical-AI training — the anchor of a ten-facility global network backed by its $1.4 billion Series C. The connective tissue of the week is data: everyone now believes that whoever accumulates the most hours of purposeful, embodied interaction wins the model race, and each of these announcements is really a bet on how to acquire those hours.

Medicine drew the week’s most consequential regulatory decision. Johnson & Johnson’s OTTAVA soft-tissue surgical robot cleared the FDA for ten general-surgery procedures, ending two decades in which Intuitive Surgical’s da Vinci system faced no direct US competitor in the category. EndoQuest, working from a very different angle, closed enrollment in the PARADIGM trial for its flexible endoluminal system, reporting a 100 percent en-bloc resection rate across 50 robotic endoscopic submucosal dissections. Between them these two milestones bracket a rebooted surgical robotics category: rigid table-integrated arms on one end, snake-like tools going through natural orifices on the other.

Defense and heavy industry supplied the reminder that autonomy is a general-purpose technology. At Farnborough, Anduril unveiled Thunder, an autonomous tilt-rotor “loyal wingman” for crewed helicopters, while Kraken Robotics booked $35 million in maritime-defense and offshore-energy orders driven by port-security demand from the Middle East and Europe. And in Xinjiang, Chinese OEM Tonly signed a deal for 800 autonomous electric mining trucks — a scale of driverless heavy-vehicle fleet operations that only a few years ago existed only in pilot projects.

Items

Johnson & Johnson Wins FDA De Novo Authorization for OTTAVA Soft-Tissue Surgical Robot

For the first time in more than two decades, the US soft-tissue robotic-surgery market has a fully credentialed challenger to Intuitive Surgical’s da Vinci. On July 22, the FDA granted De Novo authorization to Johnson & Johnson’s OTTAVA system for ten common general-surgery procedures including Roux-en-Y gastric bypass, gastrectomy, cholecystectomy, gastric sleeve, small-bowel resection, appendectomy, and hiatal-hernia repair.

OTTAVA’s central architectural bet is that the surgical robot should live inside the operating-room table rather than around it. Four robotic arms fold out of the table itself, an arrangement J&J says takes 30 to 50 percent less floor space than the boom-and-cart systems that have dominated the category. That matters operationally because operating-room square footage is one of the hardest constraints in modern hospital design, and it matters strategically because it defines a new device class rather than competing feature-for-feature with da Vinci.

The authorization is a De Novo classification, meaning the FDA has created a new regulatory category for table-integrated soft-tissue robots rather than clearing OTTAVA as substantially equivalent to an existing device. That gives J&J a defensible perimeter even as Medtronic’s Hugo and CMR Surgical’s Versius work through their own approval processes. Intuitive still holds roughly 70 percent of the market by revenue, but for the first time in the modern history of the field, hospitals planning capital-equipment purchases have a real US-approved second option in the highest-volume surgical categories.

Source: Johnson & Johnson

AMD Launches Ryzen AI Embedded X100 and Kria AI SoM for Physical AI

AMD’s Advancing AI 2026 event this week produced the most credible x86 challenge yet to NVIDIA’s Jetson platform for on-board robot compute. The new Ryzen AI Embedded X100 series pairs up to 16 Zen 5 CPU cores, an RDNA 3.5 integrated GPU with up to 40 compute units, and an XDNA 2 neural processing unit rated at up to 50 TOPS — a Strix Halo-class configuration in a package rated for continuous 24/7 operation across a -40 to 105 °C temperature range for up to ten years.

The choice AMD is putting to robot developers is architectural. NVIDIA’s edge robotics story revolves around discrete Jetson modules pushing GPU-accelerated inference; AMD is offering a single heterogeneous chip in which x86 cores handle deterministic real-time control, the GPU runs perception, and the NPU handles the learned-policy inference — all sharing up to 128 GB of unified memory. That matters for the current generation of vision-language-action models, whose parameter counts and multimodal context windows make memory bandwidth and capacity as important as raw TOPS.

Companion to the X100 is a refreshed Kria AI system-on-module built around AMD’s Versal adaptive SoC lineage, plus an open Physical AI Developer Platform and a Robotics Partner Network. Customer sampling began in June with production planned for the fourth quarter of 2026. The industrial-embedded market has historically been dominated by Intel and NVIDIA at the two ends, and AMD is now credibly in the middle with a package that reads specifically as designed for humanoids, cobots, and surgical robots rather than repurposed from datacenter or consumer parts.

Source: AMD Newsroom

Generalist’s GEN-1 Foundation Model Extends to a Range of Robot End Effectors

Generalist AI announced this week that GEN-1, its embodied foundation model, now trains and deploys across a wide variety of robot end effectors — from parallel grippers to five-fingered dexterous hands to specialized tools with novel actuation. The claim is that a single base model can learn sensorimotor policies which transfer across radically different ways of interacting with the physical world, without hardware-specific retraining for each configuration.

GEN-1 is pretrained on Generalist’s in-house dataset, which the company reports now spans more than half a million hours of real robot-interaction data across those diverse end effectors. On the tasks Generalist has benchmarked, the current model achieves roughly 99 percent success rates against a previous generation’s 64 percent while completing work about three times faster and requiring only about an hour of new robot data to adapt to a novel platform.

The significance is less about the individual metrics than about the change in what a foundation model for robotics is expected to abstract over. Language models generalize over vocabularies; image models over resolutions and styles. GEN-1 is arguing that the analogous invariant for embodied AI is the end effector: a good enough base model shouldn’t care whether it is closing a two-finger gripper or coordinating twenty-plus tendon-driven joints. If that argument holds up as third parties integrate the model, it collapses a large amount of per-morphology engineering work and pushes robotics further toward the software-driven scaling curves that have defined the rest of AI.

Source: The Robot Report

NEURA Robotics and RWTH Aachen Open a 3,000 m² Gym for Physical AI Training

On July 22, NEURA Robotics announced a partnership with Germany’s RWTH Aachen University to establish NEURA Gym RWTH Aachen, a roughly 3,000-square-meter facility on the Hightech Campus Melaten dedicated to training and validating physical-AI models in intelligent and humanoid robotics. Approximately 20 RWTH institutes will be integrated into the gym, and the facility is being run as industry-open — large enterprises and mid-sized firms can bring production workflows through for high-end robotics validation, particularly in manufacturing, automotive, circular-economy applications, and medical technology.

The Aachen site is one piece of a much larger buildout. NEURA has ten gyms in development, with five operational across Europe, the United States, and China by the end of 2026, and a TUM RoboGym opening at Munich Airport in the fourth quarter in partnership with the Technical University of Munich and its Munich Institute of Robotics and Machine Intelligence. The global rollout is underwritten by NEURA’s Series C financing of up to $1.4 billion, which the company describes as the largest ever raised by a full-stack robotics company.

The strategic bet embedded in a “gym” network is that the bottleneck for capable general-purpose robots is not model architecture or compute but access to high-quality, task-diverse, real-world interaction data captured in environments that resemble the ones the robots will eventually deploy into. By coupling those environments to a top-tier university’s research faculty and to industrial partners’ actual workflows, NEURA is trying to build a data pipeline that neither pure research labs nor pure product companies can easily replicate.

Source: NEURA Robotics

Humanoid Raises $152 Million Series A at $1.35 Billion Valuation, Becomes Europe’s First Pure-Play Humanoid Unicorn

UK-based Humanoid announced on July 21 a $152 million Series A financing at a $1.35 billion post-money valuation, bringing the two-year-old company’s total funding to $270 million and making it the first pure-play humanoid robotics unicorn to emerge in Europe. Prime Movers Lab led the round, with participation from Schaeffler, Bosch, Fubon Financial Holding Venture Capital, and Aglaé Ventures — a syndicate weighted toward strategic industrial partners rather than pure venture capital.

Humanoid’s design choice differentiates it from most well-funded competitors: rather than a fully bipedal humanoid, its first commercial robot uses a wheel-based mobile base topped with a bimanual upper body. The bet is that a wheeled platform is dramatically simpler to operate reliably in structured industrial environments than a walking one, while sacrificing very little of what matters for logistics, assembly, and materials handling — the near-term commercial applications that dominate customer demand.

The Series A proceeds will fund development of Humanoid’s next-generation robotics platform, expansion of its AI software stack, scaling of commercial deployments, and preparation for mass production, with beta systems targeted for customers in the fourth quarter of 2026. The scale of the round matters as a signal about European industrial policy: several of the strategic investors are core suppliers to European automotive and industrial manufacturing, and their participation reads as an explicit hedge against dependence on US and Chinese humanoid platforms for the coming wave of factory-floor automation.

Source: Humanoid

Unitree Launches H1 Pro Commercially in Europe

On July 22, Unitree Robotics began commercial European sales of its H1 Pro humanoid — the opening move in a coordinated three-continent rollout with Asian availability following on August 5 and North American on August 12. It is the first Chinese-made humanoid to enter Western commercial markets in a coordinated launch of this scale, and it comes just weeks after Unitree secured China Securities Regulatory Commission approval for a STAR Market IPO with a stated 2026 shipment target of 20,000 units.

Unitree’s competitive proposition is straightforward: humanoids at roughly one-tenth the price of Western competitors’ comparable platforms, with a track record of shipping more units than any Western pure-play. The H1 Pro extends that with enhanced dexterity, integrated on-board AI, and regional European distribution partners. Only about 9 percent of Unitree’s current shipments go to industrial customers, with the majority split between research, education, and commercial venues such as tourist attractions, exhibitions, and retail installations.

The launch is proceeding against a complicated policy backdrop. Every H1 Pro shipped carries an implicit obligation under Article 7 of China’s National Intelligence Law, which requires Chinese firms to cooperate with state intelligence services on demand irrespective of where their products operate. The Pentagon flagged Unitree as Chinese military technology days before the European launch. How Western enterprise customers and regulators reconcile Unitree’s cost and capability lead with those obligations is likely to be one of the defining questions of the next humanoid-adoption cycle.

Source: TechTimes

EndoQuest Completes Enrollment in the PARADIGM Trial for Robotic Endoluminal Surgery

EndoQuest Robotics announced this week that enrollment has concluded in the PARADIGM trial (Prospective Assessment of a Robotic Assisted Device in Gastrointestinal Medicine) after completion of 50 robotic endoscopic submucosal dissection procedures across 49 subjects. The pivotal multicenter investigational device exemption study was run across five institutions by six investigators — three colorectal surgeons and three gastroenterologists — evaluating the safety and effectiveness of EndoQuest’s Endoluminal Surgical (ELS) System for removing lesions in the rectum and sigmoid colon.

The preliminary numbers presented at the 2026 Society of Robotic Surgery Annual Meeting are unusually clean for a pivotal robotic-surgery trial: 100 percent en-bloc resection, 100 percent conversion-free, 95.8 percent R0 resection, and no reported device-related bleeding or perforations as of the July 18 cutoff. Those endpoints are what regulators and surgical societies look for when deciding whether a new procedural approach clears the bar to become standard of care rather than a specialist technique.

Where OTTAVA and da Vinci deploy rigid arms through small incisions, EndoQuest works through the body’s existing openings — the ELS system is a flexible robotic platform that navigates the colon like an endoscope while giving the surgeon two independently controllable robotic arms at the tip. If the final PARADIGM data holds, robotic ESD becomes a scalable outpatient alternative to open or laparoscopic colorectal surgery for early-stage lesions, with a materially shorter recovery and a far lower complication profile.

Source: Business Wire

Anduril Unveils Thunder Autonomous Attack Rotorcraft at Farnborough

At the 2026 Farnborough Airshow, Anduril unveiled Thunder — a Group 5 unmanned tiltrotor aircraft designed to fly as an autonomous “loyal wingman” alongside crewed attack helicopters. Developed in partnership with Archer Aviation, Thunder uses a series hybrid-electric powertrain and tilting rotors that let it take off vertically and cruise at speeds comparable to the Apache and next-generation Cheyenne II. First flight is targeted for sometime in 2027.

The design point is the aviation analogue to what has already happened in fixed-wing autonomy with the Collaborative Combat Aircraft program: pair a small number of crewed platforms with a larger number of autonomous ones that carry sensors, weapons, and expendable risk. Anduril’s own CCA prototype fired an AIM-120 at a digital target over the Mojave Desert on July 15, and in June the Air Force awarded Anduril and General Atomics engineering, manufacturing-development, and production contracts for CCA Increment 1 four months ahead of schedule.

Thunder extends that architecture into rotary-wing missions where crewed attack helicopters have historically operated alone in exposed low-altitude environments. From a technology-progression standpoint, the interesting element is the propulsion integration: hybrid-electric tiltrotors are a genuinely novel military airframe class, and building them for autonomous rather than crewed operation removes many of the constraints — pilot g-limits, life-support systems, escape systems — that have historically shaped rotorcraft design.

Source: DefenseScoop

Kraken Robotics Books $35 Million in Underwater and Maritime-Defense Orders

Canadian marine-robotics firm Kraken Robotics announced on July 20 that it had booked $35 million in new orders across maritime defense, offshore energy, and ocean-science customers. The company cited increased demand for port and harbor security in the Middle East and Europe as the primary driver, with orders spanning synthetic-aperture sonar systems, subsea navigation and positioning suites, and Covelya-branded monitoring hardware.

Marine robotics has historically been the quietest corner of the autonomy industry — underwater vehicles work in environments where GPS doesn’t reach, communications bandwidth is measured in kilobits, and the failure modes are unforgiving. Synthetic-aperture sonar in particular is a technology that has only recently become affordable and reliable enough to move from bespoke naval-research platforms into commercial and dual-use fleets, and Kraken is one of a small number of firms shipping it at scale.

The order book reflects a broader shift in how governments are thinking about maritime security. The combination of contested undersea infrastructure — pipelines, communications cables, offshore energy — with a persistent shortfall of crewed patrol capacity has made autonomous or remotely operated underwater and surface vehicles a priority procurement category on both sides of the Atlantic. For a mid-cap specialist like Kraken to book this scale of orders in a single quarter announcement is a signal that the category is transitioning from prototype to program of record.

Source: Kraken Robotics

Tonly Signs Deal for 800 Autonomous Electric Mining Trucks in Xinjiang

Chinese heavy-equipment maker Tonly announced this week a deal to deliver 800 autonomous “new energy” mining trucks to Xinjiang Hanxiang, announced at Tonly Heavy Industries’ Xinjiang Regional Green Intelligence Mine Innovation and Development Seminar in Ürümqi. The order is one of the largest single autonomous-vehicle purchases on record and applies to an environment where full driverless operation is already technically and economically defensible: closed private-road mines with well-defined haul routes.

Autonomous mining haul trucks are, in some ways, the most mature category of full-scale driverless operation on the planet. Caterpillar and Komatsu have been running driverless fleets in Australian iron-ore mines for more than a decade, and the operational data — around-the-clock utilization, elimination of shift-change downtime, reduction of tire wear and fuel consumption from smoother driving — has been consistent and favorable. What is new here is the scale of a single order, the fact that it is going to a Chinese OEM rather than the traditional Western duopoly, and that the trucks are battery-electric rather than diesel.

The combination matters because mining has historically been one of the hardest sectors to decarbonize: haul trucks burn enormous quantities of diesel, and the megawatt-hour class batteries required to electrify them have only recently become practical. Coupling electrification with autonomy in a single procurement means the deployment can be optimized end-to-end — the trucks can be scheduled around charging windows and dispatched with a level of coordination that human drivers cannot match, which is exactly the operational envelope where electric heavy-vehicle economics turn positive.

Source: Electrek