Robotics Weekly Review 2026-05-31

Week In Review

The week the 2026 Robotics Summit & Expo gathered more than 5,000 developers in Boston turned out to be the week the humanoid sector publicly stopped pretending it was still a research program. Figure AI signed a commercial logistics deal with Catalyst Brands covering JCPenney’s distribution operations and, days later, told the trade press it is now building one robot per hour at its California plant. At the Summit itself, Amazon Vulcan was named RBR50 Robot of the Year — not for being a humanoid, but for being the first warehouse robot with a calibrated sense of touch good enough to handle three-quarters of Amazon’s stockkeeping units. The dominant question of the Summit, captured in the opening day’s framing, was no longer whether humanoids can work, but who will own the software stack that runs them.

Tactile sensing was the throughline. The Vulcan award acknowledged what manipulation researchers have argued for years — that vision alone is not enough to grasp arbitrary objects — and the same point was made on the show floor by XELA Robotics, which demonstrated a robotic fingertip with thirty three-axis force sensors packed into a single pad. QNX, the operating-system vendor that runs much of modern automotive software, used the Summit to release its Inside the Robot Architecture Benchmark Report, a thousand-developer survey trying to capture how robotics teams actually build production-grade systems today. Together these are the kinds of unglamorous infrastructure stories that signal a field is maturing.

The week’s most arresting image came at the very end, when Noland Arbaugh — the first human Neuralink implant recipient — closed the Summit with a live brain-computer-interface demonstration, controlling a cursor by thought in front of thousands of robotics engineers. The choice of speaker mattered: the people building the next decade of physical autonomy seem increasingly willing to entertain the idea that human nervous systems will be part of the control stack. The drone industry took a similar growing-up step the same day, when Matternet raised $33 million and went public through a reverse merger, becoming a listed company on the strength of more than 60,000 commercial flights and the only FAA type-certified delivery drone in U.S. service.

Beyond the show floor, NPR’s visit to Ginkgo Bioworks and the MassRobotics Physical AI Fellowship’s nine-startup showcase both pointed at a quieter shift: the robot is no longer just an actuator, it is becoming the scientist’s hands and, in some cases, the experimenter’s eyes as well. None of these stories require believing the hype curve. They require believing that the boring work of contracts, certifications, sensors, and software stacks has begun in earnest.

Items

Figure Signs Commercial Agreement to Put Humanoids in JCPenney’s Distribution Network

Figure AI and Catalyst Brands — the apparel holding company whose portfolio includes JCPenney, Aéropostale, Brooks Brothers, Lucky Brand, and Nautica — announced a commercial agreement to deploy Figure’s next-generation humanoids inside a Catalyst distribution facility in Reno, Nevada. The robots will work alongside human associates on the Joey Pouch sorting system, an automated induction-and-packing line where individual garments are routed in fabric pouches before they are bagged for shipment.

This is Figure’s first publicly announced contract with a portfolio company of Brookfield, the asset manager that holds equity in both Figure and Catalyst. The framing in the announcement is deliberate: Figure positions the humanoid as relief for “repetitive, physically demanding sorting and packing tasks” so that associates can move to higher-skill functions, rather than as a headcount-reduction story. The Reno site will be the first phase, with the agreement structured to scale across additional Catalyst facilities if the pilot meets its performance targets.

What makes the deal notable is that it pushes humanoids out of the automotive plant — where Figure’s BMW Spartanburg pilot lived for most of last year — and into general-merchandise retail logistics, where SKUs change weekly and packaging is far less uniform than a sheet-metal subassembly. If the Reno deployment proves out, it positions humanoids as a candidate for the much larger and messier world of e-commerce fulfillment.

Source: Figure

Figure Ramps Humanoid Manufacturing to One Robot Per Hour

Days after the Catalyst announcement, Figure detailed its production cadence: the company is now manufacturing humanoid robots at roughly one unit per hour at its California facility, a step change from the prior pace of one per day. The number is significant less as a raw output figure than as a signal about which constraints the company believes it has solved.

Producing humanoids at hourly cadence implies stable component sourcing for actuators, batteries, and dexterous hands, plus an assembly process that has been decomposed into repeatable workstations rather than bespoke build cells. Both have been chronic bottlenecks across the sector. Industry watchers have pointed out that Unitree shipped over 5,500 units across 2025 with a much simpler product, while most U.S. humanoid makers were still producing in the tens or low hundreds. An hourly cadence puts Figure on a trajectory to ship many thousands of units within a calendar year if it can sustain the rate.

The production push is happening in parallel with Figure’s commercial deal book, which now includes BMW and Catalyst Brands. The strategic logic — build enough robots to meet contractual deployment milestones rather than to seed labs and demos — mirrors how Tesla scaled the Model 3 once it had reservation backlog. Whether Figure can hold the rate is the question its competitors and customers will be watching for the rest of 2026.

Source: Robotics & Automation News

Amazon Vulcan Named RBR50 Robot of the Year

The Robot Report’s 2026 RBR50 Robotics Innovation Awards, presented during the Robotics Summit in Boston, named Amazon’s Vulcan robot as Robot of the Year. Vulcan is the first Amazon warehouse robot fitted with a genuine sense of touch: AI-driven force and torque sensing that allows the system to determine how hard to grip an item before it picks it up.

The technical claim around Vulcan is that it can handle roughly seventy-five percent of the unique SKUs in an Amazon fulfillment center — a category that includes books, soft goods, irregular packaging, and small electronics that have historically defeated grippers built for boxes. The breakthrough is less about a single sensor than about closing the loop between contact estimation and grasp policy in real time, so the robot adjusts grip force the way a human would when picking up a paper cup versus a hardcover book.

The award is also a statement of editorial priorities. The RBR50 jury chose Vulcan over a field that included multiple high-profile humanoids, signaling that the most important commercial robotics story of the year is not bipedal hardware but the addition of tactile perception to manipulation systems that already exist at scale. Amazon Robotics engineers presented the Vulcan development story at the RBR50 dinner during the summit.

Source: The Robot Report

Noland Arbaugh Closes Robotics Summit With Live Neuralink Demonstration

Noland Arbaugh, the first human recipient of a Neuralink brain-computer-interface implant, closed the 2026 Robotics Summit on Wednesday afternoon with a live demonstration of neural cursor control. Arbaugh, who is quadriplegic, moved a cursor on a large screen using only neural signals decoded from the implant in his motor cortex, drawing a standing ovation from an audience of robotics engineers and developers.

The programming choice is significant. The Robotics Summit is a developer conference, not a medical-device venue, and the organizers chose to give the closing slot to a brain-computer-interface user rather than to another humanoid CEO. The implicit argument is that the next generation of physical autonomy will route through human intent in ways that pure perception-action systems cannot replicate — particularly for assistive applications, remote teleoperation, and shared autonomy in safety-critical settings.

The demo also drew an explicit line between two communities that have historically circled each other from a distance: neurotechnology and robotics. Several Summit speakers throughout the day had touched on teleoperation as a way to bootstrap humanoid skill data, and Arbaugh’s appearance closed the loop on what high-bandwidth human-to-machine intent transfer could look like once BCIs leave the laboratory.

Source: TechTimes

Matternet Raises $33 Million and Goes Public Through Reverse Merger

Drone-delivery operator Matternet announced a $33 million capital raise and a planned public listing via reverse merger, becoming one of the first dedicated cargo-drone operators to reach U.S. public markets. Matternet has logged more than 60,000 commercial flights across the U.S. and Europe since 2014 and remains the only operator to hold FAA Type Certification for a drone-delivery aircraft platform — a regulatory status that took years to secure and serves as a substantial moat.

The Matternet M2 aircraft is operated under FAA Part 135 authority by partners including UPS and Ameriflight, primarily on hospital-to-hospital medical logistics routes where rapid point-to-point delivery has demonstrable clinical value. Public-market access changes the company’s growth options: the proceeds and listing currency can fund vehicle production, route expansion, and the engineering needed to deepen autonomy and reduce per-flight operating cost.

The timing matches a broader shift in U.S. drone-delivery regulation, with the proposed FAA Part 108 framework for beyond-visual-line-of-sight operations moving through rulemaking and competitors like Zipline pursuing aggressive city-by-city expansion. The pattern is familiar from earlier waves of autonomy: once one operator clears certification, others can use the path it cut, and the public-market valuations begin to reflect a real industry rather than a speculative one.

Source: DroneLife

XELA Robotics Shows a Six-Axis Robotic Fingertip With Thirty Force-Sensing Points

XELA Robotics used the Robotics Summit & Expo to debut a new tactile-sensing fingertip that packs thirty tri-axial force-sensing points into the pulp of a single robotic finger, plus a six-axis, force-sensitive nail. The result is a fingertip that not only senses normal pressure but also resolves shear and torsional contact forces across a curved surface, the way a human fingerprint ridge does.

For dexterous manipulation, the limiting factor has long been the bandwidth and resolution of contact sensing, not the actuators. Most robotic hands can already exceed human strength and speed on simple grasps, but cannot tell when an object is about to slip, when contact has been made unevenly across a curved surface, or when a fragile object is being deformed. A pad with thirty distributed tri-axial taxels gives a control policy enough raw signal to estimate slip, surface friction, and contact patch geometry on the fly.

XELA also showed expanded integrations with leading robot arm and hand platforms — meaning the sensors can drop into existing research and industrial hardware rather than requiring a custom robot to use them. That kind of supplier-style availability is what allowed cameras and LiDAR to become commodity inputs to robotics over the past decade; tactile sensing has long lacked the same off-the-shelf ecosystem.

Source: The Robot Report

QNX Releases “Inside the Robot” Architecture Benchmark Report

QNX, the real-time operating-system vendor whose software runs in a large share of vehicle infotainment and safety systems, used its Robotics Summit appearance to publish the first edition of a new global research report, Inside the Robot: Architecture Benchmark Report. The study is based on a survey of one thousand robotics developers worldwide and aims to map how teams actually architect production-grade robotic systems today — including operating-system choices, software-stack composition, AI integration, and certification posture.

The report’s interest is less in any single statistic and more in the fact that it exists at all. Robotics has historically lacked the kind of horizontal developer survey that automotive, mobile, and web ecosystems have used to benchmark themselves. As humanoids and mobile robots begin to ship in commercial quantities, the questions developers face shift from “can we make it move” to “can we ship it safely on a schedule, with a real OS, on real silicon, with software updates over the air, while passing certification.”

QNX’s commercial interest is obvious — its safety-certified real-time operating system is a candidate to underlie those production stacks — but the dataset itself addresses a real gap. The report joins a slow accumulation of industry plumbing that suggests robotics is starting to look more like a manufacturing-grade software domain and less like a collection of academic demos.

Source: The Robot Report

NPR Inside Ginkgo Bioworks’ AI-Driven Robot Lab

NPR’s Katia Riddle filed a feature on the emerging category of AI-powered autonomous laboratories, anchored on a visit to Ginkgo Bioworks’ Boston facility. The reporting describes a working biology lab in which robots encased in glass execute experiments on a color-coded schedule, materials are routed between stations on circulating circuits, and the scientific direction is increasingly handed off to AI systems rather than human researchers at the bench.

Ginkgo’s lab handles work on new drug candidates, microbes engineered for soil fertilization, and proteins that catalyze ice and snow formation — a range that would have required separate research groups in the conventional model. The new approach centers on using AI to translate high-level scientific goals into specific protocols, and increasingly to choose which experiments to run next based on prior results. The shift is from “AI helps scientists do experiments” to “AI is the scientist deciding which experiments matter.”

The report does not soft-pedal the concerns. Domain experts NPR interviewed warned that delegating scientific judgment to AI systems risks losing tacit understanding of how good science is done, and that autonomous labs available to people with little training raise biosecurity questions. Both concerns will follow this category as it expands. But the trajectory is clear enough that mainstream science journalism is now writing about it as a category, not a curiosity — a signal that the robotic-laboratory model has moved past pilot phase.

Source: NPR

Nine MassRobotics Physical AI Fellowship Startups Debut at the Summit

The second cohort of the MassRobotics Physical AI Fellowship — Burro, Config, Deltia.ai, Haply Robotics, Luminous Robotics, Roboto AI, Telexistence, Terra Robotics, and WIRobotics — debuted publicly at the Robotics Summit & Expo this week. The fellowship is one of the more focused early-stage programs in U.S. robotics, oriented around what the field has started calling “physical AI” — the combination of foundation-model perception and reasoning with embodied actuation.

The cohort spans a wider range of form factors than the humanoid headlines suggest. Burro builds autonomous agricultural follow-robots already used in vineyards; Telexistence operates retail-stocking robots already deployed in Japanese convenience stores; Haply Robotics develops haptic devices that bring force feedback to remote operation. The fellowship is structured so that each startup gets workspace, mentor access, and a structured path to first commercial customer in a way that the broader VC-funded environment has been less willing to fund directly.

The pattern of public debuts at a single industry event matters as a market-formation signal. Three or four years ago, a U.S. robotics startup cohort would have been hard pressed to find a developer-conference stage that mattered. The fact that nine companies chose this venue as the place to launch reflects the consolidation of attention around the Boston event and around the physical-AI thesis more broadly.

Source: The Robot Report

“State of Humanoids” Opens on the ROS vs. Proprietary Physical-AI Question

The Robotics Summit’s opening framing — captured in coverage going into the event — made the central industry question explicit: now that humanoid hardware is starting to ship in production volumes, who will own the software foundation on which it runs? The opening keynote panel “Building the Next Era of Robot Autonomy” included Aaron Parness of Amazon Robotics, Anders Beck of Universal Robots, Hamid Montazeri of Locus Robotics, and John Wall of QNX. A second panel on “The State of Humanoids” included Alberto Rodriguez of Boston Dynamics and Pras Velagapudi of Agility Robotics, among others.

The strategic question they were arguing about is the open-versus-proprietary one. Robot Operating System (ROS), the open middleware that has powered the previous generation of mobile robots and research platforms, is one option. Closed, vertically integrated stacks — Figure’s Helix, Boston Dynamics’ Atlas software, Tesla’s Optimus — are another. NVIDIA’s Isaac platform, Hugging Face’s growing library of open foundation models, and several startup-driven middleware efforts are still other candidates. Whoever wins this layer captures a great deal of leverage over the humanoid economy.

Coverage of the Summit’s opening day emphasized concrete production-scale numbers to anchor the abstraction: Figure 02 contributed to more than 30,000 BMW X3 vehicles at Spartanburg and handled more than 90,000 sheet-metal components in 1,250 operating hours; Agility’s Digit moved more than 100,000 totes at GXO’s Flowery Branch warehouse. With those volumes on the table, the question of which software platform owns the next 100,000 robot units has stopped being theoretical.

Source: TechTimes