Robotics Weekly Review 2026-05-16

Week In Review

This week marked an inflection point in how robotics is talked about: less “can it work” and more “for how long, at what cost, and at what scale”. The week’s most-discussed clip was a Figure AI humanoid running for 50 hours without intervention on a package-sorting line (Figure AI Humanoid Robots Sorted Packages for 50 Hours Nonstop), an endurance demonstration that matters more than any single dexterity benchmark because it speaks directly to whether these machines can earn their per-hour service fees. Adjacent to that, Symbotic disclosed it has now processed more than two billion cases for retailers using its non-humanoid case-handling robots (Symbotic Surpasses 2 Billion Cases Processed) — a reminder that the warehouse robot economy is already large, and that humanoids are entering a market with strong incumbents.

The surrounding warehouse and factory news this week underlined the same theme. Locus Robotics introduced Locus Array, a system that pairs its mobile robots with picking arms to push end-to-end automation past 90% of fulfillment work (Locus Robotics launches Locus Array), and SAP and Cyberwave moved AI-trained robots from pilot into live operation at SAP’s own logistics warehouse in Germany (SAP and Cyberwave deploy fully autonomous AI-powered robots). On the heavy-industry side, Comau and Aptiv announced a joint development framework for AI-enabled industrial robotics (Comau and Aptiv partner on AI-powered robotics), and Agility Robotics’ Digit continued its rollout at Toyota Motor Manufacturing Canada’s RAV4 plant under a Robots-as-a-Service contract (Toyota Motor Manufacturing Canada to deploy Agility Robotics’ Digit humanoids).

Hardware progress kept pace. Boston Dynamics released new test footage of its electric Atlas executing a controlled handstand and one-legged balance routines as part of the platform’s final-phase validation (Boston Dynamics’ Atlas humanoid robot amazes with flawless handstand), and Unitree pushed prices further down with the R1-A5 and R1-A7 — dual-arm upper-body humanoids starting around $3,930 with optional dexterous hands (Unitree R1-A5 and R1-A7 humanoid robots). In medicine, Johnson & Johnson reported pivotal-trial safety and performance data for OTTAVA, its soft-tissue surgical robot, in 30 gastric-bypass procedures with no conversions to non-robotic surgery (Johnson & Johnson Announces Pivotal Clinical Study Results for OTTAVA) — a credible challenge to Intuitive Surgical’s da Vinci franchise.

Looking across these items, the underlying current is that the field is converging on shared infrastructure. The week’s IEEE Spectrum-adjacent commentary on networked AI (IEEE explores networked AI systems that enable robots to learn collectively) framed where this is heading: robots that share experience between fleets rather than each one learning from scratch. Whether Figure’s 50-hour run holds up under scrutiny or J&J’s 30-patient cohort scales to thousands, the more durable signal is that the same AI substrate — foundation models, simulation pipelines, shared data — is now showing up across humanoids, surgical arms, warehouse robots, and industrial automation alike.

Items

Figure AI’s Humanoid Runs 50 Hours Nonstop on a Package-Sorting Line

Figure AI chief executive Brett Adcock said this week that the company’s humanoid robots had sorted packages for roughly 50 hours without human intervention, in what he framed as a milestone for unattended operation. Adcock has been publicly promoting the run as evidence that the company’s Helix model can drive sustained, real-world commercial work rather than the staged demonstrations that have characterized much of the humanoid sector.

The notable claim isn’t the dexterity of any single pick — package handling is among the easier humanoid tasks — but the duration without falls, jams, or operator resets. Endurance is the metric warehouse customers care about, because labor models break down quickly when a robot needs a human present every few hours. Figure also reported that the robots maintained throughput close to a human worker’s pace through the run.

The trial follows a series of operational milestones for Figure: Figure 02 logging more than 1,250 hours and 90,000 parts handled on a BMW Spartanburg production line, and the company ramping Figure 03 manufacturing toward roughly one robot per hour out of its BotQ facility. The 50-hour run was deliberately viral content — Adcock pledged “no intervention” publicly before it began — and is likely to set a new informal benchmark that competitors will need to match or exceed.

Source: Bloomberg

Symbotic Crosses Two Billion Cases as Non-Humanoid Warehouse Robots Quietly Scale

While humanoids capture attention, Symbotic disclosed this week that its case-handling robots have now processed more than 2.23 billion retail cases for customers including Walmart and Target, and that its autonomous mobile robots have collectively traveled more than 200 million miles inside warehouses. The numbers, released as part of an industry update, place Symbotic among the largest commercially deployed robot fleets in the world by work volume.

Symbotic’s approach is deliberately not anthropomorphic: dense grids of bots move totes between high-bay storage and palletizing stations, optimized for the geometry of cases rather than the geometry of the human body. The company has argued for years that this is the lower-risk path to immediate warehouse productivity — every part of the system is purpose-built rather than asked to imitate a person.

The two-billion-case figure matters as a baseline for the humanoid pitch. New entrants targeting fulfillment work are not entering an empty market; they are competing against systems that are already moving cases at industrial scale with known unit economics. Symbotic’s continued growth suggests that the eventual humanoid value proposition will need to be in tasks that purpose-built systems cannot easily address — likely the long tail of mixed, irregular, or short-run work that resists fixed automation.

Source: Robotics & Automation News

SAP and Cyberwave Put AI-Trained Robots Into a Live Logistics Warehouse

SAP and robotics platform Cyberwave announced that they have moved beyond pilot status at SAP’s St. Leon-Rot logistics warehouse in Germany, where robots trained on Cyberwave’s platform are now performing box folding, packaging, and in-house shipping fulfillment fully autonomously. The deployment is integrated with SAP’s enterprise resource-planning stack, so the robots’ work flows into the same inventory and shipping records that human workers’ actions do.

The interesting technical claim is on the training side. Cyberwave’s platform centers on “Physical AI” — a combination of reinforcement learning and embodied-AI methods that train policies largely in simulation before transferring them to the physical robots. For a warehouse customer, that means new packaging tasks can be added without weeks of on-floor reprogramming, which has historically been the cost driver that kept smaller-volume tasks unautomated.

The deployment is also a statement of intent from SAP, which is positioning robot orchestration as an extension of its existing supply-chain software rather than as a separate vertical. If that framing takes hold, the integration layer between warehouse management systems and robot fleets becomes another front in the ERP wars — which would significantly accelerate adoption among SAP’s existing enterprise customer base.

Source: Robotics & Automation News

Locus Robotics Launches Locus Array, a Mobile-Robot-Plus-Arm Fulfillment System

Locus Robotics, best known for the autonomous mobile robots that move totes through fulfillment centers behind human pickers, this week introduced Locus Array — a system that adds robotic picking arms to its existing mobile platform to attempt end-to-end automation of warehouse fulfillment. The company describes Array as capable of automating up to 90% of pick-and-replenish work, with humans staying in the loop primarily for exceptions and edge cases.

The product positioning is notable. Locus has spent years arguing that the right warehouse-automation strategy is collaborative — robots and humans on the same floor, each doing what they are better at — and has resisted the humanoid framing. Array is its response to a market that increasingly expects fuller automation: rather than replacing humans with humanoid generalists, replace the constellation of mobile bots and pickers with a tightly coupled, purpose-built system.

Array also illustrates how the orchestration software layer is becoming the central battleground. The arms and AMRs are not novel hardware; the value is in the AI-driven order management that decides which bot picks which item from which shelf at which moment. As more vendors converge on similar physical capabilities, the gap between systems will increasingly be a software gap.

Source: Robotics & Automation News

Boston Dynamics’ Atlas Lands a Handstand in Final-Phase Validation

Boston Dynamics released new test footage of its electric Atlas humanoid robot performing a controlled handstand, transitioning from a one-legged balance pose into a full inverted position with both hands on the floor. The clip is part of what the company describes as the final phase of testing on Atlas’s underlying control architecture, in which the robot is subjected to high-intensity acrobatic trials to validate dynamic locomotion and recovery from instability.

These demonstrations matter less as parlor tricks than as evidence of how much margin the platform has. A robot that can recover from a handstand has substantial reserves of balance authority to handle the smaller, more mundane disturbances of factory work — bumps from co-workers, unstable footing, partial slips on a wet floor. Boston Dynamics has historically used acrobatic content to communicate that its hardware is closer to a generalist than a brittle specialist.

The commercial context is also relevant. Boston Dynamics, now Hyundai-owned, has said its entire 2026 Atlas production is committed, with units going to Hyundai’s Robotics Metaplant and to Google DeepMind. The acrobatic testing is therefore not speculative R&D — it is qualification work for robots that will be shipped to paying internal customers, with Gemini Robotics models from Google DeepMind providing the AI substrate.

Source: Interesting Engineering

Unitree Pushes Humanoid Prices Below $4,000 with R1-A5 and R1-A7

Chinese robotics firm Unitree extended its R1 family this week with the R1-A5 and R1-A7 — dual-arm upper-body humanoid robots with starting prices around 26,900 yuan (about $3,930). The robots support 2-finger grippers or 3- or 5-finger dexterous hands, can mount on either a fixed base or a wheeled mobility platform, and offer between 15 and 31 degrees of freedom depending on configuration.

Specification-wise, the platform is more competent than the price suggests. Repeatability is rated at ±0.1 mm for the gripper, with each arm handling payloads up to roughly 2 kilograms. That is enough to credibly perform light assembly, materials handling, and many laboratory and education tasks. By splitting the robot at the waist — selling the upper body only, fixed or on wheels — Unitree avoids the bipedal locomotion problem that drives most of the cost on full humanoids.

The pricing implication for the rest of the industry is severe. Western humanoid programs typically quote $20,000 to $50,000 or more for full-body systems intended for industrial work. Unitree’s strategy of selling capable subsystems at a fraction of that cost — backed by reported shipments of 5,500+ units in 2025 and a 2026 target of 10,000 to 20,000 — is reshaping expectations about what a robot should cost, particularly for research labs and lower-margin commercial users.

Source: CNX Software

Johnson & Johnson’s OTTAVA Surgical Robot Reports Pivotal Trial Data

Johnson & Johnson disclosed pivotal clinical study results this month for OTTAVA, its investigational soft-tissue surgical robotic system, in a prospective multicenter study of Roux-en-Y gastric bypass procedures. In the 30-patient cohort the trial met its primary safety and performance endpoints at 30 days post-procedure, and investigators completed every procedure robotically without converting to a non-robotic approach. Average weight loss at 30 days was 30 pounds, consistent with expectations for the procedure.

OTTAVA is J&J’s long-awaited entrant in soft-tissue robotic surgery, a market that has been dominated for two decades by Intuitive Surgical’s da Vinci platform. The system uses four arms mounted directly to the operating table rather than to a separate cart, with the stated aim of giving surgeons more flexibility in port placement and reducing the footprint of the system in operating rooms that were not designed for surgical robots.

J&J has also submitted a de novo request to the U.S. Food and Drug Administration for OTTAVA in general surgery, which will be the regulatory pathway for U.S. commercialization. A credible second platform in soft-tissue robotic surgery would expand patient access — particularly outside top academic centers, where da Vinci adoption has been driven as much by training pipelines and capital availability as by clinical evidence.

Source: Johnson & Johnson

Agility Robotics’ Digit Scales Up at Toyota’s RAV4 Plant in Canada

Seven Agility Robotics Digit humanoids are now active at Toyota Motor Manufacturing Canada’s Woodstock West plant, which builds the RAV4 and RAV4 Hybrid. The deployment, structured as a Robots-as-a-Service contract under which Agility charges roughly $30 per hour per robot, is the first commercial humanoid program in the Canadian automotive industry. The robots handle parts unloading and movement, freeing human workers from repetitive lift-and-carry tasks.

The Toyota engagement followed a year-long pilot in which Agility tested Digit in real factory conditions before scaling up to a multi-unit, RaaS-priced contract. That structure matters: it converts a capital purchase into an operating expense, which lowers the bar for additional plants to add Digits without separate budget approvals, and gives Agility recurring revenue with clear utilization metrics.

The deal sits in a broader trend across the auto industry. Atlas units from Boston Dynamics are headed to Hyundai’s robotics center; Figure has been running humanoid trials at BMW Spartanburg; and Mercedes-Benz, Honda, and GM all have their own robotics initiatives at varying scales. Automakers are, for the third time this century, the leading early adopters of a new generation of industrial automation — but the unit economics this time are determined by AI capability rather than by mechanical specialization.

Source: The Robot Report

Comau and Aptiv Form an AI-Robotics Joint-Development Pact

Industrial robotics specialist Comau and automotive electronics supplier Aptiv announced a partnership this week to jointly develop AI-powered robotics, autonomous systems, and warehouse-automation technology. The agreement is a framework for joint evaluation rather than a single product, but it commits the two companies to coordinate on advanced industrial robotics, autonomous systems for factories, and AI-enabled warehouse and logistics platforms.

The pairing is strategically logical. Comau brings decades of installed-base experience in heavy industrial robots — particularly in automotive body shops — while Aptiv brings the high-volume software, electronics, and sensor capabilities that came out of its autonomous-driving work. The combination is positioned to address a gap in the market: existing industrial robots are precise but not adaptive, while newer AI-driven robots are flexible but not yet hardened for factory-floor reliability.

For automotive and adjacent manufacturers, the appeal is having a single point of contact for end-to-end automation, rather than integrating brand-new AI startups with legacy industrial hardware. The deal is a signal that AI-enabled industrial automation is being absorbed into the existing supplier landscape rather than disrupting it from outside — which generally accelerates large-scale adoption even if it dampens some of the upside for pure-play startups.

Source: Robotics & Automation News

IEEE Lays Out a Vision for Robots That Learn Collectively

A widely circulated piece this week summarized IEEE-affiliated work on networked AI: the proposition that future robot and AI systems will increasingly learn collectively rather than individually, sharing data and coordinating decisions across many connected agents instead of training and operating in isolation. The argument is less a single research result than a synthesis of where the field is heading, drawing on work in federated learning, multi-agent reinforcement learning, and shared foundation models.

The practical implication for robotics is significant. Each humanoid or industrial robot generates vastly more data about a task than any single training run could ever produce; if those experiences can be pooled across fleets — with appropriate privacy and IP boundaries — every robot benefits from the cumulative experience of all others. Several of this week’s other stories sit inside that frame: NVIDIA’s GR00T foundation models, Cyberwave’s simulation-to-real training pipeline, and Boston Dynamics’ Gemini Robotics partnership all assume some version of shared learning across fleets.

The harder questions are organizational rather than algorithmic. Whose data is it? Who pays for the compute that aggregates it? What happens when an experience learned at one customer’s facility shows up as improved behavior at a competitor’s? The technical pieces are largely in place; the next year of progress in fielded robotics will be shaped at least as much by how these questions are answered as by any single algorithmic breakthrough.

Source: Robotics & Automation News