Racial Profiling Fears Erupt at Federal Vaccine Panel as Hep B Birth-Dose Guidance Shifts

At a recent meeting of the CDC’s Advisory Committee on Immunization Practices (ACIP), public commenters and nonvoting liaisons said discussions about hepatitis B prevention were tainted by anti-immigrant and racially charged rhetoric, including repeated references to Asian and other immigrant groups as sources of infection. Physician Su Wang, who lives with chronic hepatitis B, said a CDC footnote advising extra caution when infants have frequent contact with people from regions where hepatitis B is common felt like an invitation to racial and ethnic profiling. During the same meeting, ACIP voted to move away from universal hepatitis B vaccination at birth, recommending a birth dose only when mothers test positive or status is unknown, and “shared decision-making” when mothers test negative. Critics warned targeted approaches can miss cases and increase stigma. (cidrap.umn.edu)

House NDAA Tightens China Research Rules While Easing Pressure on U.S. Scientists and Universities

The U.S. House has passed the 2026 National Defense Authorization Act, reshaping federal research security policy with a mix of new restrictions and notable pullbacks. The bill includes a softened version of the Biosecure Act, limiting federal support for companies deemed national security risks, but drops an earlier provision that explicitly named five Chinese biotech firms. Instead, the Defense Department and Office of Management and Budget will designate “companies of concern,” and drugmakers are given five years to adjust supply chains. The NDAA also omits the proposed SAFE Research Act, which would have barred U.S. scientists from collaborating with researchers affiliated with China, Russia, Iran, or North Korea—an idea widely criticized as overly broad. Universities scored another win: the bill blocks the Pentagon from unilaterally cutting indirect cost reimbursements, requiring consultation and transition time before any changes. (Science)

Ghost Robotics Adds a Manipulator Arm to Its Vision 60 Robot Dog

Ghost Robotics is rolling out a major upgrade to its Vision 60 quadruped: a six-degree-of-freedom arm designed to turn the “robot dog” into a mobile manipulator. IEEE Spectrum reports the arm can lift up to 3.75 kilograms (about 8.3 pounds), expanding the platform’s utility beyond patrol and inspection into tasks like opening doors, moving objects, and handling tools. The company is also emphasizing ruggedness: the Vision 60 is rated to be submerged up to 1 meter, underscoring its pitch for harsh industrial and defense environments. The article frames the move within a broader surge of interest in legged robots, including competitive pressure from China and a growing focus on real deployments rather than demos. (IEEE Spectrum)

AI Pilots NASA’s Astrobee on the ISS, Cutting Planning Time Dramatically

Stanford researchers have demonstrated AI-assisted motion planning for Astrobee, NASA’s free-flying cube-shaped robot on the International Space Station, aiming to make autonomous navigation faster without sacrificing safety. The team used a “warm start” approach: train a machine-learning model on thousands of previously computed paths so new routes begin from an informed guess, then still pass strict safety checks. In on-orbit testing, the AI warm start reduced computation time by roughly 50%–60% compared with conventional planning, while maintaining collision-avoidance constraints inside the station’s crowded modules. Space.com notes the in-space trial ran 18 trajectories (each repeated with and without the AI warm start) over about four hours, with mission controllers able to halt runs and add virtual obstacles as safeguards. (Space)

A Grain-of-Salt-Scale Robot Points Toward Future “Inside-the-Body” Machines

Researchers have built an ultra-miniature robot—less than a millimeter across—with an onboard computer, sensors, and a motor, overcoming a long-standing miniaturization barrier that has challenged microrobotics for decades. The Washington Post describes the device as smaller than a grain of salt and powered by tiny solar cells; in lab setups, light provides energy both for computation and for propulsion as it moves through liquid. The work, led by teams at the University of Pennsylvania and the University of Michigan, was published in Science Robotics and is framed as a step toward future medical micromachines that could deliver drugs precisely, interface with nerves, or assess cellular health without surgery. Experts quoted in the piece emphasize how difficult it is to integrate power, sensing, computing, and locomotion at this scale—and why this result matters. (The Washington Post)

Pickle’s Pneumatic-Suction Warehouse Robot Targets Backbreaking Container Unloading

A New Atlas report spotlights Pickle Robot Company’s approach to one of logistics’ most punishing jobs: unloading dense, mixed cargo from shipping containers. The system pairs a one-armed robot with pneumatic suction picking and computer vision to identify and extract boxes and items efficiently, aiming to reduce injury risk and labor strain while improving throughput. The headline claim is eye-catching: the robot can clear up to 75,000 lb of cargo per hour under suitable conditions, positioning it as a serious automation play for high-volume distribution operations. Beyond speed, the promise is practical robustness—working amid clutter, variable packaging, and real-world messiness that often breaks traditional warehouse automation. The piece frames it as part of a broader trend: “AI + specialized hardware” systems designed for narrow, expensive bottlenecks in the supply chain rather than general-purpose humanoids. (New Atlas)

Soft Robots Get a Self-Focusing “Squishy Eye” That Tunes Focus With Light

Georgia Tech researchers have developed a photoresponsive hydrogel lens (PHySL) inspired by how biological eyes adjust focus, and New Atlas argues it could be a meaningful step toward practical vision systems for soft robots and biomedical devices. Instead of rigid optics moved by motors and gears, the lens “squishes and stretches” to change focal length, with control achieved optically—responding to light itself rather than requiring electronic actuation. The researchers describe the work in a Science Robotics paper, and the article emphasizes why that matters: soft robots and medical tools often need low-power, compliant components that won’t harm living tissue or fail when bent and twisted. The team also reports a prototype electronics-free camera concept integrating the lens with a light-activated microfluidic chip, hinting at future untethered imaging devices for constrained environments. (New Atlas)

Humanoid Robots Headline a Silicon Valley Summit—Hype Meets Hard Engineering

At the Humanoids Summit in Mountain View, California, more than 2,000 attendees gathered to see prototypes, debate commercialization, and confront the stubborn reality that humanoids are still expensive, complex machines. AP reports the event pulled in engineers from major organizations (including Disney and Google) and startups eager to show progress in dexterous manipulation and whole-body mobility—momentum supercharged by the broader boom in AI. But the piece is careful about skepticism: robotics remains capital-intensive, tricky to scale, and less forgiving than software when things break. The summit’s organizer and industry voices argue that better AI models are accelerating the field, while others push for a stronger U.S. strategy to keep pace with global competition. The throughline is clear: humanoids are moving from sci-fi spectacle toward industrial pilots, but timelines and economics remain open questions. (AP News)

1X Strikes a Deal to Push Its Humanoid Robot From “Home” to Factory Floors

TechCrunch reports that humanoid startup 1X has signed a strategic partnership framework tied to Swedish investment giant EQT, aimed at getting 1X’s Neo humanoids into real industrial environments. The headline number: the deal contemplates shipping up to 10,000 robots between 2026 and 2030 to EQT’s portfolio companies, with an emphasis on manufacturing, warehousing, logistics, and similar use cases. The story frames this as a telling pivot—humanoid makers often talk about home helpers, but near-term revenue is more likely in structured commercial settings where labor is costly and tasks can be constrained. TechCrunch also situates the move within an increasingly crowded humanoid field, where partnerships and distribution channels may matter as much as raw technical capability. The key test will be whether pilots translate into safe, reliable, economically compelling deployments at scale. (TechCrunch)

EQT Explains Its Humanoid Robot “Rollout”—As Access, Not a Mandated Buy

Axios follows up on the 1X–EQT announcement with an email Q&A that clarifies what the partnership is—and what it isn’t. EQT Ventures partner Ted Persson characterizes the memorandum of understanding as a framework for collaboration and learning with 1X (an existing portfolio relationship), rather than EQT “selecting a robotics partner” or committing its companies to purchases. Axios highlights EQT’s view that humanoid robotics may become supply-constrained, making early access valuable for portfolio firms that might want to pilot systems addressing labor shortages or difficult operational needs. Importantly, the piece states that any actual adoption, pricing, and volume decisions are left to individual portfolio companies, with 1X handling commercial arrangements directly. The article captures a broader pattern in robotics commercialization right now: investors want optionality—structured pathways to pilots—while the technology, safety cases, and ROI proofs are still being built. (Axios)

HyprLabs Wants to Train Self-Driving “Robots” Faster—With Run-Time Learning

WIRED profiles HyprLabs, a small startup led by Zoox cofounder Tim Kentley-Klay, arguing that advances in AI could compress the timeline and cost of building autonomous driving systems. The company is unveiling “Hyprdrive,” described as a training approach that blends ideas from camera-only end-to-end learning and multi-sensor, heavily labeled pipelines—aiming to learn in real time under human supervision and send back only novel data for fine-tuning. WIRED reports HyprLabs’ two modified Tesla Model 3 test vehicles have collected about 4,000 hours of driving data (~65,000 miles), with roughly 1,600 hours used to train the system—tiny compared with major players. The piece is candid about uncertainty: the company isn’t claiming production-ready safety, but is betting that a leaner “run-time learning” loop can scale into a new category of autonomous robots beyond cars. (WIRED)

AI-Powered Bionic Hand Reduces Mental Load for Amputees in Daily Tasks

Engineers at the University of Utah used AI to make a commercial robotic hand behave more like a cooperative partner than a device that requires constant conscious control. In a EurekAlert summary of the work, researchers integrated proximity and pressure sensors into the bionic hand and trained a neural network on grasping postures, enabling shared human–machine control that improved grip security and precision while reducing cognitive burden for users. The study involved people with transradial amputations and reports that participants could perform everyday actions—like picking up small objects and lifting a cup—using different grip styles without extensive training. The work appears in Nature Communications, and the framing is practical: advanced prostheses can be physically capable, but mentally taxing; shifting part of the control problem to onboard intelligence can make robotic hands feel more intuitive and usable. (eurekalert.org)