Physics World

Ukraine has announced a large-scale programme to build an 'army of robots', described by officials as a 'Manhattan Project for physical AI'. The initiative aims to develop autonomous ground and aerial systems at scale for battlefield use, drawing comparisons to the World War Two atomic research effort. The announcement has drawn international attention given Ukraine's ongoing war with Russia and the growing role of drones and robotics in modern combat.

Researchers publishing in Nature describe a physics-informed machine learning approach for calibrating wearable electrochemical sweat biosensors used to monitor metabolites. The method combines physical models of sensor behaviour with data-driven learning to achieve robust calibration and physiological validation, aiming to make wearable biochemical monitoring more reliable in real-world conditions.

Bloomberg reports that driverless trucks may provide a model for how autonomous AI systems can be deployed safely without harming people. The argument is that self-driving freight has operated under strict testing, regulation and limited-use conditions, offering a template for governing more advanced robots. The piece adds to ongoing debate about how to manage AI risks as automation expands into physical industries.

Gecko Robotics' chief executive is arguing that humans must remain in control as artificial intelligence moves into the physical world through robots and industrial automation. The company builds robots used to inspect critical infrastructure such as bridges, power plants and pipelines. The intervention adds a robotics industry voice to the ongoing debate over how much autonomy should be granted to AI-driven machines in safety-critical settings.

Nvidia has reached a settlement in a trademark lawsuit concerning its 'Modulus' AI software, according to a Reuters report. The dispute centred on Nvidia's use of the Modulus name for its framework for building physics-based machine learning models. Terms of the settlement were not disclosed in the report, and the company has not issued a detailed public statement on the outcome.

Niantic Spatial says its spatial AI technology is being applied to real-world robots, extending the company's geospatial mapping work beyond games and consumer apps into robotics. The move signals the firm's push into physical-world AI, where spatial understanding helps machines navigate and interact with their surroundings.

An AI system has reportedly set a record by independently computing a 9-loop physics amplitude, a Feynman diagram calculation far beyond typical human work. Most real-world perturbative calculations in quantum field theory stop at 2 or 3 loops due to their complexity. The result highlights how machine learning is beginning to automate some of the most demanding mathematical tasks in theoretical physics.

A project called Roboharm is asking whether frontier robot policies actually refuse unsafe instructions. The work examines how AI systems controlling robots respond to harmful commands, a safety question increasingly urgent as AI models are deployed in physical robotics. Discussion is focused on how well current safeguards carry over from language models to embodied systems.

The Albuquerque Journal takes readers inside Sandia National Laboratories' Z machine in New Mexico, the world's most powerful pulsed-power facility. Originally built to study materials under nuclear weapons conditions, the device is now central to research on fusion energy, exploring how the same physics that powers warheads could one day deliver clean, abundant power.

Black Forest Labs has released Flux 3 Action, a 7-billion-parameter open-weight world action model aimed at robotics, available via Hugging Face. The model is designed to let robots perceive and act in physical environments. Early discussion among developers is focused on the unusual choice of an image-model company entering robotics and on how a relatively compact 7B model performs on real robot tasks.

India head coach Gautam Gambhir is making player fitness the central priority in the team's buildup to the 2027 ICC Cricket World Cup. The approach signals that selection and workload management will be tied closely to physical readiness, with the tournament in South Africa, Zimbabwe and Namibia seen as the long-term target for the squad's core players.

Startup Midcentury has raised a $15 million seed round to build training data for physical AI systems such as robots and autonomous machines. The funding signals continued investor appetite for startups supplying high-quality data to the physical AI sector, where collecting real-world sensor and motion data is considered a key bottleneck for training capable models.

A British startup is turning video game controller inputs into training data for AI models designed to navigate the physical world. The idea is that human gameplay, including mistakes and clumsy manoeuvring, offers rich examples of how people learn to control complex systems, which could help robots and other embodied AI improve. The approach is drawing attention as a cheaper alternative to real-world data collection.

Cash hoarding is on the rise in the UK, and commentators argue it shows people still want money they can hold and use without asking permission from banks or authorities. The claim ties the surge in physical banknote holding to broader scepticism about financial institutions and renewed interest in alternative, permissionless forms of money.