A new software framework lets multiple artificial intelligence (AI) models plug directly into a fusion experiment’s control system, reading plasma measurements and issuing commands in about 20 milliseconds, far faster than any human operator. Researchers tested the framework in five experiments at the DIII-D National Fusion Facility, a Department of Energy user facility in San Diego. The framework predicted and prevented plasma instabilities and cut the time to deploy a new AI model from months to days.

Touch interfaces in devices typically require power, wiring, and specialized components, limiting where they can be used. Researchers from Japan have developed a touch interface that can be made by printing toner patterns onto a PVC sheet with a conventional laser printer. The sheet can recognize finger movements, patterns, characters, and users, while remaining stable after 1,000 sliding cycles. The approach could enable inexpensive, customizable controls for wearable and Internet of Things (IoT) devices. 

Multiple drones can be used simultaneously to quickly survey remote or dangerous areas, but keeping the fleet connected while flying efficient routes remains a major challenge. Now, researchers from China have tackled this issue by jointly optimizing flight paths and network topology, using reinforcement learning and convex optimization to maximize data throughput while keeping the fleet’s communication network robust. Their findings, validated via simulations and field experiments, serve as a foundation for drone fleet coordination.

The Perspective through collaboration between NUS CDE, University of Michigan, and the University of Washington proposes a step-by-step approach for turning patterns found by AI into hypotheses that scientists can test and, if validated, apply in research and engineering.

High pressure-temperature experiments using a large-volume multianvil apparatus show that subducted carbonate addition promotes Ca-Mg exchange with silicates and enhances Ca incorporation under Fe-rich conditions, significantly boosting davemaoite exsolution in carbonate-altered oceanic crust. This reaction concentrates davemaoite to 12–33 vol.% near 560 km depth, producing the 2–4% impedance contrast required to explain the seismic splitting of the 520-km mantle discontinuity. Consequently, deep carbon cycling plays a crucial role in actively reshaping mantle mineral equilibria.

SUTD researchers have found that older adults are most open to embracing social chatbots that feel authentic and socially engaging but not too human, offering clear lessons for designers of AI companions.