Learn how precisely dated fossils from Morocco reveal a population with a mix of archaic and emerging traits, helping clarify when African and Eurasian human lineages began to diverge.
IRVINE, Calif., Jan. 7, 2026 — Terran Orbital, a Lockheed Martin Company and a leading manufacturer of satellite solutions, announced today the appointment of Michael Vishion as vice president of […]
When people think of wind energy, they usually think of windmill-like turbines dotted among hills or lined up on offshore platforms. But there is also another kind of wind energy, one that replaces stationary, earthbound generators with tethered kites that harvest energy as they soar through the sky.
This airborne form of wind energy, or AWE, is not as well-developed as the terrestrial version, but in principle it has several advantages. Power-generating kites are much less massive than ground-based turbines, which reduces both their production costs and their impact on the landscape. They are also far easier to install in areas that lack well-developed road infrastructure. Finally, and perhaps most importantly, wind speeds are many times greater at high altitudes than they are near the ground, significantly enhancing the power densities available for kites to harvest.
There is, however, one major technical challenge for AWE, and it can be summed up in a single word: control. AWE technology is operationally more complex than conventional turbines, and the traditional method of controlling kites (known as model-predictive control) struggles to adapt to turbulent wind conditions. At best, this reduces the efficiency of energy generation. At worst, it makes it challenging to keep devices safe, stable and airborne.
In a paper published in EPL,Antonio Celani and his colleagues Lorenzo Basile and Maria Grazia Berni of the University of Trieste, Italy, and the Abdus Salam International Centre for Theoretical Physics (ICTP) propose an alternative control method based on reinforcement learning. In this form of machine learning, an agent learns to make decisions by interacting with its environment and receiving feedback in the form of “rewards” for good performance. This form of control, they say, should be better at adapting to the variable and uncertain conditions that power-generating kites encounter while airborne.
What was your motivation for doing this work?
Our interest originated from some previous work where we studied a fascinating bird behaviour called thermal soaring. Many birds, from the humble seagull to birds of prey and frigatebirds, exploit atmospheric currents to rise in the sky without flapping their wings, and then glide or swoop down. They then repeat this cycle of ascent and descent for hours, or even for weeks if they are migratory birds. They’re able to do this because birds are very effective at extracting energy from the atmosphere to turn it into potential energy, even though the atmospheric flow is turbulent, hence very dynamic and unpredictable.
Antonio Celani. (Courtesy: Antonio Celani)
In those works, we showed that we could use reinforcement learning to train virtual birds and also real toy gliders to soar. That got us wondering whether this same approach could be exported to AWE.
When we started looking at the literature, we saw that in most cases, the goal was to control the kite to follow a predetermined path, irrespective of the changing wind conditions. These cases typically used only simple models of atmospheric flow, and almost invariably ignored turbulence.
This is very different from what we see in birds, which adapt their trajectories on the fly depending on the strength and direction of the fluctuating wind they experience. This led us to ask: can a reinforcement learning (RL) algorithm discover efficient, adaptive ways of controlling a kite in a turbulent environment to extract energy for human consumption?
What is the most important advance in the paper?
We offer a proof of principle that it is indeed possible to do this using a minimal set of sensor inputs and control variables, plus an appropriately designed reward/punishment structure that guides trial-and-error learning. The algorithm we deploy finds a way to manoeuvre the kite such that it generates net energy over one cycle of operation. Most importantly, this strategy autonomously adapts to the ever-fluctuating conditions induced by turbulence.
Lorenzo Basile. (Courtesy: Lorenzo Basile)
The main point of RL is that it can learn to control a system just by interacting with the environment, without requiring any a priori knowledge of the dynamical laws that rule its behaviour. This is extremely useful when the systems are very complex, like the turbulent atmosphere and the aerodynamics of a kite.
What are the barriers to implementing RL in real AWE kites, and how might these barriers be overcome?
The virtual environment that we use in our paper to train the kite controller is very simplified, and in general the gap between simulations and reality is wide. We therefore regard the present work mostly as a stimulus for the AWE community to look deeper into alternatives to model-predictive control, like RL.
On the physics side, we found that some phases of an AWE generating cycle are very difficult for our system to learn, and they require a painful fine-tuning of the reward structure. This is especially true when the kite is close to the ground, where winds are weaker and errors are the most punishing. In those cases, it might be a wise choice to use other heuristic, hard-wired control strategies rather than RL.
Finally, in a virtual environment like the one we used to do the RL training in this work, it is possible to perform many trials. In real power kites, this approach is not feasible – it would take too long. However, techniques like offline RL might resolve this issue by interleaving a few field experiments where data are collected with extensive off-line optimization of the strategy. We successfully used this approach in our previous work to train real gliders for soaring.
What do you plan to do next?
We would like to explore the use of offline RL to optimize energy production for a small, real AWE system. In our opinion, the application to low-power systems is particularly relevant in contexts where access to the power grid is limited or uncertain. A lightweight, easily portable device that can produce even small amounts of energy might make a big difference in the everyday life of remote, rural communities, and more generally in the global south.
Robert F. Kennedy Jr. says he left a bear cub’s corpse in Central Park in 2014 to “be fun.” Records newly obtained by WIRED show what he left New York civil servants to clean up.
Spend enough time investing in space and expectations change. The industry does not advance through clean inflection points that resolve uncertainty, and progress rarely aligns with the milestones investors are accustomed to tracking. More often, space infrastructure is absorbed gradually into other systems, registering as essential only after it is already embedded. That dynamic, rather […]
Circularly polarized (CP) light is encoded with information through its photon spin and can be utilized in applications such as low-power displays, encrypted communications and quantum technologies. Organic light emitting diodes (OLEDs) produce CP light with a left or right “handedness”, depending on the chirality of the light-emitting molecules used to create the device.
While OLEDs usually only emit either left- or right-handed CP light, researchers have now developed OLEDs that can electrically switch between emitting left- or right-handed CP light – without needing different molecules for each handedness.
“We had recently identified an alternative mechanism for the emission of circularly polarized light in OLEDs, using our chiral polymer materials, which we called anomalous circularly polarized electroluminescence,” says lead author Matthew Fuchter from the University of Oxford. “We set about trying to better understand the interplay between this new mechanism and the generally established mechanism for circularly polarized emission in the same chiral materials”.
Light handedness controlled by molecular chirality
The CP light handedness of an organic emissive molecule is controlled by its chirality. A chiral molecule is one that has two mirror-image structural isomers that can’t be superimposed on top of each other. Each of these non-superimposable molecules is called an enantiomer, and will absorb, emit and refract CP light with a defined spin angular momentum. Each enantiomer will produce CP light with a different handedness, through an optical mechanism called normal circularly polarized electroluminescence (NCPE).
OLED designs typically require access to both enantiomers, but most chemical synthesis processes will produce racemic mixtures (equal amounts of the two enantiomers) that are difficult to separate. Extracting each enantiomer so that they can be used individually is complex and expensive, but the research at Oxford has simplified this process by using a molecule that can switch between emitting left- and right-handed CP light.
The molecule in question is a helical molecule called (P)-aza[6]helicene, which is the right-handed enantiomer. Even though it is just a one-handed form, the researchers found a way to control the handedness of the OLED, enabling it to switch between both forms.
Switching handedness without changing the structure
The researchers designed the helicene molecules so that the handedness of the light could be switched electrically, without needing to change the structure of the material itself. “Our work shows that either handedness can be accessed from a single-handed chiral material without changing the composition or thickness of the emissive layer,” says Fuchter. “From a practical standpoint, this approach could have advantages in future circularly polarized OLED technologies.”
Instead of making a structural change, the researchers changed the way that the electric charges are recombined in the device, using interlayers to alter the recombination position and charge carrier mobility inside the device. Depending on where the recombination zone is located, this leads to situations where there is balanced or unbalanced charge transport, which then leads to different handedness of CP light in the device.
When the recombination zone is located in the centre of the emissive layer, the charge transport is balanced, which generates an NCPE mechanism. In these situations, the helicene adopts its normal handedness (right handedness).
However, when the recombination zone is located close to one of the transport layers, it creates an unbalanced charge transport mechanism called anomalous circularly polarized electroluminescence (ACPE). The ACPE overrides the NCPE mechanism and inverts the handedness of the device to left handedness by altering the balance of induced orbital angular momentum in electrons versus holes. The presence of these two electroluminescence mechanisms in the device enables it to be controlled electrically by tuning the charge carrier mobility and the recombination zone position.
The research allows the creation of OLEDs with controllable spin angular momentum information using a single emissive enantiomer, while probing the fundamental physics of chiral optoelectronics. “This work contributes to the growing body of evidence suggesting further rich physics at the intersection of chirality, charge and spin. We have many ongoing projects to try and understand and exploit such interplay,” Fuchter concludes.
After years of talk, the Pentagon faces hard choices on commercial adoption, missile defense and whether new space capabilities become enduring programs
Connexion
Wired
Discover Mag
