To counter the threats posed by small drones, the U.S. military may have to rapidly step up its R&D timeframes, according to a new report commissioned by the U.S. Army. Small unmanned aircraft systems (sUASs) have become increasingly affordable and sophisticated. With millions of these drones now available worldwide, "It's become very easy for an adversary to use them in nefarious ways," says Albert Sciarretta, chair of the committee behind the new study and president of CNS Technologies in Springfield, Virginia. The U.S. Army asked for a detailed report from the National Academies of Sciences, Engineering, and Medicine that analyzes potential risks from these devices, especially to dismounted infantry (that is, foot soldiers) and lightly armored vehicles. For example, hobby drones could be fitted with lethal weapons such as explosive, chemical, biological, or radiological payloads--or modified to jam military radio signals, Sciarretta says.

The difference between the two pictures is that the one on the right has been tweaked a bit by an algorithm to make it difficult for a type of computer model called a convolutional neural network (CNN) to be able to tell what it really is. In this case, the CNN think it's looking at a dog rather than a cat, but what's remarkable is that most people think the same thing. This is an example of what's called an adversarial image: an image specifically designed to fool neural networks into making an incorrect determination about what they're looking at. Researchers at Google Brain decided to try and figure out whether the same techniques that fool artificial neural networks can also fool the biological neural networks inside of our heads, by developing adversarial images capable of making both computers and humans think that they're looking at something they aren't. Visual classification algorithms powered by convolutional neural networks are commonly used to recognize objects in images.

Two years after World War II, billionaire Howard Hughes personally piloted his "Spruce Goose" troop transport aircraft on the first and only flight of the largest seaplane ever built. It lasted barely a minute. Now, more than 70 years later, a U.S. startup is testing a new seaplane concept--one that could evolve into huge cargo drones that fly 109 metric tons of freight across the Pacific, touch down autonomously over water, and unload at ports around the world. The startup Natilus was founded in 2014 with a dream of building large cargo drones to deliver international freight for about half the price of piloted aircraft, and much faster than ships. In December, Natilus planned to test the water-taxiing capabilities of a small prototype drone with a 9-meter wingspan in San Francisco Bay.

Remember back when you could fly drones without having to pay the government money first, and when the only thing you had to worry about was a midair takedown by an anti-drone hit squad made up of highly-trained Dutch eagles? We're sad to have to report that we probably won't be seeing compelling videos of eagles handling rogue drones anymore, and also that the United States government has flexed its muscles and mandatory drone registration is now back on. You probably remember how the FAA finalized its mandatory drone registration rules just in time for the holiday season in 2015. Any drone that weighed more than 0.55 pounds was required to be registered before being flown outdoors, a process that involved providing your complete name, physical address, mailing address, email address, and a credit card that was charged a one-time fee of US $5. In exchange, you got a unique registration number that had to be visible on all of your drones.

Deep reinforcement learning (DRL) provides a model-agnostic approach to control complex dynamical systems, but has not been shown to scale to high-dimensional dexterous manipulation. Furthermore, deployment of DRL on physical systems remains challenging due to sample inefficiency. In this work, we show that model-free DRL with natural policy gradients can effectively scale up to complex manipulation tasks with a high-dimensional 24-DoF hand, and solve them from scratch in simulated experiments. We demonstrate successful policies for multiple complex tasks: object relocation, in-hand manipulation, tool use, and dooropening.

Robots that can be physically reconfigured to do lots of different things are, in theory, a great way to maximize versatility while saving time and effort. Okay, yeah, that may not sound super exciting, but it means you can teach a dodecapod robot to transition into a septapod robot that can carry stuff with two arms while using a third to point a camera. Programmed in advance, that is, which is fine, except that as robots get more modular and easier to physically reconfigure, it becomes more and more useful to have a generalized system that can dynamically generate gaits (and transitions between gaits) on the fly no matter what the leg configuration of your robot happens to be. The researchers are planning on extending their method to include dynamic gaits, which means things like (we hope) running and jumping, and they're also going to generalize to other morphologies like bipeds and tripeds.

During the Hands Free Hectare project, no human set foot on the field between planting and harvest--everything was done by robots. To make these decisions, robot scouts (including drones and ground robots) surveyed the field from time to time, sending back measurements and bringing back samples for humans to have a look at from the comfort of someplace warm and dry and clean. With fully autonomous farm vehicles, you can use a bunch of smaller ones much more effectively than a few larger ones, which is what the trend has been toward if you need a human sitting in the driver's seat. Robots are only going to get more affordable and efficient at this sort of thing, and our guess is that it won't be long before fully autonomous farming passes conventional farming methods in both overall output and sustainability.

I'm preparing to return to a pastime that I gave up a few years ago: flying radio-controlled model airplanes by first-person view, or FPV. Telemetry to Go: My ground station for receiving video was a battery-powered Raspberry Pi connected to two Wi-Fi dongles [top]. On board the plane, I mounted a camera module attached to a Pi Zero [bottom]. The final nail in the coffin was that I discovered that the Raspberry Pi Zero's camera connector wasn't making a very solid connection to its ribbon cable.

And if a technology incubator program in Toronto, Canada has its way, there may even be quantum machine learning startup companies launching in a few years too. Research in this hybrid field today concentrates on either using nascent quantum computers to speed up machine learning algorithms or, using conventional machine learning systems, to increase the power, durability, or effectiveness of quantum computer systems. An ultimate goal in the field is to do both -- use smaller quantum-computer-based machine learning systems to better improve, understand, or interpret large datasets of quantum information or the results of large-scale quantum computer calculations. Still, says Wittek, despite the technical objections, the number of applicants to this year's quantum machine learning bootcamp and startup accelerator in Toronto exceeded expectations.

Whitney says the device has greater torque per weight (torque density) than highly geared servos or brushless motors coupled with harmonic drives. And more significant: To build an autonomous robot, you'd need a set of motors and a control system capable of replacing the human puppeteer who's manually driving the fluid actuators [below]. John P. Whitney: The original motivation was the same as for the MIT WAM arm and other impedance-based systems designed for human interaction: Using a lightweight high-performance transmission allows placing the drive motors in the body, instead of suffering the cascading inertia if they were placed at each joint. We are learning that many of the "analog" qualities of this system will pay dividends for autonomous "digital" operation; for example, the natural haptic properties of the system can be of equal service to an autonomous control system as they are to a human operator.