DroNet uses deep machine learning for autonomous flight, a labor union takes a stand on package delivery by drone, a tiny radar for sUAS, counter-unmanned aerial systems, and a British drone survey.
Researchers have developed a drone that can autonomously fly through the streets of a city. DroNet uses minimal onboard sensing and is trained with datasets collected by cars and bicycles.
Two years ago, roboticists at the University of Zurich trained a deep neural network with photographs taken by cameras mounted on a hiker’s head. They could then fly a drone along forest paths. Now, along with researchers from a Madrid University, they have used city datasets to train the rules for navigating through streets without running into anything.
DroNet is a convolutional neural network, designed as a fast 8-layers residual network. It produces two outputs for each single input image: A steering angle to keep the drone navigating while avoiding obstacles, and a collision probability to let the drone recognize dangerous situations and promptly react to them.
The researchers publicly release all their datasets, code, and trained networks. Learn more at the DroNet project website, and the research page on deep learning. See also the video DroNet: Learning to Fly by Driving.
The Teamsters union and UPS are holding labor negotiations that cover 260,000 union workers in North America. Reportedly, one of the union demands is a prohibition on “driverless trucks, drones, robots, and other driverless technology.” In Teamsters Union Says ‘No’ to UPS Drones, we find this:
“With a smaller carbon footprint and ever-increasing sophisticated, AI-infused behavior, it makes sense to transition to individual aerial deliveries, not to mention the convenience on behalf of the customer. On the other hand, one completely understands and empathizes with the aversion truck drivers have toward this stark, autonomous future. If it feels like their jobs are being endangered by the incredible exponential growth in technology, it’s because they are. Hence, a series of discussions soon to be cemented into policy, with one side trying to slow things down in order to survive, and another eager to march into the fully autonomous future.”
Aurora Flight Sciences is collaborating with Socionext Inc. on a Radar Flight Control Module that exists as a single-chip 24GHz radar. It includes range measurement software and the radar can detect multiple objects, objects in open spaces, and target distance and speed. The RFCM talks to the flight controller and provides distance, warning and braking signals, preventing head-on collisions with obstacles. Press release: Aurora Flight Sciences and Socionext Develop Radar-Enabled Collision Protection Solution for Drones [PDF].
The DHS National Urban Security Technology Laboratory (NUSTL) System Assessment and Validation for Emergency Responders (SAVER) program conducted a market survey of counter-unmanned aerial systems. The 19 page Market Survey Report: Counter-Unmanned Aerial Systems [PDF] is intended to assist emergency responders in identifying useful products.
The thirteen systems identified range in price from $7,500 to $1.9 million. Eight systems can detect, track, classify and mitigate SUAS; two systems can detect, track and classify SUAS; and three systems can mitigate SUAS.
In a study commissioned by a UK tech firm Nominet, more than 2,000 British adults were surveyed. Nearly three-quarters of the respondents believe that flying a drone should require the equivalent of a driving licence. Thirty-seven percent said the Government should set up a body to manage drone ownership.