As reported by The Verge: It's no secret that some Model S owners have been doing truly dumb things with their carssince Tesla rolled out autopilot— the temptation to let go of the wheel and let the car take total control is strong, even though Tesla calls it a "beta" and insists that drivers keep their hands on the wheel at all times. In an earnings call Q&A this afternoon, CEO Elon Musk acknowledged that some scary videos had made it onto the internet of autopilot making scary decisions, saying that the company is planning "some additional constraints" around when the feature can be enabled in order to "minimize the possibility of people doing crazy things with it."
He didn't detail what those constraints might be, but it's easy to imagine that Tesla could, for instance, have more rigor around insisting that hands are placed on the wheel — if hands are removed for more than a few seconds, the system could automatically disengage, as some lane-keep systems in other vehicles do today. That'd reduce the chances that an inattentive driver could get caught off-guard by the car doing something unexpected.
Musk's comments come on the heels of Google's October self-driving report, where the company laid out why it decided to work on a fully autonomous car — because drivers aren't very good at taking control of self-driving cars that need to disengage their autonomous functionality. Some Model S owners are learning that better than anyone at the moment, though there's a lot of hope that Tesla's system will get better over time: the company has noted that data from cars in the field is continuously being uploaded in order to refine its maps and algorithms.
As reported by The Verge: When humans can't hack it, it's time to call in the robots.
Ford has announced today that it's working with ASI, a Utah-based firm that specializes in vehicle automation, to sell "robotic testing kits" that help automakers endurance test new cars. The kits effectively turn in-development vehicles into rudimentary R/C cars with a hint of self-driving — they can be programmed to hold a course within one inch of error and have sensor arrays to avoid nearby pedestrians and other vehicles — which is convenient for sending test cars down brutal stretches of test track designed to simulate potholes and other hazards that they'll encounter over their useful lives.
From Ford's release:
Robotically driven vehicles are expected to repeatedly perform tests on torturous surfaces with names like Silver Creek, Power Hop Hill and Curb Your Enthusiasm. These tests can compress 10 years of daily driving abuse into courses just a few hundred yards long, with surfaces that include broken concrete, cobblestones, metal grates, rough gravel, mud pits and oversized speed bumps.
Spending days on end hurtling a development vehicle down these kinds of surfaces sounds about as much fun as punching yourself in the face, which is where the notion of automating it comes into play: by installing Ford's kit, a car company can basically let the car test itself. Ford says the system can be disengaged quickly if an engineer needs to get in and drive, but for the bulk of the testing, the car can bounce and jerk all it wants without endangering the backs and necks of the poor employees trying to get the car ready for real roads.
Ford is refusing to reveal which automakers have picked up the kits so far, but it says that "several automotive OEMs" have gotten orders in.
DJI has been on the forefront of avoidance technology for drones, recently introducing its Guidance system that uses multiple stereo and ultrasonic sensors that allows the drone to automatically avoid obstacles within 65 feet.
Andrew Barry, a PhD student at MIT’s Computer Science and Artificial Intelligence Lab (CSAIL), is looking to push this technology to the next level. Barry and professor Russ Tedrake have created an obstacle-detection system that allows a drone to autonomously avoid obstacles in its flight path while flying 30 miles per hour.
CSAIL posted a fascinating video, which you can watch below, of Barry’s system helping a drone “dip, dart and dive” through a tree-filled field.
The drone in the video, which was made with off-the-shelf components for $1,700, weighs just over a pound and has a 34-inch wingspan. It has a camera on each wing and two processors that are “no fancier than the ones you’d find on a cellphone.”
CSAIL says Barry’s software runs 20 times faster than existing obstacle detection software. Operating at 120 frames per second, the open-source software allows the drone to detect objects and map its environment in real time, extracting depth information at 8.3 milliseconds per frame.
“Sensors like lidar are too heavy to put on small aircraft, and creating maps of the environment in advance isn’t practical,” Barry says. “If we want drones that can fly quickly and navigate in the real world, we need better, faster algorithms.”
So, how does it work? We’ll let CSAIL explain:
Traditional algorithms focused on this problem would use the images captured by each camera, and search through the depth-field at multiple distances - 1 meter, 2 meters, 3 meters, and so on - to determine if an object is in the drone’s path.
Such approaches, however, are computationally intensive, meaning that the drone cannot fly any faster than 5 or 6 miles per hour without specialized processing hardware.
Barry’s realization was that, at the fast speeds that his drone could travel, the world simply does not change much between frames. Because of that, he could get away with computing just a small subset of measurements - specifically, distances of 10 meters away.
“You don’t have to know about anything that’s closer or further than that,” Barry says. “As you fly, you push that 10-meter horizon forward, and, as long as your first 10 meters are clear, you can build a full map of the world around you.”
While such a method might seem limiting, the software can quickly recover the missing depth information by integrating results from the drone’s odometry and previous distances.
“Our current approach results in occasional incorrect estimates known as ‘drift,’” he says. “As hardware advances allow for more complex computation, we will be able to search at multiple depths and therefore check and correct our estimates. This lets us make our algorithms more aggressive, even in environments with larger numbers of obstacles.”
As reported by Design Boom: formed by skype co-founders, starship technologies is a european startup planning on building a fleet of self-driving delivery drones made to transport goods locally within 30 minutes. designed using readily available components, the robots are lightweight and low-cost, enabling the company to bring the current cost of delivery down by 10 to 15 times per shipment.
the fleet will reduce CO2 emissions, and will create unprecedented convenience for individuals while opening up new opportunities for businesses such as parcel firms or grocery stores. the robots are intended to slip seamlessly into the environment, traveling at slow speeds on sidewalks blending safely in with pedestrian traffic. starship technologies are currently testing and demonstrating prototypes and plans to launch the first pilot services in cooperation with its partners in the US, UK and other countries in 2016.
‘our vision revolves around three zeroes – zero cost, zero waiting time and zero environmental impact,’explains ahti heinla, a skype co-founder and CEO at starship technologies.‘we want to do to local deliveries what skype did to telecommunications. with e-commerce continuing to grow consumers expect to have more convenient options for delivery – but at a cost that suits them. the last few miles often amounts to the majority of the total delivery cost. our robots are purposely designed using the technologies made affordable by mobile phones and tablets – it’s fit for purpose, and allows for the cost savings to be passed on to the customer.’ The drones will travel on sidewalks at about 4 miles per hour for package weights of up to 20 lbs. The top lid opens to fit small and medium sized packages and the unit is opened through the use of an app on the customer's smartphone. The units can also be tracked by the deliverer or the customer. Starship's robots are said to be "99-percent" autonomous, which means that although they'll drive themselves, human oversight will be present at all times, ready to take control if the robot gets confused. Starship Technologies is clearly targeting companies that want to compete with services like Amazon's Prime Now one-hour deliveries -- although there's no reason why Amazon couldn't integrate this robot into its supply chain if it proves as cost-effective as claimed. The company still has a lot of questions to answer, though. It says the robots "consume less energy than most light bulbs," but that's not exactly the most precise metric. We also don't know how much it'll cost companies to run and maintain a fleet of bots, or whether Starship can secure the necessary regulatory permission to ride on sidewalks. Assuming it can answer those questions, and the many more that'll arise in the coming months, Starship says it'll launch "pilot services" in partnership with other companies across the US, UK, and other countries in 2016.
As reported by Slate: Digitially connecting cars to each other and to highway infrastructure promises to drastically reduce collisions and traffic jams. But that wireless vehicular chatter comes at a cost to your privacy: A car that never shuts up may be a lot easier to track.
Researchers at the Universities of Twente in the Netherlands and Ulm in Germany have found that they can use just a few thousands of dollars’ worth of equipment to track a vehicle that’s emitting the so-called “connected vehicle” wireless communications proposed for future vehicle-to-vehicle (V2V) connections.
With only two $550 devices strategically planted at intersections on the University of Twente’s 432-acre campus, they were able to follow unique signatures in cars’ radio communications, predicting which of two campus regions the vehicle was in with 78 percent accuracy, as well as the car’s more precise location on a specific road with 40 percent accuracy. Extrapolating from that proof-of-concept, the researchers believe that the same technique, expanded with a few hundred thousand dollars of hardware, could be used by governments or even amateurs to monitor vehicles over an entire small city.
“The vehicle is saying ‘I’m Alice, this is my location, this is my speed and my direction.’ Everyone around you can listen to that,” says Jonathan Petit, one of the authors of the study, which will be presented at the Black Hat Europe security conference next month and was first reported by IEEE Spectrum. “They can say, ‘there’s Alice, she claimed she was at home, but she drove by the drug store, went to a fertility clinic,” this kind of thing … Someone can infer a lot of private information about the passenger.”
The proposed connected car protocol, which the National Highway Traffic and Safety Administration (NHTSA) will consider mandating for the first time in American cars in 2017, uses the Wi-Fi-like 802.11p wireless signal and could allow cars to communicate with both each other and with highway infrastructure like roads or bridges. One NHTSA study in 2010 estimated that the protocol could prevent as many as 81 percent of all vehicle collisions.
But Petit, one of the University of Twente researchers, says that with the range of that wireless communication falling between roughly 300 and 900 feet, it could also serve as a powerful surveillance mechanism. It’s not yet clear how often connected vehicles will vary the unique wireless signatures that identify them, which could limit their use for tracking an individual car. But depending on how long those “pseudonyms” remain constant, Petit argues the connected vehicle protocol could offer a new, relatively cheap form of vehicle tracking that could bolster existing law enforcement tracking techniques like automatic license plate readers. Or, he imagines, hackers could collect and crowdsource data from the system to assemble a database of vehicle movements around entire cities.
“When you do a deployment like this, you need to think about privacy,” says Petit. “It was clear that we needed to perform this attack, to demonstrate that this information is accessible to anyone.”
For their proof-of-concept, the researchers used two Cohda Wireless MK3 radio modules and attached a pair of Smarteq antennae to each one, at a total cost of about $1,100 dollars. With those modules located at two intersections on the Twente campus, they could roughly track a vehicle that contained an active Nexcom 802.11p radio beacon. The graphs below show the location of the intersections where those two modules were planted and a “privacy heat map” of how accurately the researchers could predict a car’s location based on the resulting radio readings.
Wired
Though their two modules only gave them about a 40 percent chance of locating the vehicle at any given time down to a 65 foot area, the researchers extrapolated that a few more modules could give them much more precision. Each added radio module at an intersection offers more information about where a target car has been spotted and which direction it’s turned at a given time. If the researchers were to cover eight of the campus’ 21 intersections (at a cost of $4,400 dollars), for instance, they believe they could predict the location of the target vehicle with 90 percent accuracy.
In their paper, the researchers calculate that they could extend that surveillance to an entire city for less than half a million dollars. They write that the system could cover the nearby Dutch city of Enschede, for instance, with more than 150,000 people and 35,200 acres of land, for around $362,000. And they believe that their “sniffing stations” could easily drop in price—they even speculate that one could likely soon be built for a tenth of the cost using a Raspberry Pi minicomputer—to vastly cut the cost of that tracking. “If you have sufficient coverage of a city, you can track everyone,” Petit says.
Programming vehicles to switch their unique radio signatures more often could alleviate some of those privacy concerns, Petit notes. And Petit acknowledges that industry groups like the Crash Avoidance Metrics Partnership in the US and the Car-2-Car Consortium in Europe are considering that privacy fix. (Neither organization immediately responded to Wired’s request for comment.)
But Petit says more study is needed to understand exactly how much those pseudonym protections can foil tracking. With a high enough density of tracking modules, the researchers caution that someone could overcome even rapid pseudonym-switching to pervasively follow vehicles’ whereabouts. The chart below shows how often the researchers calculate that they could predict the location of a vehicle given different numbers of tracking modules and different lengths of time between vehicle pseudonym changes. (A “privacy score” of 1 means no chance of a vehicle being identified, and a zero means 100 percent certainty of identifying it.)
Wired
Even so, Petit argues that graph should still be evidence that the pseudonym-switching strategy needs to be built into car-to-car communication systems. Though rapidly changing the vehicle’s signature can’t combat some sort of “global observer,” he says, it can still block cheaper, lower resource types of surveillance.
“Pseudonym changing doesn’t stop tracking. It can only mitigate this attack,” says Petit. “But it’s still needed to improve privacy against this mid-size observer…We want to demonstrate that in any deployment, you still have to have this protection, or someone will be able to track you.”
As reported by Fortune: The “look Ma, no hands” approach just paid off big time.
How do you narrowly avoid a nasty car crash without even touching the steering wheel or the brakes? Lucky for one Uber driver, his ride is a Tesla. On a recent rainy and pitch-black morning in Seattle, Jon Hall had just dropped off a passenger and was cruising along the highway in his Tesla at nearly 45 mph when another car made a sharp U-turn in front of him, cutting him off. But Hall was using Tesla’s new autopilot feature, and his electric vehicle stopped immediately, saving him and his self-driving car from a potentially fatal head-on collision. “I wasn’t even able to honk the horn before the car came to a stop,” Hall wrote in an online discussion of the video of the incident, which he posted on YouTube, saying, “I did not touch the brake. Car did all the work.” The video was captured on Hall’s dashboard camera, though it does not include sound because he had disabled audio while driving Uber passengers in order to comply with Washington privacy laws. (“I screamed, honked, and yelled,” he explains.) Watch the full video of the autonomous vehicle’s feat here:
/cdn0.vox-cdn.com/uploads/chorus_asset/file/4161938/press01_autopilot_dash.0.jpg)
He didn't detail what those constraints might be, but it's easy to imagine that Tesla could, for instance, have more rigor around insisting that hands are placed on the wheel — if hands are removed for more than a few seconds, the system could automatically disengage, as some lane-keep systems in other vehicles do today. That'd reduce the chances that an inattentive driver could get caught off-guard by the car doing something unexpected.
