Eye on AI - August 27th, 2021
Welcome to Aigora's "Eye on AI" series, where we round up exciting news at the intersection of consumer science and artificial intelligence!
This week, we’ll be taking a hard look at the future of travel, including the viability of Tesla’s autopilot system going fully autonomous and a new update (with video!) on the Virgin Hyperloop mass-transit system.
Tesla Has a Camera Problem in It's Self-Driving Vehicles
Since 2018, eleven Teslas have struck stationary emergency vehicles tending to the scene of an accident. In every instance, the Tesla vehicles were on autopilot. This revelation has led to a new NHTSA investigation into the Tesla autopilot system which, according to the Slate article “Why Teslas Keep Striking Parked Firetrucks and Police Cars,” has long been known to have a pattern-matching problem.
“[Tesla’s network] is basically a super-duper, very sophisticated pattern-matching scheme,” writes Raj Rajkumar, an electrical and computer engineering professor at Carnegie Mellon University who specializes in self-driving vehicles. “The problem is that, in the real world, it is given an image where it sees an obstacle that it has never seen before. The patterns really do not match, so it will not detect it as a vehicle. For example, when the first person was killed using the Tesla autopilot in Florida, the truck [hit by the Tesla] was perpendicular to the direction of the motion. ”
Think of it this way: neural networks are trained to recognize patterns. You feed them data, and eventually, they begin to recognize patterns within that data with remarkable accuracy. It’s why some neural networks are impossible to beat at chess, and others are able to perform simple, repetitive tasks so effectively. Those systems were trained on data specific to those tasks. With neural network training, it’s all about the data.
What neural networks aren’t about is imagination. They don’t problem-solve well when unrecognized patterns appear, which happens almost every day in the world. During such circumstances, neural networks can have issues. For example, many Tesla autopilot accidents occur at night, when the flashing lights of emergency vehicles may confuse Tesla’s cameras. Tesla’s neural network reads the lights as pixels, trying to fit them into a pattern within the data it has been trained on. Because no similar images could be found, the network likely didn’t know to associate the light with a vehicle.
Images captured during accidents could be used to train the Tesla neural network to avoid accidents after-the-fact, but that won’t solve the problem entirely. There will always be new situations that don't quite match up with the network’s training, and alternatives may be needed to assist Tesla’s cameras solve the pattern recognition issue.
“With [the] combination [of radar and lindar with the camera], you can definitely detect and deal with the situation,” continues Rajkumar. “But of course, Tesla does not use radar at this point, and it’s not going to use lidar. It’s a twofold problem: With the camera alone, the training is not sufficient, and meanwhile they [Tesla] forgo all the other sensors.”
It had been Tesla’s hope to go fully-autonomous. But the addition of lidar, which is likely needed to make autonomous driving a reality, may not be economically viable. We’ll have to wait and see if the issue can be safely and financially addressed.
Virgin Hyperloop Future Plans Unveiled in Recent Video
Let’s conclude with an update on what could be the future of high-speed mass-transit: the Virgin Hyperloop! Based on Elon Musk’s hyperloop idea from 2013, the Virgin Hyperloop is one of many magnetic transportation ideas currently in development that could soon alter the future of high-speed public transit. Virgin’s differentiator is its use of separate travel “pods” as opposed to a longer train-like passenger carrier, which would allow for more navigation options.
“With the Virgin Hyperloop, passengers would sit in a pod that would not be attached to other pods—each would be a vehicle traveling on its own,” writes TechXplore contributor Bob Yirka. “By not hooking the cars together, the pod design allows for individual pods to break away from other pods around them (in caravans) and head off into a secondary tube.”
Virgin’s pods travel inside a magnetic tube to reach potential speeds of up to 1,070 km/hour, with a designed battery power option that will be more efficient than those used on maglev trains. Pods would produce no direct emissions, have limited sound interference (because they’re traveling within a tube), and could theoretically be transferring passengers by the tens of thousands across the country at airplane speeds by as early as 2027.
In theory, I love the idea! It’s fast, energy-efficient, and could reduce our reliance on fossil fuel-dependent airplanes. But there are miles to go before we hyperloop. Government regulations will likely impede development. Financial or tube-development may come into play, not to mention the other competitors. Like the space race, the hyperloop race has turned into a measuring contest for the mega-wealthy. Keep your seatbelts fastened –– this should get interesting.
That's it for now. If you'd like to receive email updates from Aigora, including weekly video recaps of our blog activity, click on the button below to join our email list. Thanks for stopping by!