I am decidedly less optimistic about what I perceive to be a rush to field systems that are absolutely not ready for widespread deployment, and certainly not ready for humans to be completely taken out of the driverâs seat.
According to reports, Cummingsâ objections focused primarily on driving in bad weather, and on cyber-security.
Both of those seem to me like known and solvable problems. And, in that vein, representatives from Google and GM testified that they were much more optimistic about self-driving cars.
But itâs helpful to remember that all the smart people arenât 100% in agreement about this.
Honda has flown below the radar in the self-driving car world, compared to manufacturers like Tesla and Ford.
This week, though, Honda announced its own ADAS suite that allows drivers to ride along with their hands off the wheel and their feet off the pedals, âas lane markings are visible and another vehicle is in front of the carâ.
The requirement for another vehicle to be out front is particularly interesting. It may map to the US Armyâs project of building self-driving truck convoys that can follow the truck ahead of them, with a human driver at the front of the line.
As manufacturers each launch their own ADAS systems, with different names and features, itâs increasingly difficult to keep track of who supports what. Iâll be curious to see whether this represents a big step into the self-driving car world for Honda, or just a natural addition of features that more or less match what competitors offer.
A big selling point for Honda is price, so maybe this represents the ability to mass product ADAS systems for $20,000 cars.
The National Highway Transportation Safety Administration has recently been an important booster for self-driving cars.
Although itâs not clear how much sway the agency has (most US transportation laws are implemented at the state level), the NHTSA has been trying to clear a legal path for self-driving cars.
Friday, however, the NHTSA qualified their support somewhat, and made clear what development direction they favor.
The National Highway Traffic Safety Administration (NHTSA) said on Friday that self-driving carsâââwhich do not have steering wheels or brake/gas pedalsâââcan be made available for purchase by in the US only after they clear some potentially âsignificantâ legal hurdles.
However, NHTSA spokesman Gordon Trowbridge said pointed out that a new report released by the agency on Friday shows there were fewer legal hurdles in deploying self-driving cars with human controls, compared to fully autonomous cars.
The NHTSA appears to taking sides in the split over whether to develop âLevel 3â vehicles, or skip right to âLevel 4â. Level 3 vehicles require mechanisms for the driver to take over from the computer, whereas Level 4 vehicles entrust the computer to drive in all situations.
Tesla is working on, and indeed has released, Level 3 technology. Google and many auto manufacturers, however, are hoping to skip Level 3, citing the complexity of transferring control between the human and computer drivers.
Image annotation is an interesting and surprising problem that many autonomous vehicle researchers are struggling with.
The issue is that its easy to send a car and a camera out into the world to collect data, but its time-consuming and expensive to label that data.
And the labeling is necessary in order to train the machine to read the data.
Think about lane lines, for example. Lots of companies can now capture millions of images of roads, in all sorts of conditions. But in order to find the lane lines, the computer models have to be trained. And training the models involves telling them where the lane lines are in the sample images.
Lots of academic researchers use Amazonâs Mechanical Turk, a job system where cheap workers overseas can pick up manual tasks from companies in the US and elsewhere. But that is both expensive – even the worldâs cheapest workers become expensive when asked to perform billions of tasks – and slow.
There doesnât seem to be a solution yet to this problem.
Auto manufacturers have made two big acquisitions this week:
Toyota hired the entire staff of Jaybridge, a Massachusetts-based autonomous driving company. This wasnât actually billed as an explicit acquisition, but rather as a type of acqui-hire or contract.
The Model 3 is a mass-market vehicle, priced at $35,000. In some locations, tax credits will lower cost to $25,000 or less.
So a big question is, will the Model 3 bring self-driving cars to the masses?
The answer would seem to be yes, given that Elon Muskâs stated timeframe of 2â3 years until self-driving cars, and the Model 3’s 20-month launch countdown.
However, some analysts wonder whether the Model 3 can include the necessary hardware and still maintain its price goal.
According to The Motley Fool:
With Model 3âs $35,000 starting price at half the starting price of Model S and well below the $80,000 starting price of Model X, Tesla may be planning to use a more advanced autopilot hardware system in its more expensive Model S and Model X.
Iâve finished NVIDIAâs introductory Deep Learning course, and Iâm now starting Udacityâs.
These courses outline the construction and use of Deep Neural Networks (DNNs) for image processing and text recognition. Theyâre great!
Here are some of the highights:
DIGITS: This is NVIDIAâs Deep Learning visualization program. It presents a GUI for both building and reviewing DNNs.
Caffe: There are several frameworks for building DNNs, but Caffe seems the most straightforward. Although it is written in C++ and provides a Python interface, no coding is required to get started. This is because Caffe can be configured with Google Protobuf, a JSON-like text format.
Theano: NVIDIAâs course advises that the various Deep Learning frameworks are âmore similar than they are differentâ, but Theano is at least syntactically different than Caffe. Theano is a Python symbolic math library, on top of which various packages offer DNN capability.
Torch: Torch is Facebookâs tool of choice for DNN research, and it gets support from Google, NVIDIA, and other major Deep Learning companies, as well. It uses the Lua programming language (yay, Brazil!).
TensorFlow: In the same way that Torch is Facebookâs go-to DNN tool, TensorFlow fills that role for Google. Like many Google projects, it is Python-based. I am just diving into TensorFlow now, via Udacityâs course, so I may have more to say later.
cuDNN: This is NVIDIAâs library for parallelizing DNN training on the GPU. It is the key to building large neural networks, and all of the DNN frameworks integrate with it. As Googleâs Vincent Vanhoucke relates, neural networks went through a period of popularity in the eighties and nineties, and then slumped in the 2000s, as CPUs werenât able to provide enough power to train large networks. The publication of AlexNet (2012), showed that the use of GPU parallelization could provide massive training acceleration. This revolutionized the field.
Convolutional Neural Networks (CNNs): Convolutional Neural Networks are a building block of DNNs that involve learning on small parts of an image and then tiling the neighboring small parts to learn over the entire image. This blocking and tiling reduces the learning complexity, which is especially important for large images.
BMW is turning its focus to self-driving cars. âThe youngest head of a major carmaker, Krueger is part of a generational shift thatâs now looking for ways to respond to new challengers such as Apple Inc. and Google, which the BMW CEO on Monday described as competitors.â
Arrowstreet is a Boston-based architectural firm that specializes in parking garage design.
So theyâve naturally been thinking about what autonomous vehicles mean for the future of parking garages.
Their hypothesis is that garages will evolve in two stages.
The first stage involves modifying conventional parking garages to support both conventional cars and self-driving cars. Conventional cars would park closer to the pedestrians, to minimize effort for drivers, and autonomous vehicles can park themselves further back (or up) in the garage.
The second stage of development will be garages oriented completely towards autonomous vehicles. These garages will have very tight parking in a restricted area, with vehicles driving out to specified zones for passenger retrieval.
Arrowstreet even thinks that conventional garages can be retro-fitted for residential and commercial space once parking is no longer a necessity.
David Silver 