From The Verge:
Audi, like pretty much every other automaker under the sun, continues to introduce more and more automated driving features for its cars. The next step, says the company, will be machine learning-powered driver assist features that work in constrained situations like driving in traffic and parking. They’ll be available in the new generation of Audi A8 saloons in 2017, but for the moment the German company is testing these features in a much more compact model — specifically, a model car 1/8th the size of the real thing.
You can see this pint-sized vehicle in action in the video above. Using deep reinforcement learning (a type of machine learning that’s essentially trial and error with "rewards" for the computer when it gets something right), Audi’s technicians have enabled it to autonomously search and park in a 3 x 3 meter arena. "An algorithm autonomously identifies the successful actions, thus continually refining the parking strategy," writes Audi in a blog post. "So in the end the system is able to solve even difficult problems autonomously."
Audi isn’t the only player in the auto industry to use machine learning to develop autonomous driving (startup Drive.ai is doing something similar) but it is one of the biggest, and technology it introduces will affect more drivers. Just like Audi’s miniature training car, pretty soon technology will make us all model drivers.
Comments
The build quality on the cars is really impressive. There is a Youtube channel for competitors that starts with an unboxing video:
...followed by a series of videos that go through all the features of the car and the software development environment. All the videos I watched were in German, but auto-generated, auto-translated closed captions were available on most of them and I think I was able to understand about half of what he was saying. Maybe.
The only other robotics competition I've seen that a sponsor put this much work into is the big DJI robotics competition (Rise of the Robomasters - The Verge) Are there any other super-nice competitions out there we should know about?
Instead of simulating a 1:8 car on a 1:8 parking space, they should simulate a 1:64 car on a 1:8 road, with simulated shadows & obstacles. It would accurately simulate the autonomous car most of us will actually have: a 1:8 car on a normal road.
https://www.audi-autonomous-driving-cup.com/fahrzeug/
This vehicle has a different setup:ASUS Xtion RGB-D camra + and what seems to be LIDAR on bumper
Some additional info:
For the development, we rely on the industry standard environment ADTF (Automotive Data- and Time-Triggered Framework). A developer license for the software is available on every vehicle computer. This makes a comfortable development possible directly on the vehicle. The operating system of the measuring computer is Ubuntu Linux, the software of the Arduino control boards a self-development, which is available as source code. Software updates can be downloaded from the website during the preparation phase. New softwares are announced in the forum.
The block diagram from AADC manual is like this.
Interesting -- I didn't know about that one. I've added it to our document of autonomous car resources.
Audi Autonomous Driving Cup 2016: piloted driving on a 1:8 scale
Wow, Audi holds nice competition.
HC SR04 modules are an excellent choice for preventing all manner of high speed collisions.
At 1:44 you can see an Intel based embedded board looking like a KONTRON and probably powered by I7.
Arduino are used to interface the Sonar and outputting serial to Main Computer
At min. 2:02 the video shows 2 Arduino boards being part of the sistem.