Autonomous cars are usually associated with extra careful driving, simply to make them as safe, for both occupants and pedestrians, as possible. So much so that there have been incidents of people attacking them in frustration on road. But what if we these cars were made to handle curves on a track at high speeds, just like a human. Would you believe they can do it? You better!
The confirmation comes from the recent work of engineers at Stanford University. The engineers created a neural network that enables driver-less cars to operate just like race car drivers, on their own. That means performing high-speed, low-friction turns as well as top of the power accelerations.
The artificial intelligence that the team used for this involved creating a human brain-like neural network which worked on models that learn from pre-recorded data. Regular patterns is then sought from this data which in turn enables the model to make appropriate choices on the track. The physical form of these neural networks is a high-powered GPU, stored in the trunk of vehicles.
Autonomous Audi TT (Image: Stanford/ YouTube)
Also read:?India's First AI-powered Electric Motorcycle Is Already Under Testing, These Spy Shots Reveal
In this instance, the team from Stanford used data from 200,000 motion samples, including test drives performed on surfaces like snow and ice. The program was embedded in two different cars, one named Shelley which is an autonomous Audi TTS, and another called Niki, a self-driving Volkswagen GTI. The cars were tested out at Thunderhill Raceway in the Sacramento Valley.
Shelley performed comparable lap times on the track as an amateur driver while using just a simple physics-based autonomous system. Similar results were received when the team loaded its own neural network on the car, even though the neural network did not have enough information about road friction. You can watch the autonomous Audi making fast paced turns in the video below:
Also read:?Hackers Trick Tesla Model S Into Turning Towards Oncoming Traffic Using Stickers On The Road
The bigger motive here though, is the fact that such autonomous measures can make the self-driving cars operate much like humans. This is desirable in certain scenarios where quick decisions at high speeds need to be made in order to avoid a crash. Also, the addition of high-speed manoeuvring?will help autonomous cars expand their capability beyond those of the majority of humans, which is necessary as around 94 percent of crashes are caused by human errors.
As per the team, additional data would further help the cause to make the system learn of more such scenarios. Nathan Spielberg, a mechanical engineer graduated from Stanford said, ¡°our work is motivated by safety, and we want autonomous vehicles to work in many scenarios, from normal driving on high-friction asphalt to fast, low-friction driving in ice and snow.¡±