Autonomous Cars Before They Get To The Road

in StemSocial2 years ago

For the first time, autonomous vehicles that were never trained in the real world might be safe to drive.


Image by Free-Photos from Pixabay

Experts from MIT created a new simulation system for training self-driving cars. Its goals are to help autonomous systems orientate themselves in several different critical scenarios before the self-driving cars ever get to the streets.

Based on the documentation the self-driving cars should learn to safely react to different situations. Many data sets from several different dangerous scenarios that took place in real life will be included. The simulations will put them into virtual worlds and use them to train artificial intelligence.

Hard Training Make The Combat Easier

Currently, we mostly use systems that rely on data sets gathered from experienced drivers to train autonomous vehicles. But situations in real life can be much more dangerous and the autonomous vehicles need to know how to properly react to them. For example, when there is a risk of a car accident sometimes the best solution might be to get off the road or a line change – even though this would be too dangerous under any other circumstance.

So far there have existed several computer programs that tried to simulate these types of situations on virtual roads. Sadly, the learned reaction often proved to be wrong in real life. This is why the scientists created a photorealistic simulator called Virtual Image Synthesis and Transformation for Autonomy (VISTA).

The driving system should react to situations in it - which takes the system on the brink of a collision or crash and the goal is the make the system react in a way that is safe and ideally allows the vehicle to continue on its journey. MIT cooperated with the research institute of the Japanese car-maker Toyota on the tests.

The First Tests Were Good

After going through 10,000 simulated kilometers the experts tried out the autonomous vehicle in real life. This is the first time when the system was trained in this way – without ever being trailed in normal traffic. The vehicle was capable of safely navigate in environments it has never “seen”.

Scientists are now continuing – implementing more factors such as rainy weather, bright sunlight, the day and night cycle and other changes in the conditions. And apart from this, they want to simulate more complex interactions with other vehicles on the road.


  • If you like the content I’m producing about science maybe you will like the content I produce about gaming as well! Be sure to check out my other posts!

science blog news steemstem steem