Autonomous cars: New AI prevents accidents

Autonomous cars: New AI prevents accidents

06.11.2019: Detailed risk assessment to prevent collisions at confusing intersections MIT test drives in a toy city (Photo: mit.edu) Researchers at the Massachusetts Institute of Technology (MIT) http://mit.edu, together with developers from the Jap

Detailed risk assessment to prevent collisions at confusing intersections


 
MIT test drives in a toy city (Photo: mit.edu)





Researchers at the Massachusetts Institute of Technology (MIT) http://mit.edu, together with developers from the Japanese car manufacturer Toyota, have developed a method with which the collision risk of autonomous vehicles can be almost completely eliminated, especially when entering intersections. It incorporates several factors into the decision-making process.



A bundle of data is included



First, the system takes into account its own uncertainty in estimating the risk of colliding with an initially invisible vehicle. It then takes into account all obstacles, such as houses or hedges, that affect visibility. But it also takes into account noise, the possible inattentiveness of other drivers and the speed of cars driving on the crossroads.


From these factors, the system calculates the probability that there will be no collision when the vehicle enters the intersection and gives appropriate instructions. If the result is unsafe, it lets the car advance slowly to collect more data.



It also calculates the time it takes the vehicle to cross the intersection. If the calculated probability is high enough for the vehicle to pass without a collision, the system instructs the driver to leave quickly and without a break.





Over 100 tests in Toy City



"In a way, it is a predictive model based on probabilities," says Daniela Rus, director of the MIT Laboratory for Computer Science and Artificial Intelligence at http://csail.mit.edu . The researchers have tested their system, which can also be used to assist human drivers, in over 100 left-turn manoeuvres on busy intersections with no traffic lights in a model city. In all cases it turned out well.



On the other hand, vehicles that only use conventional sensors such as cameras and radar to orient themselves could not reliably prevent collisions. Before the prediction model can actually be used to support drivers and make driverless vehicles safer, the system has to be tested in real vehicles but in real traffic, say the researchers.



Although there are other solutions, they cannot be used because of the great computing effort involved in the vehicles. Next, the scientists want to consider other risk factors, such as pedestrians.





Wolfgang Kempkens

Permanent-URL: http://www.automobilsport.com/autonomous-cars-new-ai-prevents-accidents---200049.html

06.11.2019 / MaP

More News

2019 PARIS-SACLAY Autonomous Lab - Groupe Renault starts public trial
14.10.2019
.... of its on-demand car service using autonomous, electric and shared ZOE Cab •          From 14th October to 8th November 2019, a panel of around 100 people  more >>
Audi publishes user typology and emotional landscape of autonomous driving
25.09.2019
From the suspicious driver to the tech-savvy passenger: In a representative online study, the initiative “&Audi” shows how attitudes to autonomous driving are related to lifestyle.    Initiative “&Audi&r more >>
Hyundai Motor Group and Aptiv to Form Autonomous Driving Joint Venture
23.09.2019
 •    Joint venture advances the development of production-ready autonomous driving systems for commercialization of Level 4 and 5 self-driving technologies•    Autonomous driving platform will be available for robotaxi providers, fleet operators, and automotive manufacturers in 2022•   more >>

2001-2020 automobilsport.com full copyright