Startup developed an AI model for driving different types of cars
The British startup Wayve created a machine learning model that drives two different types of cars: a car and a delivery van. Technology Review writes about this.
According to the company, both types of vehicles are equipped with the same set of six cameras. However, due to the structural features of the van body, the sensors were placed above and at other corners than that of the sedan.
As a result, the input data for the model differ between vehicles. However, AI learned to manage them from any viewing angle, the company said. The algorithm was also adapted to greater size and mass of the van.
To train the model, WAYVE specialists used reinforcement training and thousands of hours of driving data. Engineers launched driving simulation, which in just 80 hours learned to manage the van.
According to the developers, the model adapted to new cars faster than expected at the design stage.
Then the engineers tested the algorithm on the streets of London. According to the Wayve Security operator Naomi Standard, the van managed well with the narrow streets of the city and successfully overcame various obstacles.
The developers argue that this is the first time when the same AI driver learned to manage different types of cars. In their opinion, this will help the system scale faster and will be introduced in various cities without significant improvements.
“This is like when you go to a new place and rent a car-you can still drive a car,” said Wayve Vayve Vaya President Jeff Hawk.
In the future, the company plans to become the first in the market that placed unmanned vehicles based on computer vision in 100 cities.
Recall that in September Cruise began to deliver orders from Walmart supermarkets using unmanned cars.
In August, Waymo launched a robotaxi without a security operator driving in the center of Phoenix.
Subscribe to FORKLOG news at Telegram: Forklog AI – all news from the world of AI!