The British startup Wayve has built a machine-learning model that can operate two different types of vehicles: a passenger car and a delivery van, according to Technology Review.
According to the company, both vehicle types are equipped with the same six-camera sensor suite. However, due to the van’s body construction, the sensors are mounted higher and at different angles than those on the sedan.
As a result, the input data for the model differ between the vehicles. Yet the AI learned to operate them from any viewpoint, the company said. The algorithm has also adapted to the van’s larger size and mass.
The Wayve team trained the model using reinforcement learning and thousands of hours of driving data. Engineers launched a driving simulation that, in just 80 hours, learned to drive the van.
The developers said the model adapted to the new vehicles faster than anticipated during the design phase.
Subsequently, engineers tested the algorithm on the streets of London. Wayve’s safety operator, Naomi Standard, said the van performed well in the city’s narrow streets and successfully navigated a range of obstacles.
The developers say this is the first instance of a single AI driver learning to operate different types of vehicles. They believe it will help the system scale more quickly and be deployed in various cities with fewer modifications.
“It’s like going to a new place and renting a car—you can still drive,” said Jeff Hawk, Wayve’s vice president of technology.
Looking ahead, the company aims to be the first to deploy computer-vision-based autonomous cars in 100 cities.
Earlier in September, Cruise began delivering Walmart supermarket orders using autonomous vehicles.
In August, Waymo launched a driverless robotaxi with no safety driver at the wheel in central Phoenix.
Subscribe to ForkLog News on Telegram: ForkLog AI — all the news from the world of AI!