Thursday, 18 June 2020

Autonomous driving: UniTrento builds cars that learn while dreaming

The prototype was built within the Horizon2020 Dreams4Cars project

Versione stampabile

The car that dreams, built at UniTrento, is as fascinating as a replicant. But it does not aim to become the main character in a science fiction movie. Its only goal is to attract the interest of the car industry. Given that it has already proved its value in a European research project, it may now, with some more engineering, be ready for the market.

The new technology may play an important role in the smart car industry, that is self-driving or assisted driving vehicles which, despite the resources invested, have not succeeded so far in achieving the goal of one billion miles driven with no accidents and need therefore an innovative approach.

Mauro Da Lio, professor of the Department of Industrial Engineering of UniTrento and coordinator of the project, believes in the potential of the system that was developed in the past three years by “Dreams4Cars” within Horizon2020. 

The robotic technology of Dreams4Cars, like a very rudimentary artificial brain, takes inspiration from biology and simulates human cognitive processes both during the driving and after that, in that phase that for humans corresponds to sleeping and dreaming.

In this way the artificial driving agent self-trains to face difficult circumstances that may occur on the road by recombining aspects of real-world experience.

The goal of the project is to develop the artificial cognitive system of these cars when they are not running. The project adopted a novel approach, inspired by the idea that thoughts are chains of simulated actions and simulated perceptions.

The initial goal of Dreams4Cars was indeed to increase the ability of smart vehicles to operate correctly in the case of rare events, those that lead to accidents.

One of the limits of current approaches to achieve autonomous driving is that the system is not enough flexible and capable to manage unexpected events which are rare and therefore cannot be foreseen by designers. In other words, the autonomous driving systems developed so far are not autonomous enough when it comes to the ability to tackle unexpected and unpredictable events. There are now many examples of self-driving vehicles that were not able to perform an action or to perform the right action. 

Now we have one more chance to do better. Weeks ago, the artificial brain of Dreams4Cars was admitted to IP Booster, a European programme that provides free professional advice on intellectual property to develop the best strategies to realise value from research results.

The research team was supported in the application process by the Research support and technology transfer division of the University of Trento and by Fondazione HIT - Hub Innovazione Trentino.

The advanced services for intellectual property strategy offered by the European Commission through IP Booster aim to explore the potential market for the software that mimics human behaviour and to assess the competition.

In particular, researchers requested two high added value services: an evaluation of the invention's potential and possible positioning on the market, to obtain the best possible placement; and advice and support in the negotiation of technology transfer agreements, which are essential for the economic exploitation of research results.

The press release includes comments by Mauro Da Lio, coordinator of the Horizon2020 project “Dreams4Cars”, and further information.