Self-driving Truck That Learns Like an Animal

May 26, 2016
Self-driving Truck That Learns Like an Animal

Researchers at Chalmers University of Technology are finding inspiration in evolution’s biological counterparts in the development of a driverless truck. The first public demonstration of the Volvo FH16 will take place on a Dutch A270 motorway on 28 May. That’s when the truck will take part in a competition for autonomous vehicles, within the framework of an EU project called the Grand Cooperative Driving Challenge. an EU project and collaborative competition in which 10–15 universities compete against each other with autonomous vehicles.

Chalmers researcher Ola Benderius says the traditional approach to developing driverless vehicles has been to start with a base vehicle and add on new functions.

Benderius' team has instead chosen to regard the self-driving vehicle as a completely new type of vehicle. A vehicle that is more like an animal, a biological organism, than a technical system. 

“Biological systems are the best autonomous systems we know of. A biological system absorbs information from its surroundings via its senses and reacts directly and safely, like an antelope running within its herd, or a hawk pouncing on its prey on the ground. Before humans walked the earth, nature already had a solution, so let’s learn from that,” says Ola Benderius. “We are trying to design a system that adapts to whatever happens, without pointing to specific situations – and this is something that even the simplest animals can usually do better than existing vehicle solutions.”

The team is developing software called OpenDLV (stands for driverless vehicle) which is available as open source code on the Internet, that complies all the information from the vehicle's sensors and cameras, then converts that data into a format that resembles the way humans and animals interpret the world via our senses.

Instead of just one large program with dedicated functions for all conceivable situations, the OpenDLV team is working on small and general behavioral blocks that aim to make the truck react to various stimuli and adapt to unexpected situations. The truck is programmed to constantly keep all stimuli within reasonable levels, and it will continuously learn to do this as efficiently as possible. This makes the framework extremely flexible and good at managing sudden and new dangers, according to Ola Benderius, just like an animal does.

“Traditionally, the aim has been to try to separate and differentiate all conceivable problems and tackle them using dedicated functions, which means that the system must cover a large number of scenarios. You can cover a large number of different cases, but sooner or later the unexpected occurs, and that’s when an accident could happen,” says Ola Benderius.

Ola Benderius and his project group hope that other researchers around the world can join the project by running and developing the software in their own vehicles. OpenDLV is intended to serve as an academic platform for researchers in many different scientific disciplines, such as vehicle engineering, adaptive systems, computer science and engineering, perception, neurology, and biology, where they can exchange knowledge about how autonomous vehicles should be made to enable their safe, large-scale introduction into society.

Source: Chalmers University of Technology