Autonomous vehicles sharing sensory data would improve safety, study suggests
Image credit: Dreamstime
A team of Swiss researchers have developed a software framework which enables multiple intelligent cars to pool their sensory data, improving safety and reliability on the road.
Arguably the greatest challenge facing manufacturers of autonomous vehicles is demonstrating to policymakers and public that they are safe to use. All autonomous vehicles under development now are equipped with a range of different sensors, including Light Detection and Ranging (Lidar) sensors – which rapidly measure the distance to surrounding objects by reflecting pulses of laser light – as well as optical cameras, mapping and navigation systems.
While these systems appear to provide a sufficient degree of situational awareness to navigate with on test drives, researchers from Ecole Polytechnique Fédérale de Lausanne (EPFL) suggest that safety and reliability on the roads could be improved by pooling information between cars.
For instance, if an autonomous car on the motorway is assessing whether or not to overtake the car in front, it could use information from the other car to assess the relative speeds of both vehicles, as well as how busy the roads out of view are, allowing for a more informed decision to be taken.
A software framework developed by researchers at EPFL allows for the fusing of data from one vehicle’s sensors with data provided by the sensors of nearby vehicles, significantly extending the field of view for autonomous cars.
According to Milo Vasic, the EPFL PhD student who developed the algorithms used as the basis for the software, this “cooperative perception” makes manoeuvres such as overtaking safer and more fluid.
The team, based at the Distributed Intelligent Systems and Algorithms Laboratory at EPFL, tested their algorithms under various simulated scenarios, with and without the presence of “cooperative vehicles” before bringing the method to the road.
The researchers fitted two Citroën electric cars with a computer to run the software, a battery, optical camera, localisation system and router to enable Wi-Fi communications between the two vehicles.
“These were not autonomous vehicles,” said Professor said Professor Alcherio Martinoli, who heads up the Distributed Intelligent Systems and Algorithms Laboratory. “We made them intelligent using off-the-shelf equipment.”
There were multiple obstacles to overcome when testing the system on the road. It proved particularly challenging to fuse the data accurately, as without relative localisation, a single pedestrian or other object could appear as two figures in the fused image rather than one. By using other signals provided by the sensors, the researchers were able to adjust their algorithms to improve the accuracy of localisation.
“Although the tests involved only two vehicles, the longer-term goal is to create a network between multiple vehicles as well with the roadway infrastructure,” said Vasic.
Although the researchers still have some way to go in refining their algorithms and wrangling with questions regarding security and litigation before these cooperative networks reach the road, they hope that, eventually, this technology could not just enhance safety, but also improve traffic flows and save energy.