Driverless or autonomous cars are becoming popular these days. They were designed keeping in mind the safety of drivers and people on the road. However, researchers at the University of California, Irvine (UCI) have identified a possible risk that could trick the collision-avoidance system of driverless cars leading to an inevitable accident. Collision avoidance is installed to prevent or decrease the chances of a collision. It monitors the vehicle’s speed, the speed of the counter vehicle and the distance between the two to avoid a crash. The system contains various sensors and technologies, including radars, cameras and lasers, to spot an imminent collision.
Chen and his colleagues focused their research on identifying security vulnerabilities of a planning module to know how exposed it is to accidents. A planning module is a part of a software code that builds the collision-avoidance system to control the motion of driverless cars. This component checks the decision-making process of the vehicle while changing lanes, cruising, slowing or stopping down.
“The vehicle’s planning module is designed with an abundance of caution, logically, because you don’t want driverless vehicles rolling around, out of control,” said Ziwen Wan, UCI PhD student in computer science. “But our testing has found that the software can err on the side of being overly conservative, and this can lead to a car becoming a traffic obstruction, or worse.”
The researchers designed a tool named PlanFuzz which can efficiently identify vulnerabilities or risk possibilities of accidents in popularly used automated driving systems. They used it to analyze the three different parameters of behavioural planning implementations of open-source as well as industry-grade autonomous driving software systems, Apollo and Autoware as shown in the video demonstration.
During the experiment, the researchers observed that bicycles and cardboard boxes positioned on the roadside caused vehicles to stop permanently on empty intersections and thoroughfares. In another experiment, automatic cars whose collision system detected a nonexistent risk did not change lanes as expected.
Follow us on Medium!