Reasons Why Autonomous Vehicles Are Still Prone to Accidents


Photo By smoothgroover22 via Flickr


Even with high-tech sensors and expertly designed computational system, autonomous vehicles still pose certain dangers, like what happened to Elaine Herzberg, the woman killed by a self-driving car owned by Uber while crossing a street.

Autonomous Vehicles Do Not Have True Sight

AVs rely on light detection and ranging systems to remotely sense objects using a pulsed laser. In highly maintained roads, LIDAR can work in measuring distance and collecting map details. However, in poorly maintained roads, LIDAR suffers difficulties because it could not detect the worn out lines. Another drawback of LIDAR is it lacks the capability to deal with certain weather conditions, such as rain and snow. The particles from the pulsed laser simply bounce off in the air particles instead of bouncing off from road obstacles and pedestrians.

Image Recognition Prone to Slight Changes

Using LIDAR, AVs can recognize obstacles on the road while the camera takes a snapshot to find a reference. The captured image is sent to the deep neural network algorithm to determine what the obstacles might be. If recognized, the vehicle should slow down or stop to avoid a collision. But the technology for image recognition remains vulnerable to slight changes. According to researchers from multiple universities in the United States, DNNs can easily be fooled by minor changes, such as mistakenly identifying black people as gorillas.

Poor Reaction to Unexpected Situations

Unlike humans, AVs do not possess a shift reaction when startled. These vehicles have a limited processing power and follow a guided protocol whenever an object appears in front of them. AVs rely on calculations of every factor when approaching objects in motion. Human drivers have the instinct to hit the brake if something like that appears out of thin air.

Driverless self-driving vehicles have a long way to go before it can be used in all kinds of public roads. Developers need to improve the systems used by AVs to avoid hitting a person accidentally.