Autonomous vehicles (AVs) are facing challenges in effectively identifying children, according to a recent study. The study, conducted by an internal safety assessment at GM’s autonomous robotaxi division, Cruise, highlighted the difficulties AVs encounter when it comes to recognizing children due to their smaller size, erratic behavior, and potential concealment inside objects such as strollers. While it is reasonable to expect AVs to exercise extra caution around potential child pedestrians, the study suggests that Cruise AVs did not demonstrate additional care in such situations.
Cruise internal materials reviewed by The Intercept revealed that despite being aware of the AVs’ inadequate ability to recognize children, the company continued to expand its services. This revelation follows the recent revocation of Cruise’s AV taxi permits by the California Department of Motor Vehicles (DMV). The decision came after one of Cruise’s vehicles struck a woman who had already been hit by another vehicle and was dragged along the street. Further investigation revealed that Cruise had withheld crucial footage of the incident from state regulators.
In response to The Intercept’s findings, Cruise personnel emphasized the company’s commitment to caution around children. However, internal documents suggest otherwise, as an unreported safety assessment within the company noted the AVs’ potential failure to exercise additional care around children. These concerns have also been reflected in simulated tests conducted by Cruise.
One of the challenges faced by AV technology, including fully autonomous vehicles and advanced driver aid suites, is the reliance on machine learning to improve performance. However, experts suggest that machine learning is flawed in terms of safety technology, as it primarily learns from past incidents. Phil Koopman, an engineering professor specializing in AV safety, explains that this approach is not well-suited for safety purposes since it cannot anticipate unseen scenarios.
The study underscores the need for AVs to improve their ability to detect and respond to children, particularly given the potential risks associated with their behavior and size. While these findings raise concerns about Cruise’s approach to safety, the temporary shutdown of its AV operations by the California DMV provides an opportunity to address these issues and ensure safer autonomous driving experiences in the future.
1. How do autonomous vehicles currently detect pedestrians?
Autonomous vehicles use a combination of sensors, such as cameras, lidar, and radar, to detect pedestrians and other objects in their surroundings. These sensors collect data, which is then processed by the vehicle’s software to identify and track pedestrians.
2. Are children more difficult for autonomous vehicles to detect?
Children can pose challenges for autonomous vehicles due to their smaller size, unpredictable behavior, and potential obstruction by surrounding objects like strollers. These factors make it harder for AVs to accurately recognize and respond to children compared to adults.
3. How can AV technology improve the detection of children?
Improving the detection of children by AVs requires advancements in sensor technology, software algorithms, and machine learning techniques. Enhanced capabilities in object recognition and interpretation can help AVs better understand and respond to the unique characteristics and behaviors of children.
– The Intercept: [insert URL]
– Phil Koopman’s profile: [insert URL]