Self-driving cars are repeatedly tested to avoid collisions with pedestrians and bicycles that suddenly jump out
When driving a car, a car or a human suddenly jumped out on the course, and there are many people who have suddenly braked or turned the steering wheel. Waymo, an autonomous vehicle development company under the same Alhabet umbrella as Google, is a test method ' Collision Avoidance ' for evaluating collision avoidance ability against potential dangers such as humans running ignoring red lights and cars suddenly changing lanes. Testing (CAT) ' is explained.
Collision Avoidance Testing of the Waymo Automated Driving System - Collision Avoidance Testing of the Waymo Automated Driving System.pdf
(PDF file) https://storage.googleapis.com/waymo-uploads/files/documents/safety/Collision%20Avoidance%20Testing%20of%20the%20Waymo%20Automated%20Driving%20System.pdf
Waypoint - The official Waymo blog: Waymo's Collision Avoidance Testing: Evaluating our Driver's Ability to Avoid Crashes Compared to Humans
Fully autonomous systems with fully automated driving are much more thoroughly tested than typical driver assistance systems, as they must handle all driving tasks without a human in the driver's seat. Waymo seems to use a method called 'scenario-based test' that combines virtual driving, test driving, and actual driving.
Waymo simulates various data such as existing driving data obtained from past experience, police accident database, collision accident data recorded by drive recorder, area and road conditions where Waymo's self-driving car drives. put into. As a result, the test reproduces common situations that can occur almost anywhere, such as ``a car protruding from the roadway'' and ``a pedestrian crossing the street ignoring the traffic light'' Based on this, he seems to give feedback to himself while taking this test many times.
Below is where multiple pedestrians are walking right next to Waymo's automatic driving car in San Francisco. A pedestrian suddenly jumps out of a group of such pedestrians to cross.
Below is a CAT reproduction of a pedestrian actually jumping out of a place similar to the above situation.
Waymo has created a number of test scenarios that require emergency avoidance operations to avoid collisions while changing the position and speed of humans in order to respond to pedestrians and bicycles that jump out of the blind spot. By clicking on the thumbnail below, you can see the GIF animation of Waymo's experiment that slows down to avoid a bicycle that suddenly jumps out of the blind spot.
According to Waymo, Waymo's automatic driving system showed a higher collision avoidance ability than the results of manual driving by humans. In the test conducted in September 2022, a virtual human named ``non-impaired, with eyes always on the conflict: NIEON'', assuming that he is always alert and does not get tired. When the driver was driving, the collision perfect avoidance rate was 62.5%, while Waymo's automatic driving system's automatic driving system recorded 75%.
Waymo said, “This CAT is very important in the safety assessment process as we continue to bring Waymo’s autonomous driving to more people in more places. We will continue to share more information about our testing and safety evaluation methods so that people and drivers can better understand our approach to safety.'