|an autonomous car/ Photo By cheskyw via 123RF|
Makers of self-driving cars face one of the biggest challenges in autonomous driving: the unpredictable behavior of pedestrians. But a startup company, Perceptive Automata, managed to create a solution to this problem.
Man's natural ability in understanding how people behave and react can be difficult to incorporate in software. So, developers at Perceptive Automata used a unique way to train autonomous software in understanding how pedestrians, cyclists, and other people move around the car.
Compared to traditional machine-learning methods which are trained with data measured objectively, the startup company utilized algorithms based on the subjective judgment of other people. In this way, the software can learn from the point-of-view of actual human beings.
To do this, the developers asked people to watch some video clips and asked to label every pedestrian they see. The subjective angle here was that each viewer judged the video at the best of their ability to notice each pedestrian who tried to cross the street and who might not notice an oncoming vehicle. After the test, the engineers collected the dataset of labeled videos and applied it to train machine-learning algorithms.
"We're building a module that allows autonomous vehicles to understand the state of mind of humans out on the road… Give autonomous vehicles the ability to look at a person and say, in a human-like way, 'This person wants to cross the road, this person knows that my car is there,'" said Sam Anthony, co-founder of the company, as quoted by ARS Technica.
The company’s software module could be purchased by self-driving car makers and have it added to the existing database. Anthony said that some automakers might outsource such module to ease the numerous tasks of engineers in self-driving software development.
Moreover, the module offers a different predictive model compared to the usual one deployed by self-driving car companies. This is because pedestrians will always be unpredictable, some can simply stand at the side of the road and wait for several seconds before crossing the street.