New Startup Seeks to Help Self-Driving Cars Understand Humans
A new approach
The company is experimenting with using the subjective judgment of other humans to train algorithms. Co-founder Sam Anthony articulates the goal of Perceptive Automata’s technology. The new software will “give autonomous vehicles the ability to look at a person and say, […] ‘This person wants to cross the road, this person knows that my car is there.'”
Reputable and Family-Friendly: Discover the 2018 Buick Regal
How it works
Perceptive Automata enlists human volunteers to watch video clips. During the footage, they must label the pedestrians they see. They must also guess whether or not each pedestrian noticed the car and whether certain pedestrians were about to cross the street. Next, the company’s team uses the labeled videos as a dataset to train machine-learning algorithms. These algorithms trigger automated vehicles’ responses on the road.
The challenge, of course, is how to get an inanimate technology to have a human-like awareness and response for making these driving decisions. And, what happens if a pedestrian lingers on the side of the road instead of crossing it right away? There’s a high risk of confusing scenarios that could puzzle the AV’s technology. This, in turn, could result in a delayed and/or inaccurate response from the car.
Despite these obstacles, Perceptive Automata has high hopes for the technology. If all goes as planned, the software module will be available for self-driving automakers to incorporate into autonomy stacks. We anticipate more news on this exciting development, as the startup forges software to help self-driving cars understand humans.
Financial Lingo: Get the lowdown on gap insurance
News Source: Ars Technica