BostInno
This Boston Startup Is Teaching Self-Driving Cars to Think Like Humans
By Srividya Kalyanaraman
Collision costs resulting from car accidents totaled $5.5 billion in 2015 in Massachusetts. Self-driving cars will reduce this number to $3.4 billion by 2050, says a fiscal study released in July by Conservation Law Foundation, an environmental advocacy organization based in Boston.
The report predicts that autonomous vehicle entering the roads will increase congestion in the Boston area by as much as 17% but overall, reduce collisions and traffic injuries to an extent that safety benefits from self-driving cars will be “considerable.” Why? Because 90% of collisions take place due to human errors.
While human error may result in the occasional collision, it can be argued that discerning human judgment prevents more accidents than it causes. And so, here is one Boston startup that’s on a mission to teach self-driving cars to perceive, distinguish and judge like a human.
Perceptive Automata, which emerged out of stealth mode in July with a $3 million seed round from First Round Capital and Slow Ventures, claims to solve the hardest part for autonomous vehicles — understanding humans. “Humans are good at gauging intention and making a judgment,” said Sid Misra, CEO, and co-founder of the company. “We expect everyone on the road to do that, but cars don’t..not yet.”
Backed by a team of neuroscientists and computer vision experts, Perceptive Automata builds software that teaches autonomous vehicles to watch humans as humans do. How does the team do this? By capturing streams of data on frame, which in this case is pedestrians on the street, and formulate real-time responses. What this means is the software will look at pedestrians’ posture and their facial expressions to anticipate what they will do next. Based on these inputs, the software will perceive and judge like a human and prompt an appropriate response in the situation. This, is especially crucial in states like Massachusetts where crosswalk courtesy is a law.
The law states, “If a motorist is approaching a crosswalk, pedestrians have the right of way if they are in the path of a motorist or if they are within 10 feet of the halfway point in the road. Motorists are not allowed to pass a vehicle that has yielded the right of way to a pedestrian, nor are they to block a crosswalk.” Laws like these will require ‘presence of mind’ from a driver to make judgments and predict actions of pedestrians on the street.
Perceptive Automata CTO Sam Anthony is a vision scientist who studies human predictive behavior. And one of the co-founders, David Cox, is a vision science cum computer science expert who puts this research into action (in a software) with machine learning tools.
The technology benefits more than just self-driving cars and can be applied to robots, autonomous forklifts, low-flying drones; basically, any machine that interacts or is in the vicinity of humans. For now, Misra said that the company is focused on furthering partnership talks with manufacturers to help get the software installed in cars, as well as looking into deploying the technology in autonomous shuttles.
