The dream of creating an artificial intelligence is one that has been fluttering in and out of humanity’s greatest minds for centuries. How would we create such a thing? What would it look like? What would it do?
San Diego-based tech startup TuSimple has brought those questions out of the realm of abstract dreams and into the real world, to solve one of modern society’s most vexing problems: how do we stop killing ourselves in car crashes?
Though we have completely integrated automobiles into our societies, building our cities and public spaces and even homes around their existence, it turns out that we just aren’t very good at driving them.
According to the Association for Safe International Road Travel, automobile accidents cause more than 1.3 million deaths around the world each year – an average of 3,287 deaths a day. That doesn’t include the 20-50 million others who are injured in accidents each year.
In the United States alone, 40,200 people died in motor vehicle accidents in 2016, with another 2.35 million injured. In addition to the human toll, ASIRT estimates the economic cost from crashes in the US to be $260.3 billion dollars.
But the most devastating statistic of all: 94% of all motor vehicle accidents are caused by human error.
TuSimple wants to change this. Fast.
Their mentality is simple: to remove the human error, remove the human. Focusing on trucks – a disproportionately deadly segment of crashes – TuSimple is developing an AI system to act as a safer driver.
Co-founder Xiaodi Hou has a PhD from Caltech in computation and neural networks. He has spent nearly every day of the past two years coding algorithms for a deep learning artificial intelligence system that will form the core of TuSimple’s self-driving truck.
TuSimple rose to fame last fall when its scientists took home ten first-place titles in computer vision benchmark competitions released by KITTI and Cityscapes.
Now it’s offering a challenge of its own.
TuSimple’s data scientists have created a massive dataset of 1-second videos designed to simulate real-world continuous driving conditions for a self-driving car. They are challenging any teams who want to test their AI driving systems to use their dataset benchmarks and see how accurately they can recognize lanes or analyze the speed of adjacent cars.
To encourage computer vision teams to test their algorithms, TuSimple is offering cash prizes of $1,000 for first place in the competition, $500 for second place, and $250 for third place. The winning teams will also receive special mention at this summer’s upcoming CVPR Conference in Honolulu, HI from July 21-26.
“Benchmark dataset competitions like KITTI, Cityscapes, and now ours foster innovation and creative solutions,” Dr. Hou explains. “We believe that development in the field of AI is a collective process, not a zero-sum game, and any growth in knowledge or processes improves the entire field of study.”