The venture capital firm that backed Facebook and Lyft is making its first investment in an autonomous trucking startup that was born of artificial intelligence research from the Massachusetts Institute of Technology.
Founders Fund is leading a $15 million investment in iSee. The fund, whose general partners include Paypal co-founder Peter Thiel, has also invested in Flexport and PostMates.
“ISee’s approach to autonomy gives them structural advantages over competitors,” said Scott Nolan, a general partner at Founders Fund. “By training vehicles to think like human drivers, rather than only follow narrow rules or mimic an infinite long tail of edge cases, iSee has achieved industry-leading autonomous trucking performance in just two years.”
Based in Cambridge, Massachusetts, iSee was founded by Yibiao Zhao, Debby Yu and Chris Baker in 2017. Zhao and Baker were doctoral students under Josh Tenenbaum, a professor of cognitive science and artificial intelligence at MIT.
In a 2018 speech at MIT’s EmTech conference, Tenenbaum said his research attempts to “reverse engineer” human intelligence for use in autonomous vehicles and other applications. He said the current state of the art in AI focuses mostly on pattern recognition. While that allows a host of digital applications to take over tasks once relegated to humans, the machines that use pattern recognition solely can perform only those specific tasks, and have no ability to learn about dynamic situations such as highways.
Tenenbaum’s research focuses on how children learn from visual or other cues and attempts to reverse engineer those processes to build a machine that “grows into intelligence the way a person does, that starts like a baby and learns like a child.”
“If you can capture basic learning in simple math and scale it up to Silicon Valley scale, it can change the world,” he said.
Tenenbaum, who also serves as an adviser to iSee, said of his former students’ work that “they are trying to take the basic insights about how we understand other people’s actions.”
“Even to do something as simple as changing lanes and entering and exiting the highway, that’s a serious challenge, and they are trying to take those insights from reverse engineering human intelligence and use that to make self-driving car technology that works better and is safer than what we have today.”
ISee describes its approach to artificial intelligence as “Multi-Agent Tensor Fusion,” which takes apart visual data from the background scene and each individual agent moving on that scene. ISee’s probability models allow autonomous vehicles to react to given situations based on how humans would react based on similar situations.
“Human drivers use their social intelligence to predict how others’ future motions will depend on interactions with themselves, the surrounding agents, and the scene context,” iSee said.
A paper co-authored by Baker, Tenenbaum and MIT researchers Julian Jara-Ettinger and Rebecca Saxe provides an example of how machines can be taught to read scenes and infer outcomes; in this case, what kind of food truck does a person prefer based just on visual cues.
In the scenario, a student goes out for lunch to one of three preferred food trucks that come to their campus. But the campus only provides two parking spots, each on opposite sides of a building.
The student walks toward the one truck they can see upon exiting their side of the building, walks past that first truck to look down the other side of the building at the second food truck, then walks back toward the first truck.
The mathematical model developed by the researchers parsed the scene fairly close to how human observers did: the student most likely wanted the third food truck that was unable to park, followed by the first food truck.
The company received $2.7 million in seed funding led by MIT’s venture capital incubator, The Engine, according to PitchBook. One of its trial projects was retrofitting a Lexus SUV owned by co-founder Yu to drive through Boston autonomously.
ISee joins the increasingly crowded market for autonomous trucking startups that includes PlusAI, which came out of Stanford University, Perrone Robotics and TuSimple. Approximately 200 trucks with some degree of automation are currently undergoing testing on U.S. roads, according to one estimate.
This article has been amended to reflect that Scott Nolan is the partner at Founders Fund.