This autonomous trucking startup out of MIT wants computers to think like human drivers, and it just became the first self-driving truck investment for Founders Fund
- iSee, an autonomous driving startup for trucks, announced $15 million in Series A funding from Founders Fund on Monday.
- Unlike other autonomous driving and logistics startups, iSee uses proprietary deep learning and computer vision technology to try to give its trucks "common sense," an area that other technologies struggle to replicate.
- The startup is Founders Fund's first investment in the red-hot logistics sector. Still, partner Scott Nolan told Business Insider that the firm is investing in the iSee team and its technology specifically, not the larger trend.
- The team is comprised of a notable computer research lab under cognitive science professor Josh Tenenbaum at MIT. Maintaining a high bar for new hires is one of the key challenges facing the team of former academics, Nolan said.
- Visit Business Insider's homepage for more stories.
There's a new autonomous driving startup on the block, and it just brought on one of the most sought-after investors in Silicon Valley.
That startup, iSee, announced $15 million in Series A funding from Founders Fund on Monday. Although Founders Fund has backed a self-driving startup in the past, this is its first foray into the red hot automated logistics and shipping industry, given iSee's focus on the trucking industry.
"The other self driving companies out there were operating with the same tech stack and weren't differentiated at all. They are all using the same sensors, the same data sets, and the same approaches to collecting that data," Founders Fund partner Scott Nolan told Business Insider.
The main challenge with other self-driving technologies, according to Nolan and iSee cofounder Yibiao Zhao, is that there are infinite edge cases - or statistically improbable situations - to be discovered in trucking, with more cropping up every day.
Even larger players like Uber struggle to correctly classify objects in the road with self-driving tech, and according to a recent report, has difficulty accounting for human unpredictability. It's the billion-dollar challenge facing the industry, which is why Nolan wasn't satisfied with the scores of self-driving startups knocking down his door previously.
"The hardest thing to do is navigate other drivers because you have to think about those around you. So why not train the computer to drive like a human?" Nolan said.
Unlike other autonomous driving and logistics startups, iSee aims to use proprietary deep learning and computer vision technology to gives its trucks "common sense," an notion that other technologies struggle to replicate. Zhao and his team of academics hail from a notable computer research lab under cognitive science professor Josh Tenenbaum at MIT, and Zhao himself got his Ph.D in computer vision from the University of California, Los Angeles.
Perhaps one day, all cars will be autonomous. Until then, Zhao said, tech like that of iSee can help make these vehicles safer for the real humans who populate the road.
"We realized that driving is actually a social activity," Zhao told Business Insider. "Autonomous vehicles have to drive in open environments populated with human drivers and pedestrians. If we completely remove humans, the autonomous vehicle problem is solved."
Robots in the real world
The other constraint facing many robotics and machine learning startups is the need to set controls within any given environment. Until humans are entirely out of the picture, operating robots in an entirely closed environment is unrealistic, Zhao explained.
"The biggest challenge is to close the gap between research and engineers, because we consider ourselves an AI company but we are doing a different type of AI," Zhao said. He said that firms like Google's DeepMind and the OpenAI think tank do good work, "but little of that can be applied to real work machines and real robots."
The biggest test of a robot's ability to operate in a human-dominated environment is whether it can explain to its handlers why it reacted a certain way. For iSee's trucks, that means being able to see a car driving erratically and understand that the driver may be aggressive so the truck needs to allow the car to pass. Other autonomous vehicle startups, according to Nolan, fail this test because they are just replicating what the car in front of them is doing.
"It's a theory-of-mind approach because they are modeling out the thinking and intent of those around them," Nolan said. "It's a very elegant solution, but it does take more computing power on board the vehicles because they need to project out in real time what they are doing and others' intent. It works well in trucking because you can fit the computing in the larger vehicle."
Growing beyond the lab
Even for a capital-intensive business like autonomous trucking, iSee has managed to keep a small team. According to LinkedIn, it has only 20 employees even though it was founded more than 3 years ago. The fresh funds will help the team grow beyond its academic roots, but that will present another challenge for the company, Nolan said.
"One of the things they are going to be doing over the next year and two years is hiring the core team, and now it's about expanding with A-plus talent," Nolan said. "You have to make sure your hiring bar stays extremely high."
iSee cofounder Debbie Yu explained that the Cambridge company's proximity to top-tier research institutions and his connections to MIT will help, but there is a knowledge gap among engineers that have worked at previous self-driving startups. He is looking for a team that is open to approaching the problem in a fundamentally new way, she said.
"Smart people want to work with smart people," Yu said. "To tackle the most advanced and difficult problems in the field, we need people who like the challenge and see unique advantage of our approach."