According to the World Health Organization, human error causes more than 90 percent of traffic accidents across the globe. Some 1.25 million people die in crashes, and 20 to 50 million people are injured each year. Automakers are scrambling to develop cars with advanced driver-assistance systems that are so good they’re “crash-proof,” or fully autonomous vehicles, each safer than the next. But the automakers are somewhat limited by the computational resources onboard their vehicles.
Now, a startup out of Mountain View, Calif. called DeepScale has raised $3 million in seed funding to help automakers use industry-standard low-wattage processors to power more accurate perception. Alongside sensors, mapping, planning and control systems, perception, (sometimes referenced as “computer vision”) enables vehicles to make sense of what’s going on around them in real time.
DeepScale co-founder and CEO Forrest Iandola previously attained a doctorate at UC Berkeley working on deep neural nets and computer vision systems. His work there focused on, among other things, making AI systems learn to do something new in just a day instead of months of training. Iandola said, “Lots of deep neural network software out there is tied to processors like Nvidia’s. They make really good libraries for running neural net stuff quickly. We built our own libraries for fast, deep neural net computation on a variety of affordable processors, Qualcomm, Intel and the like. We’re using a lot of what we learned in academia on the commercial side now.”
While he did not have permission to name the companies, he said DeepScale is already working on pilot projects with some major original equipment manufacturers in automotive. With the market for automotive systems, data and services forecast to reach $70 billion in revenue annually by 2030, DeepScale is focused strictly on automotive for now. But, the CEO said, the technology could be adapted over the long term for use with unmanned drones or robotics, which also need to accurately understand and respond to changes in their environment.
DeepScale’s seed investors included: Bessemer Venture Partners, Greylock, Auto Tech Ventures, Andy Bechtolsheim (who was the first investor in Google) and Jerry Yang. A partner with BVP, Alex Ferrara, said, “Cars are moving from systems today where they have a large number of small computers, called ECUs, in them, to working with smaller more powerful computers for perception. But you have all these little sensors, lidar, radar, ultrasound, and each one brings its own view of the world. There’s a really interesting opportunity here for DeepScale to pull everything together and use info from all those sensors to make computer vision accurate and efficient.”
DeepScale is competing for a share of this burgeoning market versus some 800-lb. gorillas in automotive tech, like Mobileye, now owned by Intel, or Bosch, but also other funded startups like Comma.ai, Argo and Drive.ai, which are trying another approach of building their own, fully autonomous vehicles or retrofit systems.