What does DeepScale do?
DeepScale develops perception software for vehicles that takes sensor data from the surrounding area as input and determines where is everything, what is everything called, and how it is moving.
What is this used for?
Our primary market is the automotive industry. Highly-accurate perception software is a key requirement for assisted and autonomous driving. In Fall 2017, automakers in Europe are integrating our perception software into prototype vehicles for semi-automated and fully-automated driving.
Where does the software run?
It is increasingly common for mass-produced cars to have a central processing platform called an “ADAS Domain Controller.” Companies including NVIDIA, NXP, Qualcomm, Intel, and Renesas design these processors. These processors are integrated into automotive-grade hardware modules by suppliers including Visteon and ZF TRW. Our software runs on a car’s ADAS Domain Controller.
How do I use the product?
DeepScale is a deep technical play, and our software will be deeply embedded in vehicles that will begin selling around 2020, as well as autonomous taxis that will launch in the early 2020s. You will be using our product if you drive or ride in one of these vehicles.
What’s the business model?
We license software to automakers and automotive suppliers. Each deal is a little different, but we sell the same core technology to each customer. Each customer deal can be worth 1 million cars or more.
What’s the point of all this?
Long term, autonomous vehicles will save human time and reduce congestion. Our core objective is to reduce our planet’s 1 million traffic deaths per year to zero. Our software contributes to all of this.
Why does DeepScale have an advantage in this area?
To win this market, you have to do a bunch of things well, such as: develop custom deep neural nets which achieve high accuracy, integrate data from several sensors, and run efficiently on embedded processors. We are the right people in the right place at the right time; our core team is comprised of experts in all of these areas, who have published papers such as SqueezeNet and Boda.
What’s unique about DeepScale’s approach?
Autonomous cars have many sensors. Traditional approaches use different sensors to solve different perception problems – for example, LIDAR data to find the objects, and Camera data to classify the objects. However, our solutions integrate all the sensors to solve each perception problem, yielding higher accuracy and safety.