Tesla has so far been mostly alone in the autonomous driving world trying to achieve fully self-driving capability using a camera-based system. That’s because almost all other company developing self-driving technology are focusing on lidar sensors.
It’s about to change. A new startup in the field is now trying a similar albeit even more radical approach to self-driving with a camera-only system.
AutoX, which was created by Dr. Jianxiong Xiao, a computer vision professor at Princeton, recently obtained a self-driving test permit from the DMV in California and posted last week a first video demonstration of its prototype.
While Tesla’s new Autopilot program aimed at achieving fully self-driving capability is making extensive use of computer vision through its 8 cameras, it also uses a forward-facing radar, 360-degree ultrasonic coverage, and high-precision GPS data.
The lack of lidar sensors is what differentiate them from most other self-driving programs, but Xiao’s AutoX is taking it one step further and doesn’t use any radar, sonar or GPS – only cameras.
Of course, the biggest advantages are hardware cost and integration. Cameras are much cheaper than radar antennas and lidar sensors. The disadvantage is that they are collecting less data on which their system can make decisions on.
For their current prototype, Xiao told Business Insider that they simply bought a few $50 cameras from Best Buy and strapped them to a 2017 Lincoln MKZ. That’s their first prototype hardware suite to develop their computer vision and self-driving software.
6 months later, they made enough progress in image processing and autonomous driving decision-making to come up with a decent prototype to drive on public roads.
The result is fairly impressive in this new video demonstration:
Again, while it’s more radical than Tesla’s approach, it’s interesting that Tesla is not alone investing in camera-based systems for self-driving.
AutoX is currently about 20 people, most of them computer vision experts hired from companies like Apple, Magic Leap, and Microsoft.
Xiao says that he doesn’t plan to build a vehicle from the ground up, but he instead plans to work with automakers to license self-driving software suites based on their computer vision tech, à la Mobileye, which was just acquired by Intel for $15 billion.
He didn’t offer any timeline for the commercialization of AutoX’s technology.
The competition in the sector is increasing almost daily. As we recently reported, there are now 26 companies in California alone that are testing self-driving cars on public roads.