This video demonstrates the Torc self-driving car smoothly navigating a tunnel—entering and exiting a multi-lane tunnel in bright daylight.
Navigating tunnels presents unique challenges for self-driving cars due to the sharply contrasting lighting outside and inside the tunnel. Additionally, GPS has limitations in position accuracy, and requires a clear view of the sky to be effective.
The Torc self-driving car relies on a fusion of multiple sensors and cameras to gather a 360 degree view of its surroundings. We run the cameras at a higher frame rate, interleave multiple exposure settings, and combine the frames into an image with much higher dynamic range. This helps the vehicles recognize objects in the mixed lighting conditions found on the road, such as inside a dark tunnel and outside in the bright sun.
Additionally, our localization system uses information about the car’s position beyond GPS, so the autonomous system maintains its place on the road without a clear view of the sky. We rely on inertial and vehicle sensors to dynamically estimate how the vehicle moves, and then refine this estimate by matching what the vehicle sees to features stored in our map. This allows the vehicles to maintain a positional accuracy better than 4 inches, whether or not we have GPS.
We’re showcasing the capabilities of our autonomous car system on real public roads with real scenarios the car has encountered. To get updates on the latest videos, check back on our blog or subscribe to our YouTube channel.