Finding Multiple Lanes with Vision and Lidar

Authors: Albert Huang.

Time: 14:20-14:45

To successfully utilize the millions of kilometers of existing road networks around the world, an autonomous land vehicle must have a wide range of perceptual capabilities. One of these is the ability to accurately and reliably estimate the shape and geometry of the roadway and its travel lanes. To do so, the system must utilize information from its on-board sensors, and any other a priori information it has available (e.g. from a road map).

In this talk, I will describe the vision and lidar based lane estimation system developed at MIT for the 2007 DARPA Urban Challenge. Our system was incorporated into a closed-loop controller to successfully guide an autonomous vehicle through a 90km course at speeds up to 40 km/h amidst moving traffic. I will discuss its successes, its limitations, and the ongoing research we have conducted since the Urban Challenge to improve the reliability and accuracy of our system.