Today I listened to a talk by
Sebastian Thrun from Stanford University and Google on When will we get our robotic car? Abstract of the talk can be found
here. Sebastian talked about the CS project that he has in AI at Stanford that deals with drive-by-wire, which means cars that automatically drive by themselves. His research work uses Volkswagen Toureg cars which are equipped with a laser system for automatic steering. The car can detect obstacles through the shining of light, and do laser terrain mapping, driving in a laser map, and online motion planning by predicting future motion. They applied these techniques and other AI algorithms to creating a robotic car and formed the
Stanford Racing team to compete in the 2005 DARPA racing challenge which required maneuvering a car automatically through desert-style terrain.
They explored navigation problems using AI, false positives are not good, and they try to solve hard problems like if the car goes off the road around the obstacles. They use statistics and probabilistic error models with Markov chains. How to determine if an obstacle is nearby, they do a trigger if the difference between Zf and Zi is at least 15 cm. The car that they entered into the 2005 DARPA racing challenge was called Stanley. Stanley's adaptive vision was designed with finding the road with computer vision. This involved how to find roads in the desert and they tested their method in a desert in Arizona. Stanford was selected as a semi-finalist in the 2005 DARPA racing challenge. Sebastian Thrun also showed how some competitors didn't fare so well in videos, which was pretty hilarious by how the audience laughed. While the cars were running the course, the teams were in a tent looking on a web page to see where their robot cars were. In the end and all the hard work, Stanford won the racing challenge.
Stanford then entered the next racing challenge which was The Urban Challenge in November 2007 (last year) and the objective was to navigate the car in an urban setting with car traffic and human traffic. For this, Stanford created Junior and used surround laser technology. Thrun showed a video of how Junior merged into traffic (using a laser view) when there were cars in both directions on the roads. In order to determine where the car will move based on cars in front, Junior uses a hybrid A* search tree for planning to navigate. They did lots of simulation. Finally, the race involved a dirt road, passing disabled robot cars, and parking. Stanford came in second place while Carnegie Mellon came in first, however Junior was the first robot car to return. Apparently, Junior lost some time compared to the Carnegie Mellon team.
Thrun also showed an example of the next part of the research which was high speed driving where he showed the car zooming right through traffic. What are the implications of robotic cars? Thrun mentioned the huge cost of human driving and commuting, saying how many lives are lost due to driving. So he feels that robotic cars can be used to help save lives and make us more productive than having to waste time commuting. During the question period, someone mentioned that the robotic car is a way of trying to solve the commuting and traffic problem, however it doesn't really kind of solve this problem because you still have cars on the road. He mentioned that we should rethink the commuting problem instead of replacing the human, and asked if there were alternative ways. This made me think about the Segway, which is the human transporter vehicle created by famed inventor Dean Kamen. Also, Thrun mentioned about the need for clean technologies and alternative fuels.
So will we see any robotic cars any time soon? Thrun mentioned that even though the technological problems are largely solved, the main thing is political and social. He feels it will be probably 20 years out before we see self-driving cars. Since these self-driving cars will obey the rules of the road, does this mean we will see very few accidents and there won't be any need for police to monitor the roadways? Overall, this was an interesting talk and shows the practical applications of computer science research.
On Technorati:
U of T,
robotic car,
Stanford
