Jimi van der Woning. 30 November 2010

Transcription

1 Jimi van der Woning 30 November 2010

2 The importance of robotic cars DARPA Hardware Software Path planning Google Car Where are we now? Future Jimi van der Woning 2/17

3 Currently over 800 million cars in the world. 87% of the working population travels by car, thereby consuming 34% of the nation s energy. 42,000 traffic deaths in 6 million accidents per year in the US. 3.7 billion hours are wasted due to traffic jams. 8.7 billion litres of fuel are wasted in traffic jams. Cars are only used 4% of their lifetime Jimi van der Woning 3/17

4 2004 Grand challenge 142 miles through the Mojave desert. None of the participants finished Grand challenge 132 miles through flats, dry lake beds and mountain passes. Out of 23 finalists, four cars reached the finish. Winner: Stanford University s Stanley Jimi van der Woning 4/17

5 2006 Urban Challenge 16 miles through an urban environment with 11 robotic and about 35 human controlled cars. Cars had to follow the Californian traffic rules. Six robotic cars finished. Winner: Carnegie Mellon University s Boss Jimi van der Woning 5/17

6 Any type of car can be used. Custom controllers for throttle, brake, steering, etc. A Linux based computer system, two quad cores. Some radars and laser range finders. A camera. Some GPS antennas Jimi van der Woning 6/17

7 Three main areas: Perception Planning Control Modular programmed, with asynchronously running modules. Maximizes flexibility Minimizes reaction time to about 300 ms Jimi van der Woning 7/17

8 Laser data is integrated into a 3D point cloud. To prevent noise, the presence of an obstacle is a problablistic test Jimi van der Woning 8/17

9 Lasers range is only 26 meters, increased with adaptive vision: Small part of drivable area is extracted with data from lasers Rest of the image is scanned for similar color distributions Done ten times a second, increasing the obstacle detection range to 200 meters and maximum speed to 40 mph Jimi van der Woning 9/17

10 Moving objects are tracked using temporal differencing: Uses the differences between two subsequent scans. Further tracking is provided by radar sensors Only when trying to merge into moving traffic Jimi van der Woning 10/17

11 Global path selection Expected time to target is calculated from any point. Local path selection Uses a modified A* algorithm. Finds the shortest path using search trees. Performed on the fly, in less than a second Jimi van der Woning 11/17

12 Developed by some of DARPA s best engineers. Six Toyota Priuses and an Audi TT. Driven 230,000 km semi- and 1,600 km fully-autonomous. Crossed Golden Gate Bridge Safely traversed San Francisco's Lombard Street One accident: the car was hit from behind Jimi van der Woning 12/17

13 Jimi van der Woning 13/17

14 What s good? Control What needs improvement? Perception Reacting to low-frequency change Reliability Comfort Jimi van der Woning 14/17

15 Initially, robotic cars will take over highways. Later, robots will provide door-to-door transport. Unoccupied cars will pick up people. Cars will be shared a lot more. There will be less or no traffic jams Jimi van der Woning 15/17

16 Developed by General Motors and Segway. Current prototypes can be driven manually. Will be made autonomous in the (near) future. Should be finished by Jimi van der Woning 16/17

17 Thrun, S. (2010), Toward Robotic Cars, Communications of the ACM, 53:4, Thrun, S. (2010), What we re driving at, The Official Google Blog, [visited 28 November 2010] Markoff, J. (2010), Smarter than you think: Google cars drive themselves, in traffic, The New York Times, [visited 28 November 2010] Siegler, M.G. (2010), Google Has A Secret Fleet Of Automated Toyota Priuses; 140,000 Miles Logged So Far, TechCrunch: GreenTech, [visited 28 November 2010] Motavalli, J. (2010), G.M. EN-V: Sharpening the Focus of Future Urban Mobility, The New York Times, [visited 28 November 2010] Jimi van der Woning 17/17