The Ben Franklin Racing Team and the DARPA Urban Challenge

Transcription

1 The Ben Franklin Racing Team and the DARPA Urban Challenge Jon Bohren, Jason Derenick, Tully Foote, Jim Keller, Alex Kushleyev, Daniel Lee, Brian Satterfield, John Spletzer, Alex Stewart, Paul Vernaza Department of Computer Science & Engineering P.C. Rossin College of Engineering and Applied Science

2 A Pressing Mandate In the 2001 National Defense Authorization Act for Fiscal Year 2001, Public Law , Congress mandated in Section 220 that: It shall be a goal of the Armed Forces to achieve the fielding of unmanned, remotely controlled technology such that by 2015, one-third of the operational ground combat vehicles are unmanned.

3 Additional Motivation Automobility for the Elderly Automated Transport & Retrieval System (ATRS) The age of 65 is projected to increase from 36 million in 2004 to over 70 million by 2030, this number will only increase Increasingly, these people live in suburban and rural areas where public transportation services are limited Loss of personal automobile = loss of independence

4 The Urban Challenge Summary November mile race in an urban setting 6 hour time limit $2M 1 st prize, $1M 2 nd, $500K 3 rd International Funding line for Track A teams 35 Semi-finalists; 11 finalists Final Result: CMU s Boss finishes 1 st (250 min) Stanford s Junior 2 nd (269 min) Virginia Tech s Odin 3 rd (277 min) BFRT Little Ben finishes 4 th in 305 min (est.)

5 The Ben Franklin Racing Team

6 The Big Three

7 Drive by Wire System Installed by Electronic Mobility Controls (EMC) Computer interface to car steering, brakes, shifting, lights, horn, turn signals, etc.

8 Sensors Velodyne LIDAR SICK LIDARs Hokuyo LIDARs Bumblebee stereo camera OxTS 3050 GPS/IMU Gigbait ethernet Interface

9 Ben s Computer Network 6 Mac Minis running Ubuntu Spread Toolkit for IPC monit for process monitoring

10 Now it s a Just Software Problem Ben runs on less than 5,000 lines of Matlab code Other teams had ,000 lines of code

11 LIDAR Signal Processing Ben relied almost entirely on LIDARs for exteroceptive sensing Two fundamental problems in sensing Reconstruction Data association LIDARs excel at the former, but outliers can compromise the latter Strategy: Spatial Consistency Temporal Filtering Beam-based calibration Reflectivity for lane markings Robust estimation strategies for road segmentation

12 Velodyne HDL-64E

13 Sick LMS291s for Road/Lane Segmentation 5m 7m

14 Iterative Re-weighted Least Squares Incorporates simple temporal filtering Assumes true model is slow changing w.r.t. time Computing g k = m k b k [ ] T More weight! Less weight! g k 1 Default Calibration Bias

15 LIDAR Based Road Classification

16 Sick LD-LRS (and Velodyne) for Vehicle Tracking No time or money for RADAR Obstacles grouped into local line features Line features tracked across multiple scans using a multihypothesis Kalman filter Binary filter based on upon prior size and velocity constraints

17 Vehicle Tracking Requirements Velodyne & LD-LRS could reliably detect moving vehicles at ~60 meters Sufficient sensor range to detect a vehicle at a relative speed of 60 mph with sufficient time to properly react and stop

18

19 Hokuyo URG-04LX LIDARs

20 Vision Based Lane Tracking

21 Planning & Path Follower Global Mission Planner computes optimal sequence of waypoints Local mode-specific planner computes detailed path in dynamic map Lane following Intersections Zones Path planner uses bicycle model to predict vehicle poses after a set of possible steering commands Choose the steering command that minimizes the cost function

22 The National Qualification Event Test Courses Three test areas Area A: Merging traffic Area B: Navigation, blocked roads, parking Area C: Intersection precedence, blockages, dynamic course

23 NQE Proving Grounds - El Mirage Dry Lakebed

24 Curbs are Hard to See (and just plain hard)

25 Sick LD-LRS Proves Useful in Merges NQE Course A signage (and speed limit LED display) occluded the Velodyne at times to oncoming traffic LD-LRS units on corners were able to detect RADAR may have been a better solution (?) * Course A Figure from Wikipedia

26 Still Adding Hardware!

27 UFE: Ben s out of the Chute!

28 Ben s Perspective

29 Not so Urban

30 Little Ben (Almost) Retired

31 The Highlight of the Race!

32 The Highlight of the Race!

33 Crossing the Finish Line!

34 No Prize Money, but

35 The Winnowing Process October 06: May 07: Aug 07: 26 Oct 07: 01 Nov 07: 03 Nov 07: 89 teams (78 Track B) 53 teams after video submission (42 Track B) 36 teams after DARPA site visits (26 Track B) 35 NQE semifinalists (26 Track B) 11 teams chosen as Finalists (4 Track B) 6 teams finish the UFE (1 Track B)

36 BFRT at Victorville Penn Jon Bohren Tully Foote Alex Kushleyev Dan Lee Alex Stewart Paul Vernaza Lehigh Jason Derenick John Spletzer Lockheed Heeton Choxi Drew Houston Brian Satterfield

37 Many Other People to Thank at Home! Allen Biddinger, Gilad Buchman, Kostas Daniilidis, Rich Fritz, Daniel Garofalo, Chao Gao, Erika Gross, Ani Hsieh, Steve Jamison, Jim Keller, Vijay Kumar, Bob Lightner, Gerry Mayer, Tom Miller, George Pappas, Ray Quinn, Ellen Solvibile, CJ Taylor, Chris Wojciechowski and many others CSE398/ Spring 2007 Class LU Police Department Mitch Herbets, President/CEO, Thales Communications

38 Thanks for Listening!