A description of the race and early results of the DARPA robot car contest from Scott Ettinger.Today was the day of the DARPA Urban Challenge final event and people turned out in force to watch high-tech cars drive themselves through city streets for a two million dollar prize (for a full description of the Urban challenge, please see my previous posts). I knew DARPA would put together a well organized event, but I have to admit it was better than I had expected – complete with stabilized helicopter video coverage and audio commentary from Jamie and Grant of Discovery Channel’s show “Mythbusters” shown on three giant projection screens in the main tent. Impressively, this coverage could also be viewed on the internet via webcast for free. The production was great (like major TV network sporting event coverage but without commercials) and it allowed spectators to get involved in the event which would be difficult to observe without the aerial coverage. [Photo 1: DARPA’s giant screens in the main tent] Eleven teams survived the qualification rounds held earlier this week. The top three qualifiers were Carnegie Mellon University, Virginia Tech University, and Stanford University. The rest of the qualifiers can be found on DARPA’s web page. I have been lucky enough to be working with the Stanford Team to help develop this very futuristic technology. Everyone on the team was a bit nervous since this is the first event of its kind. The race began with the top three qualifiers lined up in starting gates in front of the main tent. From here, DARPA hands each of the teams a USB flash drive containing the mission that each vehicle must follow. Each team loads the data into their vehicle and then sends it out to drive off on its own out onto the course. This is the last interaction that a human will have with the vehicle until it returns from completing its mission. To finish the race, the vehicles must complete approximately 60 miles of driving through the course which contains urban terrain, parking lots, and dirt roads – all of which are swarming with human driven traffic vehicles – and other robots. To reiterate, these cars are not remote controlled. Computers on-board the vehicle make all decisions and perform all actions. Carnegie Mellon was scheduled to start first, but they had a technical problem, so Virginia Tech sent their vehicle out on the course first followed by Stanford. Junior (Stanford’s vehicle) drove out onto the course and disappeared around the corner into the open course. Carnegie Mellon got their vehicle started soon after along with the remaining vehicles. All eleven finalists were now on the course together. [Photo 3: Stanford’s vehicle “Junior” navigates the course] DARPA had a large staff of drivers on the course along with the robots. Most were traffic vehicles which the robots must interact with properly – following all traffic rules in the California driver’s handbook. Each robot also has a chase car that follows behind it to monitor it for any scoring penalties. The chase car also has a button called an E-stop that can disable the robot in case it gets into a dangerous situation. Some of them would be used during the race. Remarkably, every human driven vehicle on the course has a video camera sending video back to the DARPA command center at the event. DARPA gave us a quick look at the command center in their video coverage and it did look like a NASA mission control center – video screens everywhere. The race was broken up into 3 distinct missions that each vehicle has to complete with each of those missions broken up into 6 sub-missions. At the end of each mission the vehicle returns to the starting gates to be loaded with the next mission. Since the course is a very large area and it is difficult to see much of the action from the few spectator areas, we headed into the main tent to watch the video coverage. DARPA not only had the aerial video, but they also had a number of cameras set up on the ground around the course so they could catch most of the action as it happened. During the first set of missions, all eleven competitors were out on the course, but the field quickly thinned. Two teams actually had run-ins with buildings on the side of the road and were removed from the course (both were E-stopped before any major damage occurred). Another team ran off the side of the dirt section and got stuck in the loose dirt lining the road. Others simply got into situations where they stopped and would not move. There were only 6 teams that were able to survive the first mission. [Photo 3: Carnegie Mellon’s vehicle “Boss”] Virginia Tech was the first robot to return to the starting gate to complete the first mission. Junior was next to return. Many of the robots had to be paused due to other robots having problems in front of them. The time while the robots are paused does not count against their total time. During the second missions, it became clear that the top three qualifiers were more capable than the other three remaining teams. While the other three robots were able to make their way through the course, they made frequent stops and took long pauses at intersections. Perhaps the most intense action of the day happened when two of the slower robots actually collided with each other – every team’s worst nightmare. Remarkably both teams were able to continue and in fact finish the race. Throughout the race DARPA’s coverage would periodically update the status of which sub-mission the vehicles were currently working on. Towards the end of the second mission, Virginia Tech was out ahead followed by Carnegie Mellon. We knew that Junior had been paused for a long time behind the collision between the other two vehicles, so we were not concerned that Junior would be coming in later. Through the third mission, we did not see a lot of Junior because of incidents with other teams attracting the coverage. Virginia Tech looked to be doing very well until their vehicle unexpectedly drove over a curb and waited for a time before correcting itself. Watching the race video, Junior’s driving looked consistently very smooth and well centered within the lane. Software written by Jesse Levinson of the Stanford team uses the reflection intensity from the laser scans to correct any GPS shift and accurately position Junior squarely in the center of the lane. [Photo 4: Virginia Tech’s vehicle “Odin”] As time ran on into the third mission, DARPA unexpectedly announced that Junior was on the final sub-mission of the race. We had assumed that Junior was still behind the other teams. A few minutes later, they announced that both Junior and the Carnegie Mellon vehicle were heading for the finish. A large crowd gathered outside in the bleachers at the finish line. It was a tense 5 minutes while the crowd waited for a robot to turn the corner into view. After a few false alarms from traffic vehicles passing by, finally a spinning laser could be seen in the distance coming around the corner. As it did, you could clearly make out the the Stanford “S” on the windshield. Junior crossed the finish line first. Carnegie Mellon finished only moments later followed by Virginia Tech. The other three teams that survived the first mission all completed the course as well. Now, however, we have to wait. The robot that crosses the line first is not necessarily the winner. First, there are the pause times to factor out. Then there are the scoring penalties. DARPA has not defined exactly how they will score the race. From what we gather, things look good for Junior but nothing is certain. DARPA will announce the winners tomorrow morning. In talking with the team, we all agree that regardless of the scoring outcome, we are happy with today’s performance. We got the vehicle back in one piece and made a step forward in research. For the six teams that finished the course, this is a huge feat of research and engineering. To watch vehicles navigating the course through traffic with no driver inside is something to see.
Connect With Us
- nhat phat on How do you package ‘must-have’ security in the Internet of Things world?
- fille infidele on How do you package ‘must-have’ security in the Internet of Things world?
- Divya Kolar on Face Age Progression: Technology that can help bring missing children home
- Edilizia popolare on Face Age Progression: Technology that can help bring missing children home
- Divya Kolar on Intel Labs at Intel Developer Forum 2014
- Big Data
- Connected Car
- Context Aware
- Data Society
- Energy Efficiency
- High Performance Computing
- Intel Labs
- Intel Labs Europe
- People & Practices Research
- Research Day
- Social Computing
- US Innovation
Tags#IntelR&Dday @idf08 Big Data circuits Cloud Computing Ct CTO energy efficient Future Lab Future Lab Radio HPC IDF IDF2008 IDF 2010 Immersive Connected Experiences innovation Intel Intel Labs Intel Labs Europe Intel Research ISSCC Justin Rattner many core microprocessor mobility multi-core parallel computing parallel programming radio Rattner ray tracing research Research@Intel Research At Intel Day Robotics security silicon photonics software development Stanford technology terascale virtual worlds Wi-Fi WiMAX wireless