Robotics

DARPA Grand Challenge entrant talks about the event

DARPA Grand Challenge entrant talks about the event
View 0 Images

One of the most interesting entries in the recent DARPA GRAND CHALLENGE was an autonomous motorcycle entered by the GhostRider Robot team, headed by 23 year old Anthony Levandowski and is Berkeley University’s official entry. Levandowski, is a serial entrepreneur at just 23 years of age and has invested a lot of his own money in getting the project up.We http://www.gizmo.com.au/public/news/news.asp?articleid=2652 prior to the event and subsequent to the event to see what he’d learned. Gizmo: What did you learn? We learned a tremendous amount. From an organisational perspective, the larger teams had a clear advantage. They could send people to get food, get spare parts, talk with DARPA, conduct media and PR interviews, and test new code all at the same time. We are a small team and had two people, five if you count my family that came down to help out. I was shocked that the QID was simpler than outlined. It was by no means trivial - it seems that most teams had to “enhance” waypoints to make it through the course. It was obvious who had money and who did testing. Testing was perhaps the most important thing anyone could do. If your vehicle could do only one thing but do it well every time you were better than most teams. We had a human error that killed us… on the morning of the race after waiting for hours in the sun and prepping the vehicle, revising everything and doing it again, we were ready to launch. Finally our turn to depart the chute came and we got very excited. Unfortunately our designated DARPA rep, who launched our vehicle during QID, was not there as he had been reassigned to crowd control in Dagget. That morning (3am) I trained another individual to launch the vehicle. We didn’t get a chance to rehearse and when he launched the bike, the stability control software, that keeps the bike uprught, was not activated. The bike obviously crashed since it wasn’t activated. The reason we needed to turn on and off the stability control software is because when moving the bike by hand we had to turn it off otherwise the stability control would “fight you” to stablise itself, and injure you, as it injured one of our team members during testing (sorry Sing!). From this we learned that user interface is critical, even for an autonomous robot. Had I put a light on the handlebars that indicated the status of the stability control software, we would have been fine, but instead we provided a “comical ending” to the GC. It also seemed that the vehicles that performed the best were those that were modified existing vehicles, rather than those built from scratch. Gizmo: What did everyone learn? The first day of QID allowed for 8 teams to test, I think only one moved! Again everyone saw that testing is the most important aspect of the projects. The next day almost everyone moved! Sensors are very finicky and if they worked well in the pit area doesn’t mean they will work on the course (when everyone is watching). Robots tend to turn right… at least it seamed all the vehicles that crashed hit the right wall more than once. Every effort is rewarded. It was very nice of DARPA to let all the teams, that demonstrated that they were safe and could follow the course somewhat, compete in the GC. Some teams, I won’t name any, used GPS very effectively to almost navigate all the course with minimal sensing needed. GPS is much less reliable when you really need it to work and some times you don’t get your positioning accuracy as good as is possible because you can only see four satellites. Gizmo: If you were assessing the strengths of each of the competitors, who would you say is ahead in each of the technologies and which technologies aren't yet there to the degree required? I was very impressed with the Israeli team (Elbit systems – sciautonics II). They had a very small vehicle and used ultrasonic sensors (which are slow to update at 4Hz) CMU showed that when you push people they can rebuild a vehicle in 3 days. They also did a great job using Lidar (Riegl 2D) to detect obstacles. Gizmo: What will it take to win the challenge next time? Lots of testing and money! Vehicles failed for two main reasons this time: they hit something or they we not able to start again after being paused by DARPA. This is what everybody will be focused on. Gizmo: When will the Grand Challenge be won? GC WILL BE WON in 2006 (next time). Looking at the levels the teams reached from starting work in late 2003, they all came a long way in five months. Extrapolate the progress over 18 months and add some lessons learned … there will be a real competition next time out. The odds are that 50% of the starters will go more than 10 miles. Gizmo: Yours was easily the bravest challenge - it obviously cost you a chunk of your "make ASL a wealthy young man" fund. Was it worth it? Absolutely! It took a lot of my personal resources but the experience was unbelievable. We also have something quite unique and are solving the problems that will allow us to have a superior platform. Personally I spent over $100K on the bike but the costs were a lot more and I couldn’t have done it without our sponsors, especially AMD, Raytheon, Agilent and Crossbow. Gizmo: Will you do it again? We will be back in 2006, and we are currently seeking sponsors who want to be associated with the most innovative team. Our goal for GC ’04 was to demonstrate our concept and prove the vehicle could work (we achieved both). We entered a “pre-prototype” vehicle into the race to show our resolve. Now we are getting the vehicle in race shape. We take our platform very seriously. Mechanically it is THE BEST way to complete GC. Gizmo: Do you believe you are developing commercially applicable technology which the Japanese giant motorcycle manufacturers are not? This is a double question. I am developing the technology that will enable a single track vehicle to be used as a platform for robotics. A single track vehicle is the best platform using wheels for ground locomotion by measure of obstacle handling, cost, weight, size etc. Japanese companies are definitely interested in this. I just wasn’t able to talk to the right people to get them involved last time. As a 23 year old student, albeit with a good track record, it is hard to get to speak to the people who make the decisions. Last time, they said no, but it wasn’t the business development or management that said no. We are now talking to several motorcycle companies. Unfortunately, I cannot devote all my time to GC. Autonomous motorcycles are in their infancy and need lots of nurturing. I am working part time on my other venture called Construction Control Systems.We make software and hardware for automatically managing drawings, change orders, and documents for the construction industry. We have a test model under evaluation at the Ghetty remodel in LA and we made some presentations at the Pentagon for their remodeling project too. This has more short term income potential than the GC project but I only work part-time in the other venture so I can continue with the Grand Challenge. Right now we’re focussed on speaking with companies who can see the benefits of being part of our next Grand Challenge campaign and who can help with funding and/or equipment. Anthony can be contacted at big_vic@uclink4.berkeley.edu

No comments
0 comments
There are no comments. Be the first!