The DARPA Grand Challenge was the first race for autonomous robots from LA to Las Vegas on March 13, 2004 with a US$1 million prize. The race was be contested by 24 cars and one autonomous two-wheeler. This interview was conducted with the Team leader of the Ghostrider Robot Team, Anthony Levandowski just prior to the event.
Gizmo: Why two wheels?
AL: We are looking at where robotics will be 15 years from today. Single track (two wheel) vehicles cost less, are more efficient and can traverse more extreme environments because they only have to conform to the terrain in one axis.
For these reasons, two wheeled vehicles are the second most common means of locomotion for humans, second only to walking! Achieving stability is much more complex than on any other platform, which is why no one has succeeded at creating an autonomous motorcycle.
By understanding and mastering the stability of this platform we hope to create a vehicle with performance characteristics that no other robot can match.
Gizmo: How are you progressing with a few weeks to go?
AL: We are doing great! Our vehicle is up and running. We finally found the reason why the vehicle kept crashing (besides gravity). Our maximum steering angle correction speed was too low. Our GPS navigation is field (but not on the actual vehicle) tested and works great. The vision system shows promising results from sample data feeds collected during desert visits. We will see how well they all work as one system at QID! In any case we consider our project a success because we have demonstrated how to control a single track vehicle for autonomous operation.
Gizmo: Why did you choose a motorcycle over four wheels?
AL: Good question. The fastest way for a human to go from point a to point b off-road is on a motorcycle, why would robots be different? Let's look at why the motorcycle is more flexible.Four, three or six wheeled vehicles are great for stability. However their platform is physically attached to the surface of the ground.
By this I mean that they are designed to be supported by the ground in at least three points, forcing them to be subject to terrain variations in two dimensions. This is great because you can use them with little initial engineering investment and very simple (and slow) controls (speed and desired steering angle). Single track vehicles (like a motorcycle) require constant fine tuning, this fine tuning is extremely difficult to achieve (we currently have over 140,000 lines of code for stability only).
But once achieved we have a platform that sees the world as one dimensional (versus two dimensional for surfaces). This reduces the amount of obstacles the vehicle needs to avoid and simplifies the path planning. Overall we are:
Narrower, so it's harder to hit trees if you are 18 inches wide versus 8 feet, and easier to chase foes down narrow roads when your vehicle actually fits on the sidewalk) and we are subject to less roll inducing obstacles, like rocks hitting the left and right wheels at different times.
Lighter - have you ever seen a motorcycle get stuck in the mud?
Cheaper - I am 23 and can't afford a hummer!
More robust - we have crashed over 100 times, hitting everything from the ground to trees to teammates not paying attention during testing.
More stealthy - if we are going to use this as a weapon you need to be able to sneak up without looking like a tank on the radar.
Fuel efficient - we get better than 50 miles per gallon.
Gizmo: What base motorcycle are you using?
AL: No motorcycle company had enough vision of the future to sponsor us so I cannot disclose the brand.
Gizmo: What extra gear goes on the bike to substitute for the rider?
AL: Tons! Our most important components are a solid state Crossbow MEMS 6axis gyro and an AMD microcontroller. These two components are like the rider 'intuition and feel' and provide the robotic reflexes needed to keep the bike standing up. Next up is the CMG (control moment gyroscope) allows the bike to stand on its own while stopped.
Gizmo: Will the challenge be won this year? Next?
AL: I hope not! It would be a shame to have this contest only once. Having the contest reoccurring every two to four years would be a great way to stimulate robotics similarly to the Olympics. Funding which is hard to come by, provides a big correlation with team performance. It would be extremely difficult to complete the challenge with less than $ x million, some teams have this amount many times over. I wish them the best of luck.If there is no winner on March 14th, the next challenge will occur in 2006. If the course is similar, some vehicle will complete the route.
Gizmo: Will it be won by a two or four wheeled vehicle?
AL: If the contest is not won this year we will be a force to reckon with.
Gizmo: How many years do you think it will take to build an autonomous motorcycle (or car) capable of averaging the same speed as a human-piloted bike/car in the desert with less risk of crashing?AL: Wow, we can hardly go from point a to point b without destroying the vehicle. It took millions of years of evolution to create the sensors and processing power that humans possess. It will not take as long to synthetically exceed the capabilities in all domains. Making vehicles truly safer under automated control than when human operated will take a long time though - maybe 50 years. It's hard to say.