Drivers not required

10 December 2005

digital human opening spreadA bunch of off-road cars and trucks are lined up as dawn breaks over the Nevada desert. There are no drivers at the wheel: these cars have computerised minds of their own. Their engines running, they are just waiting for the command from their electronic overlord to start driving. As the first one gets the signal, it lurches out of the starting gate and, instead of driving straight out into the desert, it suddenly turns left towards a grandstand crowded with spectators. But, just as suddenly it changes its mind and it is off on a 200km course across the desert in the second race of its kind: a race that involves no human drivers, except for those in the safety cars.

The people looking on as each car heads off in this robo-rally include teams of engineers who stand to collect a $2m prize if one of these vehicles makes it to the finish in one piece. For close to two years, these teams have completed qualifiers that test the cars' ability to dodge cones and concrete slabs. Now, each car would have ten hours to pick its way round a course that crossed dry lake beds, ran along rutted dirt tracks and through narrow underpasses. Finally, the course wound through a high pass with cliffs on one side and a steep, metal-bashing drop on the other. On the way, the vehicles would have to deal with obstacles as tricky as tank traps. Some of them looked like they could drive over most of the obstacles. The biggest was a monster truck from Oshkosh, with its cab sawn off so that it could get through the low tunnels on the route.

This combination of Robot Wars and Wacky Races had a purpose. In ten years, US military, which put up the $2m prize for the race, wants to have a third of all military vehicles - whether planes, trucks or tanks - to be able to drive themselves with no help from soldiers. And robo-vehicles will not be restricted to tanks and trucks. It did not make it to the final race, but one team has successfully made a robotic motorbike that can stay upright without a rider.

Maybe 20 years on from now, you won't need a driving licence because the car will be doing all the driving itself. And even when you take over and try to point it in one direction, computers under the bonnet will be working out whether that is a good idea or not.

Some of this technology is already in cars. Some posh cars have radar systems similar to those used by aircraft. They look out for other cars, and let you know loudly if you are likely to hit one. Cameras in some cars look at the white lines on the motorway and use them to make sure you cannot drift out of one lane, and end up decorating the front of a truck piling along in the next lane. Soon, there will even be smart headlights that follow the road around at night, making sure you don't miss any sharp corners and discover off-roading the hard way.

In 2004, it looked as though technology had a long way to go. The first race saw even the best-designed car that year - Carnegie-Mellon University's Sandstorm - forced to stop after just 12km. The others did not even get that far. Some ended up in ditches, most had to stop because their computers became confused. One managed to get itself stuck in a barbed-wire fence.

This year and almost 18 months on, five vehicles finished a different course, including OshKosh's massive truck TerraMax and the two cars entered by Carnegie-Mellon. The difference? According to the veteran robocar designer at Carnegie-Mellon, William "Red" Whittaker, who came up with the concept behind Sandstorm and its successor H1ghlander, the cars could "see" a lot better.

Robocar makers are careful people, which is handy. No-one wants to be run over by a tractor driven by Silicon Sam just because its cameras were playing up. The engineers don't just rely on one set of eyes to guide a vehicle. They use lots of them. And, because these cars have a digital map, they see how their car is doing by asking orbiting satellites where they are.

Every competitor had a Global Positioning System (GPS) receiver that it could use to check its position on the map. But that was not enough on its own. The sudden left turn by the car at the start that went to be the winner - Stanford Racing's Stanley - was because the GPS told it to go that way. Later on in the race, several teams noticed that the GPS was telling the vehicles to head straight over the cliffs on the treacherous Beer Bottle Pass. Plus, GPS will not tell you about rocks, tank traps or animals in the way.

Sitting on top of each vehicle was a forest of cameras and sensors. Stanley alone has four laser rangefinders, radar and two cameras for stereo vision, with yet another camera for getting close-ups of objects in the road. The laser rangefinders paint the landscape with light - invisible to the human eye - and see what gets reflected back to work out what is in the way. The vehicles back that information with images from cameras.

Gary Schmiedel, in charge of the TerraMax project at Oshkosh, explained why his team also used stereo cameras: "Stereo vision finds things like fences or objects like cones that are very different in colour to the rest of the scene. But you need to be careful with the information you pay attention to. These sensors report a lot of things that are not of interest, things that are not in your way."

Whittaker's Sandstorm and H1ghlander have laser scanners and cameras sitting in a dome on top of the vehicle that swings from side to side. "Driving with your head in a neck brace gives you an idea of what these sensors give to you without this," said Whittaker. "If you're coming up to a T-junction and want to see what is, the sensor can turn and look directly out the passenger side."

Although the cameras on TerraMax do not move around as much as Whittaker?s vehicles, they do have a trick up their sleeve that goes way beyond what people or animals are able to do. "For stereo vision, you need two cameras with overlapping fields of vision. We have three cameras," said Schmiedel. Two are close together, like human eyes, one is further away from those two. "At slow speeds, we use the two cameras that are close together. As speed picks up, we switch to use two cameras that are space further apart."

The advantage of TerraMax's approach is that by moving its eyes further apart, it gets a much more accurate range for obstacles that a long way away. Switching back to the cameras that are close together, it can get a better look at things that the vehicle is near. For Schiedel, that ability translates into an ability to drive faster safely than human drivers. "Where you have a known area and the path is pretty well-defined, you could run the truck a lot faster," said Schiedel. For roads with tight turns and big hazards, the truck could slow right down. "We negotiated Beer Bottle Pass very gingerly."

Whittaker said he sees robocars taking part in races alongside vehicles with human drivers to demonstrate what they can do. Robot cars running in the Paris-Dakkar rally is just one possibility. "We are looking for another race. There are so many classic races out there." He likens the situation to the early days of flight: "Then, no-one knew what planes could do, so they went around barnstorming to show the machines off. Today, most people haven?t seen driverless vehicles and even fewer have ridden in them." Having cars compete would let the public see how the robotic versions can perform.

Unlike the early pilots, the robo-vehicle designers have some ideas about what their creations will be able to do. Whittaker wants vehicles for his ranch that will fix fences and harvest crops. Others will drive around airports, plouging snow out of the way. And Schiedel forecasts we will see convoys of trucks following nose to tail along motorways, perhaps with a human driver at the head to let the robots know where to go. But not necessarily at the front. Whittaker sees the situation where the robots will be out in front: the driver will just decide which road they take. Soon we will all be back-seat drivers.

Written for the December 2005 issue of Flipside.